Академический Документы
Профессиональный Документы
Культура Документы
Markus J. Buehler
Atomistic Modeling
of Materials Failure
123
Markus J. Buehler
Laboratory for Atomistic and Molecular Mechanics
Department of Civil and Environmental Engineering
Massachusetts Institute of Technology
77 Massachusetts Avenue, Room 1-235A&B
Cambridge, MA 02139
USA
mbuehler@mit.edu
springer.com
To my wife, for inspiration and loving support
Preface
This book has evolved from lecture notes of undergraduate and graduate level
subjects as well as review articles and journal papers. The book provides a
review of atomistic modeling techniques that successfully link atomistic and
continuum mechanical methods. It intended to be a reference for engineers,
materials scientists, and researchers in academia and industry. The writing of
this book was motivated by the desire to develop a coherent set of notes that
provides an introduction and an overview into the field of atomistic-based
computational solid mechanics, with a focus on fracture and size effects.
The book covers computational methods and techniques operating at the
atomic scale, and describes how these techniques can be used to model the
dynamics of cracks and other deformation mechanisms. A description of molec-
ular dynamics as a numerical modeling tool covers the use of interatomic
potentials (pair potentials such as the Lennard-Jones model, embedded atom
method (EAM), bond order potentials such as Tersoff’s and Brenner’s force
fields, as well as the first principles based ReaxFF Reactive force field) in
addition to the general philosophies of model building, simulation, interpre-
tation, and analysis of simulation results. Example applications for specific
materials (such as silicon, nickel, copper, carbon nanotubes) are provided
as case studies for each of the techniques, areas, and problems discussed.
Readers will find a physics-motivated discussion of the numerical techniques
along with a review of mathematical concepts and code implementation issues.
Using specific examples such as investigations of crack dynamics in brittle
materials or deformation mechanics of nanomaterials, this volume conveys
how atomistic studies have helped to advance developing new theories, or
provided insight into the molecular deformation mechanisms, explaining or
supplementing experimental results. Many of the examples are adapted from
studies carried out by the author of this book, and some of the discussion
should therefore not be considered as a comprehensive and inclusive review
with respect to the wider range of available results. Rather, they represent a
set of specific examples to illustrate the application of the atomistic simulation
techniques reviewed here.
VIII Preface
Completing this book would not have been possible without the help and
support of numerous people. The author is most greatly indebted to all who
have contributed to this book in some way. In particular, sincere gratitude goes
to those individuals from whom he had the opportunity to learn from over
the years, in particular his graduate advisor Huajian Gao and postdoctoral
advisor William A. Goddard III. The author is deeply humbled by the many
contributions that have pioneered the development of this research field over
the past decades. The author would also like to thank the Editor Mrs. Elaine
Tham of Springer and her staff for the continuous support for this project.
The efforts by the reviewers of the manuscript are greatly acknowledged, as
they provided valuable suggestions for revisions in the final manuscript.
The study of materials failure using atomistic simulation has been a
rewarding journey that continues to bring so much joy, excitement, and inspi-
ration. The author hopes to convey some of the excitement about this research
field in this book.
Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LXV
Part I Introduction
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.1 Materials Deformation and Fracture Phenomena:
Why and How Things Break . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Strength of Materials: Flaws, Defects, and a Perfect
Material . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Crystal Structures and Molecular Packing . . . . . . . . . . 8
1.2.2 Cracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.3 Dislocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.2.4 Other Defects in Crystals and Other Structures . . . . . 12
1.3 Brittle vs. Ductile Material Behavior . . . . . . . . . . . . . . . . . . . . . 12
1.4 The Need for Atomistic Simulations . . . . . . . . . . . . . . . . . . . . . 15
1.5 Applications: Experimental and Computational Mechanics . . 18
1.5.1 Experimental Techniques . . . . . . . . . . . . . . . . . . . . . . . . . 18
1.5.2 Example Applications: The Significance
of Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.6 Outline of This Book . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
List of Figures
4.9 The figure plots the elastic properties under uniaxial loading
without Poisson relaxation for the harmonic potential. In
the plot, stress vs. strain, as well as the number of nearest
neighbors are shown. The lower two subplots show Young’s
modulus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138
4.10 Illustration of the parameters used in the biharmonic
potential defined in (4.44). The plot defines r, k0 , k1 , ron ,
rbreak , as well as the “atomic” strain . . . . . . . . . . . . . . . . . . . . . . 139
4.11 Elastic properties of the triangular lattice with biharmonic
interactions, stress vs. strain in the x-direction (a) and in
the y-direction (b). The stress state is uniaxial tension, that
is the stress in the direction orthogonal to the loading is
relaxed and zero . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.12 Bond breaking process along the fracture plane and
calculation of fracture surface energy for (a) direction
of high fracture surface energy and (b) direction of low
fracture surface energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.13 Elastic properties associated with the harmonic potential,
[100] crystal orientation, with Poisson relaxation. Poisson
ratio is ν ≈ 0.33 and is approximately independent of the
applied strain. The plot shows the elastic properties as a
function of strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.14 Elastic properties associated with the harmonic potential,
[100] crystal orientation, without Poisson relaxation. The
plot shows the elastic properties as a function of strain . . . . . . 145
4.15 Elastic properties associated with the harmonic potential,
[100] crystal orientation, triaxial loading. The plot shows
the elastic properties as a function of strain . . . . . . . . . . . . . . . . 146
4.16 Elastic properties associated with the harmonic potential,
(a) [110] and (b) [111] crystal orientation, uniaxial
loading with Poisson relaxation. The plot shows the elastic
properties as a function of strain . . . . . . . . . . . . . . . . . . . . . . . . . 147
4.17 Elastic properties associated with (a) LJ potential, and (b)
an EAM potential for nickel [15], uniaxial loading in [100],
[110] and [111] with Poisson relaxation . . . . . . . . . . . . . . . . . . . 148
4.18 This plot depicts a series of snapshots of a single molecule
with increasing length L, at constant temperature. The
longer the molecule, the more wiggly the geometrical shape . . 150
XXX List of Figures
6.64 Crack tip history a(t) and crack tip velocity v as a function
of time, suddenly stopping mode I crack. The limiting speed
according to the linear theory is denoted by the black line
(Rayleigh velocity), and the super-Rayleigh terminal speed
of the crack in the nonlinear material is given by the blueish
line. When the crack stops, the crack speed drops to zero . . . . 275
6.65 Evolution of principal maximum stress and potential energy
along the prospective crack line; for a mode I nonlinear crack 276
6.66 Maximum principal stress field for various instants in time,
for mode I nonlinear crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
6.67 Variation of stress at fixed distance ahead of the stopped
nonlinear mode I crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
6.68 Schematic of waves emitted at a suddenly stopping mode
II crack; (a) stopping of daughter crack, (b) stopping of
mother crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
6.69 Evolution of (a) principal maximum stress and (b) potential
energy along the prospective crack line; for linear supersonic
crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
6.70 Potential energy field for various instants in time, mode II
linear crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
6.71 Variation of stress at fixed distance ahead of the stopped
intersonic mode II crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
6.72 Crack extension history vs. time for the supersonic mode II
crack. The dashed line is used to estimate the time when
the mother crack comes to rest . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
6.73 Evolution of (a) principal maximum stress and (b) potential
energy along the prospective crack line; for nonlinear
supersonic crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
6.74 Potential energy field around the crack tip for various times,
suddenly stopping mode II crack . . . . . . . . . . . . . . . . . . . . . . . . . 284
6.75 Normalized stresses σ ∗ vs. time, suddenly stopping
supersonic mode II crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
6.76 Geometry of the simulations of cracks at bimaterial
interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
6.77 Crack tip history and crack velocity history for a mode I
crack propagating at an interface with Ξ = 10. Subplot (a)
shows the crack tip history, and subplot (b) shows the crack
tip velocity over time. A secondary daughter crack is born
propagating at a supersonic speed with respect to the soft
material layer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
List of Figures XLI
6.78 The plot shows the stress fields σxx , σyy , and σxy for
a crack at an interface with elastic mismatch Ξ = 10,
before a secondary crack is nucleated. In contrast to the
homogeneous case, the deformation field is asymmetric. The
dark grey shades corresponds to large stresses, and the
lighter grey shades to small stresses . . . . . . . . . . . . . . . . . . . . . . . 290
6.79 The plot shows the particle velocity field (a) u̇x and (b) u̇y
for a crack at an interface with elastic mismatch Ξ = 10,
before a secondary crack is nucleated. The asymmetry of
the particle velocity field is apparent . . . . . . . . . . . . . . . . . . . . . . 291
6.80 The plot shows the potential energy field for a crack at an
interface with elastic mismatch Ξ = 10. Two Mach cones in
the soft solid can clearly be observed. Also, the mother and
daughter crack can be seen. In the blow-up on the right, the
mother (A) and daughter crack (B) are marked . . . . . . . . . . . . 291
6.81 The plot shows the stress fields σxx , σyy , and σxy for a crack
at an interface with elastic mismatch Ξ = 10. In all stress
fields, the two Mach cones in the soft material are seen. The
mother crack appears as surface wave behind the daughter
crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
6.82 The plot shows the particle velocity field (a) u̇x and (b) u̇y
for a crack at an interface with elastic mismatch Ξ = 10.
The shock fronts in the soft solid are obvious . . . . . . . . . . . . . . 292
6.83 Atomic details of nucleation of the secondary crack under
tensile dominated loading. The plot shows the shear stress
field σxy near the crack tip. Atoms with the energy of a
free surface are drawn as larger atoms. The plot suggests
that a maximum peak of the shear stress ahead of the crack
tip leads to breaking of atomic bonds and creation of new
crack surfaces. After the secondary crack is nucleated (see
snapshots (2) and (3)), it coalesces with the mother crack
and moves supersonically through the material (snapshot (4)) 293
6.84 Crack tip history for a mode II crack propagating at an
interface with Ξ = 3. The plot illustrates the mother–
daughter–granddaughter mechanism. After a secondary
daughter crack is born travelling at the longitudinal wave
speed of the soft material, a granddaughter crack is born
at the longitudinal wave speed of the stiff material. The
granddaughter crack propagates at a supersonic speed with
respect to the soft material layer . . . . . . . . . . . . . . . . . . . . . . . . . 294
XLII List of Figures
6.91 Crack tip velocity history for a mode III crack propagating
in a harmonic lattice for two different choices of the spring
constant ki . The dotted line shows the limiting speed of
the stiff reference system, and the dashed line shows the
limiting speed of the soft reference system. Both soft and
stiff systems approach the corresponding theoretical limiting
speeds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 301
6.92 Mode III crack propagating in a thin elastic strip that is
elastically stiff. The potential energy field is shown while
the crack propagates supersonically through the solid. The
stiff layer width is h = 50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
6.93 Check of the scaling law of the mode III Broberg problem.
The continuous line refers to the analytical continuum
mechanics solution [24] of the problem. The parameters
γ0 = 0.1029 and τ0 = 0.054 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
6.94 Suddenly stopping mode III crack. The static field spreads
out behind the shear wave front after the crack is brought
to rest . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
6.95 Geometry used for simulating mode I fracture in silicon.
The systems contain between 13,000 and 113,015 atoms
with Lx ≈ 550 Å and Ly ≈ 910 Å. The numerical model is
capable of treating up to 3,000 atoms with ReaxFF in Ωrx . . 308
6.96 Crack propagation with a pure Tersoff potential (subplot
(a)) and the hybrid ReaxFF-Tersoff model (subplot (b))
along the [110] direction (energy minimization scheme).
The darker regions are Tersoff atoms, whereas the brighter
regions are reactive atoms. The systems contain 28,000
atoms and Lx ≈ 270 Å × Ly ≈ 460 Å . . . . . . . . . . . . . . . . . . . . . . 309
6.97 Crack dynamics along the [110] direction at finite
temperature (T ≈ 300 K), 10% strain applied . . . . . . . . . . . . . . 309
6.98 Crack dynamics along the [100] direction at finite
temperature (T ≈ 300 K, 10% strain applied). Shortly after
nucleation of the primary crack two major branches develop
along [110] directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
6.99 Crack speed as a function of load, for the (110) system
(subplot (a), and the (111) system (subplot (b)) . . . . . . . . . . . 311
XLIV List of Figures
7.22 Subplots (a) and (b) show a view from the [110]-direction
(a) before nucleation of dislocations on secondary glide
systems (therefore only straight lines), and (b) after
nucleation of dislocations on secondary glide systems (which
appear as curved lines) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
7.23 Subplots (a) and (b) show detailed views on the formation
of sessile Lomer–Cottrell locks, with its typical shape of a
straight sessile arm connected to two partial dislocations . . . . 352
7.24 The final network from a distant view, including a blow-up
to show the details of the network [39]. The characteristic
structure of the network is due to the fact that all sessile
defects (both trails of partial and complete point defects)
as well as sessile dislocations as part of the Lomer–Cottrell
locks assume tetrahedral angles and lie on the edges of
Thompson’s tetrahedron. The wiggly lines in the blow-up
(see the right half of the figure) show partial dislocations,
and the straight lines correspond to sessile defects . . . . . . . . . . 354
7.25 Development of the density of different defects during the
simulation using a method of separating defects of different
energies [39] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
7.26 Simulation geometry, lattice orientation, and time-sequence
of the EAM simulation. (a) Simulation geometry and (b)
lattice orientation, also defining the directions for the x, y, z
coordinate system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
7.27 Sequence of snapshots that illustrate the dislocation
nucleation process from the crack tip, for the case of an
EAM potential. This snapshot depicts the early stages of
dislocation nucleation, depicting how dislocations grow from
the tips of the crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
7.28 Closer view on the dislocation structure at a later stage in
the simulation with the EAM potential (for a system with a
single crack at one surface) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359
7.29 Centrosymmetry analysis that shows the activation of
secondary slip systems, at later stages in the simulation
with the EAM potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360
7.30 Centrosymmetry analysis of the details of the dislocation
structure, for the EAM simulation. In contrast to the LJ
simulations, here complete dislocations (that is, leading and
trailing partial dislocations) are emitted at the crack tip.
As a consequence, the dislocation cutting products (partial
point defects) have a finite length. Subplot (a) depicts a
sequence of two snapshots, illustrating the growth of the
dislocation network at the crack tip. Subplot (b) shows a
detailed view of the dislocation structure . . . . . . . . . . . . . . . . . . 361
List of Figures XLIX
7.36 Partial dislocation emission from crack tip for the 110112
crack orientation (only part of the crystal close to the
crack tip is shown), under presence of oxygen molecules
in the void inclusion (same coordinate system as shown
in (Fig. 7.35) [40, 41]. Subplots (a),(b) and (c),(d) depict
results for two different starting positions of a single O2
molecule in the crack ((a) and (b): oxygen molecule at the
tip of the crack, (c) and (d): oxygen molecule at the side of
the crack face). Defects produced at the crack tip are circled
and the dotted line indicates the approximate region of
atoms modeled by ReaxFF. Subplot (a) shows the structure
of the crack tip after 10,000 equilibrium molecular dynamics
integration steps, and (b) shows the first partial initiating
at a strain of 0.049. Subplot (c) depicts the resulting crack
structure after 10,000 molecular dynamics integration steps
for a different starting position of the O2 molecule (placed
at the side of the crack), and (d) shows the partial initiation
for this case at a strain of 0.041. The figures show lattice
defects produced in the bulk crystal surrounding the crack
as a result of oxidation at crack surface, even before strain
is applied . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368
7.37 Partial dislocation emission from crack tip for the 111112
case, results for the case without oxygen molecules. Subplot
(a) shows brittle crack opening for the no oxygen case at a
strain of 0.05, but the results depicted in subplot (b) shows
that the crack tip emits dislocations as well at a strain of
0.051 [40, 41] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 369
7.38 Partial dislocation emission from crack tip for the 111112
case, results for the case with oxygen molecules. Subplot (a)
shows the structure of the crack tip with O2 after 10,000
equilibrium molecular dynamics integration steps. The
differently shaded region to the left of the crack represents a
grain boundary (GB) that has formed during the oxidation
process. In subplot (b) at a strain of 0.03 a void has
initiated close to crack tip. The crack tip starts to open
up in subplot (c) at a strain of 0.05, indicating a brittle
fracture mode [40, 41] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
List of Figures LI
Variable Description
U , Utot Total potential energy of a system
S Entropy of a system
T Temperature (in Kelvin)
F Free energy of a system
K Kinetic energy of a system
H Total Hamiltonian of a system
kB Boltzmann constant
Ui Potential energy of atom i
φ Interatomic potential energy function
rij Distance between atom i and atom j
k, ki Spring constant of the harmonic potential
0 Energy well depth, Lennard-Jones potential
σ Length parameter, Lennard-Jones potential
F Atomic force vector
r Atomic position vector
Rcut Cutoff radius for interatomic potential
Rtrans Size of transition region in multiparadigm model
ṙ Atomic velocity
r̈ = a Atomic acceleration
Rbuf Size of buffer region in multiparadigm model
wi Relative weight of force field engine
in multiparadigm model
BOij Bond order of pair of atoms (ReaxFF model)
E Young’s modulus
µ Shear modulus
ν Poisson’s ratio
ρ Material mass density
cijkl Elasticity tensor coefficients
S, Ψ Strain energy density
LXVI Nomenclature
Introduction
1
Introduction
Catastrophic phenomena that afflict millions of lives, ranging from the failure
of the Earth’s crust in earthquakes, to the collapse of buildings, to the failure of
bones due to injuries, all have one common underlying theme: the breakdown
of the basic constituents of any material ultimately leads to the failure of its
overall structure and intended function. Understanding how materials fail has
always been of great importance to enable and advance technologies. Over
several thousands of years, the knowledge about the behavior of materials in
extreme conditions has furnished the way for modern technologies, by enabling
the use of materials for a variety of applications.
At this moment we are at the crossroads to a new era where humans, for
the first time, start creating structures and technologies at the scale of single
atoms and molecules (see Fig. 1.1). Such nanotechnology could revolutionize
the way we live, learn, and organize our lives in the decades to come. Computa-
tional modeling, in particular atomistic and molecular simulation, is becoming
increasingly important in the development of such new technologies. At
nanoscale, the effects of single atoms, individual molecule, or nanostructural
features may dominate the material behavior. Thus novel modeling and sim-
ulation approaches are a vital component in enabling the engineering design
process. Atomistic models, by providing a material description that starts at
a fundamental scale, are thus expected to be important not only for scien-
tists but also for engineers. How will engineers of the future treat nanoscale
systems? Atomistic, mesoscale and multiscale modeling, and simulation as
discussed in this book may provide an important tool for future engineers.
How can these techniques be applied to solve problems? What kind of
information can be obtained, and how can this be related to existing theories,
methods, and concepts, and most importantly, experimental studies? How
do existing engineering theories relate to the results obtained from advanced
simulation approaches? This book will address some of these questions by
providing a fundamental description of the various aspects of atomistic and
molecular simulation approaches. In addition to practical questions regarding
algorithms and implementation, this book explores the science behind fracture
4 Atomistic Modeling of Materials Failure
Fig. 1.1 Illustration of how the characteristic material scales of technological eras
have been reduced from the scales of meters to the scales of individual molecules
and atoms. The current technological frontier is the development of molecular and
atomistic structures at the interface of physics, biology and chemistry, leading to a
new bottom-up approach in creating and characterizing materials
is how materials fail under extreme conditions, and how the microscopic or
macroscopic failure processes are related to atomistic details.
Fig. 1.2 The plot shows simple, schematic stress–strain diagrams characteristic
for a brittle and a ductile material. Similar curves are found for other materials,
including polymers or rubber-like materials. The cross symbol (“x”) indicates the
point of material failure [1]
6 Atomistic Modeling of Materials Failure
Fig. 1.3 Homogeneous material (subplot (a)) and material with elliptical hole (sub-
plot (b), length of elliptical hole is 2a). The presence of the elliptical void leads to
a magnification of the stress in the vicinity of the tip of the defect (see schematic
illustration of stress profile)
There exists a large variety of atomic structures with distinct packing arrange-
ments and different chemical bonds, giving rise to the numerous classes of
materials. For instance, metals or ceramics typically feature a densely packed,
highly organized crystal structure. Often, the bonds between these atoms are
quite strong, as it is the case for metallic bonding or in covalent bonding (e.g.,
forming metals like titanium, copper, nickel, iron, or ceramics such as alumina,
quartz, or feldspar). Metals or ceramics without an organized crystal struc-
ture are also found, with glass being a prominent example. These materials
contain a random, amorphous arrangement of the atoms.
Many polymers such as rubber also have a rather amorphous microstruc-
ture, where molecules are arranged in a less-ordered fashion. These materials
are typically less dense and softer, because fewer or less strong atomic bonds
per unit volume resist mechanical deformation. Once stretched significantly,
the disordered molecules in amorphous polymers often arrange into regular
patterns, and the material stiffens significantly. This phenomenon occurs for
instance when a rubber band is stretched further and its resistance increases.
1 Introduction 9
Biological materials and tissues, for instance materials made out of pro-
teins, represent an intermediate case – they can feature highly organized and
ordered, quite precise molecular structures in which arrays of intermolecu-
lar H-bonds and ionic interactions provide the basis for their properties. In
addition to differences in the molecular packing, many biological and natu-
ral materials feature hierarchical structures, that is, a different characteristic
organization of atoms, molecules, supermolecules at different lengthscales.
The particular molecular deformation mechanisms that emerge under
large loads depend very strongly on the atomic, molecular, or hierarchical
microstructure and the associated chemical bonds. Similarly, the definition of
defects depends on the reference geometry of the material. A large part of this
book is focused on the deformation and failure of crystalline materials. For this
class of crystalline materials, we will explain how some of the basic deforma-
tion and fracture mechanisms can be studied using atomistic methods. Many
crystalline materials are associated with great technological significance: met-
als being used as structural materials, as materials for electronic applications,
or as decorative components in buildings. Crystalline materials such as silicon
or silica play a key role in microelectronics and micromechanical (e.g., MEMS)
devices. Silicon, in particular, is of great importance due to its semiconductor
properties and due to its availability in a very pure form.
Important crystal structures frequently found in metals or ceramics include
simple cubic (SC), face centered cubic (FCC), body centered cubic (BCC),
and hexagonal closest packing (HCP). A few crystal structures are shown
in Fig. 1.5. Variations of these crystal structures exist for two-dimensional
geometries. In this book, we will focus on deformation behavior of the simplest
crystal structures, primarily studying FCC crystals and triangular lattices.
Fig. 1.5 Overview of different crystal structures, showing the SC, FCC, and BCC
crystal structure
1.2.2 Cracks
Cracks are typically characterized by their geometry, most often by their size,
since this parameter controls the influence of a crack on the stresses in the
surrounding material (this can also be seen in (1.1)). As will be discussed
later, an extensive theoretical framework has been developed that deals with
the stress fields around cracks, the geometric effects for various kinds of cracks,
the link between the applied load further away from the crack, and the amount
of stress magnification observed in the vicinity of the crack.
There exist intermediates between cracks and a perfect crystal lattice
material. For instance, grain boundaries feature a reduced traction or reduced
bonding strength across a geometric interface.
1.2.3 Dislocations
Dislocations are crystal defects that enable the deformation of crystals, con-
stituting the fundamental carriers of plasticity or the mediators of plasticity.
The existence of dislocations was discovered when it was observed that the
predicted strength of materials (based on the strength of individual atomic
bonds) was much larger than the strength of materials actually measured.
This meant that somehow it was impossible to take advantage of the actual
bond strength and that deformation was controlled by mechanisms that oper-
ated at much lower critical stress levels. For instance, the theoretical shear
strength of a material can be estimated by
µ
τth ≈ √ , (1.3)
30
1 Introduction 11
where µ is the shear modulus (often between 25 and 100 GPa for many metals).
The predicted shear strength would then be close to several GPa and larger.
In contrast, the maximum shear resistance of crystallographic planes mea-
sured in real metals is many orders of magnitudes smaller. In 1934, Orowan,
Polanyi, and Taylor proposed that this difference is due to existence of crys-
tal defects referred to as dislocations [57–59]. The discrepancy between the
theoretical shear strength and the measured values can be explained straight-
forwardly. Whereas the estimate in (1.3) assumes that all interatomic bonds
participate collectively in shear deformation, the concept behind a dislocation
is that only a few atomic bonds participate in deformation of a larger piece
of a crystal. Instead of homogeneously shearing a crystal, a localized wave
of displacements moves through the solid that deforms it step by step. The
localization of displacements is energetically more favorable over the homo-
geneous deformation path, and thus, it is the dominating mechanism in the
deformation of metals. Since their discovery in the early 1930s, the concept
of dislocations has helped to explain many of the perplexing physical and
mechanical properties of metals [57].
Fig. 1.6 Dislocations are the discrete entities that carry plastic (permanent) defor-
mation; measured by a “Burgers vector.” The snapshots illustrate the nucleation
and propagation of an edge dislocation through a crystal, leading to permanent
deformation
Crystals may feature many other defects, in addition to cracks and dislocations
that were introduced above. For instance, if grains of crystals of different
orientation interface, defects referred to as grain boundaries are formed. At
the atomistic scale, grain boundaries constitute high-energy regions in which
the ideal atomic lattice is disturbed in order to provide a bridge between the
two crystal orientations. The presence of these high-energy domains can play
a critical role in the deformation of materials. Grain boundaries may serve
as sources for dislocations and as obstacles for the motion of dislocations.
The interplay of these competing effects is controlled by the interaction of
elastic energy of dislocations, the size of the crystal grains, and the particular
structure of the grain boundary. It leads to intriguing material phenomena,
such as the strengthening of materials by reducing the size of crystal grains.
Other crystal defects include point defects, that is, missing atoms at lattice
sites typically referred to as vacancies, interstitial atoms represented by an
additional atom situated between the regular lattice sites, or impurities where
foreign atom types are inserted into a crystal lattice, often substituting the
atom at an actual lattice position.
Each of these defects plays a significant role in the material behavior,
as they influence how the carriers of plasticity (dislocations) nucleate and
propagate or how other deformation mechanisms such as diffusion or crack
extension become activated. We will discuss some of these in later chapters.
Fig. 1.7 Schematic of brittle (a) vs. ductile (b) materials behavior. In brittle frac-
ture, the crack extends via breaking of atomic bonds. In ductile fracture, the lattice
around the crack tip is sheared, leading to nucleation of crystal defects called dislo-
cations. Which one the two mechanisms is more likely to occur determines whether a
material is brittle or ductile; this distinction is closely related to the atomic structure
and the details of the atomic bonding
not possible. Thus they mostly exhibit brittle failure with little ductility. The
atomistic mechanisms of brittle fracture or ductile failure is quite different.
While atomic bonds are broken by stretching the solid in brittle fracture, the
sliding between planes is achieved by shearing the solid in ductile failure.
The ductile vs. brittle failure of materials is schematically summarized in
Fig. 1.7. Figure 1.7a shows brittle materials failure by propagation of cracks
and Fig. 1.7b depicts ductile failure by generation of dislocations at a crack tip.
The atomic details of such different behavior is shown in Fig. 1.8, providing a
snapshot of an atomistic simulation.
Fig. 1.8 Brittle (a) vs. ductile (b) materials behavior observed in atomistic com-
puter simulations. In brittle materials failure, thousands of cracks break the material.
In ductile failure, material is plastically deformed by motion of dislocations
The origin of brittle vs. ductile behavior of materials has been linked to
atomic processes in work by Rice and Thomson [66]. They considered quan-
titatively the energy penalty associated with shearing atomic planes on top
of each other vs. the mechanism of creating new surfaces. These studies sug-
gested the concept that there exists a competition between ductile (dislocation
emission) and brittle (cleavage) mechanisms at the tip of a crack. The model
by Rice and Thomson has been extended further in many following studies,
to include additional material parameters that required to characterize this
behavior. Most importantly, the introduction of the unstable stacking fault
energy γus [30, 67] describes the resistance of the material to the nucleation
and motion of dislocations. This provides an corresponding parameter to the
fracture surface energy, which describes the resistance of materials to fracture
by creating new surfaces. It has been shown that once both of these mate-
rial parameters are known, it may be possible to quantitatively predict the
material behavior, that is, if a material is rather ductile or brittle.
Dislocation-based processes and cleavage are not the only mechanism for
deformation of materials. Materials under geometric confinement, also referred
to as materials in small dimensions or interfacial materials, often show a
1 Introduction 15
dramatically different behavior. This is caused by the fact that the behav-
ior the carriers of plasticity or fracture changes at very small dimensions, so
that they either shut down or other mechanisms are activated. Examples for
such materials are nanocrystalline materials [68, 69] or ultra-thin submicron
metallic films [51].
It was shown that in materials with ultra-small grain sizes of tens of
nanometers and below, deformation can be completely dominated by grain
boundary processes such as grain boundary diffusion at high temperatures or
by grain boundary rearrangements and grain boundary gliding [68–70]. Due
to the small sizes of the grains, dislocations cannot be generated (e.g., sources
for dislocations, including Frank–Read sources are too large to fit within a
grain, or because dislocations are energetically very expensive under very
small geometrical confinement [71–73]). Even though such material behav-
ior is still perceived as ductile (since materials can be bent without fracturing
catastrophically), no dislocation motion is required to mediate deformation.
Fig. 1.9 Overview over timescales and lengthscales associated with various problems
and applications of mechanical properties (adapted from [2])
hard indenter tip into the material is used to the extract information about the
load–displacement relationships, which enables one to characterize the mate-
rial hardness. The combination of these approaches with temperature control,
environmental chambers to control humidity enables the study of material
behavior at a diverse range of controlled conditions.
For more details, we refer the reader to several textbooks that describe
mechanical testing at macroscales and microscales (for example, [1, 65]).
Until more than 20 years ago, the scales accessible to mechanical test-
ing of materials was largely limited to millimeter scales. Certainly, material
properties of individual molecules or assemblies of molecules could not be
determined. The early 1990s marked a decade of many new developments that
enabled the study of material properties at ultra-small scales. The experi-
mental analysis of such nanomechanical properties at submicrometer scales
is possible with techniques such as the atomic force microscope (AFM),
nanoindentation, or optical tweezers. These techniques are based on the devel-
opment of finer instrumentation and control mechanisms that enable one to
characterize forces on the order of hundreds of pN, with displacements on
the order of nanometers. Optical tweezers are based on the concept of optical
traps, and allow one to probe comparable force and displacement levels. This
has led to new applications of the study of mechanical properties in a variety
of fields, including materials science, biology, and physics. Figure 1.10 pro-
vides a summary of experimental techniques, here focused on applications in
20 Atomistic Modeling of Materials Failure
Fig. 1.10 Experimental techniques for conducting mechanical tests in single cell and
single molecule biomechanics. Reprinted from Materials Science and Engineering
C, Vol. 26, C.T. Lim, E.H. Zhou, A. Li, S.R.K. Vedula and H.X. Fu, Experimental
techniques for single cell and single molecule biomechanics, pp. 1278–1288, copyright
c 2006, with permission from Elsevier
Mechanical properties are important for many areas in science and engi-
neering, ranging through a broad spectrum that includes biology, physics,
materials science, and medicine. Here we briefly review the significance of
mechanical properties of materials and structures via a brief review of a few
studies in different disciplines. Much recent interest in mechanical properties is
sparked by the enhanced experimental capabilities that enables a simultaneous
comparison between computational and experimental testing, and the ability
to probe fundamental deformation mechanisms at the scale of molecular unit
events.
1 Introduction 21
also substantially larger than the values reported for nanocrystalline metals.
Such AFM studies therefore provide important insight into the deformation
mechanisms of metals at ultra-small lengthscales.
Similar AFM-based testing has been carried for several other nanomate-
rials and nanostructures, including carbon nanotubes of different geometry,
ceramic nanowires, and thin films. In many of these studies, the behavior was
found to differ substantially from the properties known from the macroscopic
counterparts.
Fig. 1.12 Mechanical deformation of a red blood cell (RBC) with optical tweezers.
Subplot (a) depicts a schematic of the experimental approach. Subplot (b) depicts
optical images of a healthy RBC anda RBC in the schizont stage of malaria, in PBS
solution at 25◦ C. The left column depicts results prior to stretching, the middle
column depicts results at a constant force of 68 ± 12 pN, and the right column
plots results at a constant force of 151 ± 20 pN. The P. falciparum malaria parasite
can be seen inside the infected RBCs. Reprinted from Acta Biomaterialia, Vol.
1, S. Suresh, J. Spatz, J.P. Mills, A. Micoulet, M. Dao, C.T. Lim, M. Beil, T.
Seufferlein, Connections between single-cell biomechanics and human disease states:
gastrointestinal cancer and malaria, pp. 15–30, copyright c 2005, with permission
from Elsevier
a drastic change of the mechanical properties of cancer cells has been observed
by atomic force microscopy analysis of the mechanical stiffness [4]. Figure 1.14
displays some of the images and analyses, in particular illustrating the change
of mechanical properties of cancer cells. The analysis suggests that tumor cells
feature a lower Young’s modulus. The analysis method reported in [4] might
be used to diagnose cancer based on a mechanical analysis, enabling one to
develop new inexpensive methods of cancer detection. The medical signifi-
cance of using mechanical signals as a method to diagnose cancer is based
on the fact that the appearance of cancer is biochemically extremely diverse,
making it difficult to find unique biochemical signatures for detection. In con-
trast, the analysis reported in [4] suggests that there exists a signature of the
change of mechanical properties that is common to variety of cell types, even
for different tumor types and different patient fluids.
These effects of a variety of phenomena on the mechanical properties illus-
trate the possibility to provide new paths toward diagnosis of diseases, and
maybe also for the development of treatment methods.
The mechanical deformation of cells and tissues plays an integral role in
the functioning of organisms. For example, the mechanical deformation of
24 Atomistic Modeling of Materials Failure
Fig. 1.13 Images of an RBC being stretched from 0 to 193 pN. Subplot (a) shows
images obtained from experiment, while subplots (b) and (c) depict a top view and
a three-dimensional view of the half-model corresponding to the large deformation
finite element simulation of the RBC, respectively. The contours of shades of grey
in the middle column shows the distribution of constant maximum principal strains.
Reprinted from Materials Science and Engineering C, Vol. 26, C.T. Lim, E.H. Zhou,
A. Li, S.R.K. Vedula and H.X. Fu, Experimental techniques for single cell and single
molecule biomechanics, pp. 1278–1288, copyright c 2006, with permission from
Elsevier
Fig. 1.14 Change of cellular mechanical properties in cancer cells. Subplot (a)
depicts an optical image demonstrating the round, balled morphology of visually
assigned tumor cells, and the large, flat morphology of presumed benign mesothelial,
normal cells. Subplots (b–d) show histograms of the associated Youngs modulus E
for cytological samples collected from patients with suspected metastatic cancer.
Subplot (b) shows the histogram of E for all data collected from seven different
clinical samples, indicating that there exist two peaks in the distribution. Subplot
(c) shows a Gaussian fit for all tumor cells, and subplot (d) shows a log-normal fit
for all normal cells. The analysis suggests that the presence of tumor cells leads to a
sharp peak due to a lower Young’s modulus. This method might be used to diagnose
cancer based on a mechanical analysis. Reprinted with permission from Macmillan
Publishers Ltd, Nature Nanotechnology [4] c 2007
(e.g., SEM, AFM imaging) provides key advances in the understanding of how
molecular and multiscale hierarchical features in these materials participate
in their deformation.
The analysis of mechanical properties is not limited to tissue or cellular
scales. Recent experimental and computational progress enabled the study
of the deformation and fracture behavior of individual protein molecules. For
example Fig. 1.16 depicts the force–extension profile of an immunoglobulin-like
domains of human cardiac titin [6]. Fig. 1.16a depicts the large-deformation
regime, showing characteristic peaks that correspond to repeated unfolding of
protein domains. Fig. 1.16b depicts a close-up view of the first peak.
26 Atomistic Modeling of Materials Failure
Fig. 1.15 Experimental study of fracture mechanics of bone. Subplot (a) shows the
geometry of the three-point bending test with an initial notch as a seed for frac-
ture. Subplots (b) and (c) show scanning electron micrographs of microscopic bone
fracture mechanisms, obtained from carrying out fracture experiments (arrangement
of figure adapted from original source). Reprinted with permission from Macmillan
Publishers Ltd, Nature Materials [5]
c 2003
Fig. 1.16 AFM experiments of protein unfolding. Subplot (a): Force peaks corre-
sponding to the sequential unfolding of a immunoglobulin-like domains of human
cardiac titin (human cardiac I band titin encompassing the immunoglobulin-like
domains I27–I34). The results show large hump-like deviations from the WLC model
of entropic elasticity (continuous lines indicate WLC fits, the arrow illustrates the
point of deviation). Subplot (b): Detailed view of the first force peak of a saw-
tooth pattern. Reprinted with permission from Macmillan Publishers Ltd, Nature [6]
c 1999
comprehensive and inclusive review with respect to the wider range of avail-
able results. Rather, the examples used in this book should be regarded as
examples to illustrate application of the numerical techniques reviewed in the
book.
Part II
2.1 Introduction
Molecular dynamics simulation techniques are very widely applicable, for
a range of materials and states, including gases, liquids, and solids. The
first molecular dynamics studies were focused on modeling thermodynamical
behavior of gases and liquids in the 1950s and 1960s [81–84]. It was not until
the early 1980s when the first molecular dynamics modeling works were pub-
lished that were applied to the mechanical behavior of solids. As a consequence
of the general applicability of molecular dynamics, many of the methods and
approaches described in this book can also be useful for the study of gases and
liquids, as well as the interaction of those with solids in systems that contain
both solids and liquids.
The outline of this chapter is as follows. We begin with a presentation of
the basic formulation of molecular dynamics (sometimes also referred to as
“MD”). After discussing of the numerical strategies associated with molec-
ular dynamics, we introduce interatomic potentials that mimic the energy
landscape predicted by quantum mechanics (we emphasize here that quan-
tum mechanics will not be discussed explicitly in this book, and the reader
32 Atomistic Modeling of Materials Failure
Fig. 2.1 Molecular dynamics can be used to study material properties at the inter-
section of various scientific disciplines. This is because the notion of a “chemical
bond” as explicitly considered in molecular dynamics provides a common ground as
it enables the cross-interaction between concepts used in different disciplines (here
exemplified for the disciplines of biology, mechanics, materials science and physics)
The significance of the atomistic viewpoint for failure processes and the enor-
mous computational burden associated with such problems makes modeling
and simulation of failure a promising and exciting area of research. In this
section we discuss some fundamental concepts associated with model build-
ing and the solution of the particular numerical problems to be computed in
molecular dynamics simulations.
The terms modeling and simulation are often used in conjunction with
the numerical solution of physical problems. However, it is important to note
that the two words have quite distinct meanings. The term modeling refers
2 Basic Atomistic Modeling 33
Fig. 2.2 This figure illustrates the concept of model building. Panel (a) on the left
shows the physical situation of a map of the subway lines. This representation makes
it quite difficult to determine a strategy to use the subway system to travel from
the cities of Braintree to Revere, for instance. The model representation depicted
in panel (b) on the right enables one to determine quite easily which subway line
to take, where to change the subway line, and how many subway stops there are
in between. This example illustrates that even though the model representation on
the right misrepresents the actual distances and directions, it elegantly displays the
connectivity. This figure was created based on a snapshot from the Massachusetts
Bay Transportation Authority (MBTA) web site (URL: http://www.mbta.com/),
reprinted with permission from the the Massachusetts Bay Transportation Authority
along with a definition of how atoms interact with each other (for a discussion
of Newton’s law, see Sect. 3.1). Despite this quite simple basis, very complex
phenomena can be simulated. Unlike many continuum mechanics approaches,
atomistic techniques require no a priori assumption on the defect dynamics
or its behavior.
Once the atomic interactions are chosen, the entire material behavior is
determined. This aspect of atomistic modeling provides a terrific opportunity
to build insightful models, that is, models that capture the essentials, elucidate
fundamental mechanisms, and thereby provide an elegant representation of
the key principles that underline the key observations. While in some cases it
is difficult to find an appropriate interatomic potential for a material, atomic
interactions can also be chosen such that generic properties common to a large
class of materials are incorporated (e.g., describing a general class of brittle or
ductile materials). This approach refers to the design of “model materials” to
study specific materials phenomena. Despite the fact that such model building
has been in practice in fluid mechanics for many years, the concept of “model
materials” in materials science is relatively new. On the other hand, atomic
interactions can also be chosen very accurately for a specific atomic interaction
using quantum mechanical methods such as the density functional theory
(DFT) [87], which enables one to approximate the solutions to Schrodinger’s
equation for a particular atomistic model.
Richard Feynman has also emphasized the importance of the atomistic
viewpoint in his famous Feynman’s Lectures in Physics [88], where he stated
that the atomic hypothesis (or the atomic fact, whatever you wish to call it)
that all things are made of atoms – little particles that move around in per-
petual motion, attracting each other when they are a little distance apart,
but repelling upon being squeezed into one another [...] provides an enormous
amount of information about the world, if just a little imagination and thinking
are applied.”
This underlines atomistic simulations as a natural choice to study materials
at a fundamental level. This is particularly true for studies of materials failure!
The atomistic level provides the most fundamental, sometimes referred to as
the ab initio, description of the failure processes. Many materials phenom-
ena are multiscale phenomena. For a fundamental understanding, simulations
should ideally capture the elementary physics of single atoms and reach length
scales of thousands of atomic layers at the same time. This can be achieved
by implementation of numerical models of atomistic models on very large
computational facilities.
In atomistic simulations, the goal is to predict the motion of each atom in the
material, characterized by the atomic positions ri (t), the atomic velocities
vi (t), and their accelerations ai (t) (see Fig. 2.3). Each atom is considered
as a classical particle that obeys Newton’s laws of mechanics. The collective
behavior of the atoms allows one to understand how the material undergoes
deformation, phase changes, or other phenomena by providing links between
the atomic scale to meso- or macroscale phenomena. Extraction of information
from atomistic dynamics can be challenging and typically involves methods
rooted in statistical mechanics.
The total energy of such as system is
H = K + U, (2.6)
where K is the kinetic energy of the entire system and U the potential energy.
The kinetic energy is a function of the kinetic energy of all N particles,
38 Atomistic Modeling of Materials Failure
1
N
K= mi v2i , (2.7)
2 i=1
and the total potential energy is the sum of the potential energy of all particles:
N
U (r) = Ui (r), (2.8)
i=1
noting that the potential energy of each particle Ui depends on the position
of itself and all other particles in the system, expressed by r = {ri } as defined
above. For now we leave this expression as an unknown.
The total energy H is also referred to as the Hamiltonian. We note that
K = K(p) and U = U (r), that is, the kinetic energy depends only on the
velocities or the linear momenta of the particles and the potential energy is a
function only of the position vectors.
To satisfy Newton’s law Fi = mi ai for each particle i in the system, the
following equation governs the dynamics of the system:
d2 ri dU (r)
mi 2
=− . (2.9)
dt dri
The right-hand side corresponds to the gradient of the potential energy, which
is the force (note that the potential energy of the system depends on the
positions of all particles r). Equation (2.9) represents a system of coupled
second-order nonlinear partial differential equations, corresponding to a cou-
pled system N -body problem for which no exact solution exists when N > 2.
2 Basic Atomistic Modeling 39
Equation (2.13) provides a direct link between new positions (at t0 + ∆t) and
the old positions and accelerations (at t0 ). The accelerations can be obtained
from the forces by considering Newton’s law,
Fi
ai = . (2.14)
m
This updating scheme is referred to as the Verlet central difference method.
There exist many other integration schemes that are frequently used
in molecular dynamics implementations. In the following few sections we
summarize a few additional popular algorithms.
Leap-Frog Algorithm
When the equations reviewed in the previous section are integrated, the
resulting thermodynamical ensemble is N V E, which means that the parti-
cle number N , the system volume V , and the total energy of the system E
remain constant throughout the simulation. Other thermodynamical ensem-
bles can be realized by modifying the equations of motion in an appropriate
way, leading to N V T or N P T ensembles. Table 2.1 shows an overview over
various thermodynamical ensembles.
Fig. 2.4 Schematic of the atomic displacement field as a function of time. The
atomic displacement field consists of a low-frequency (“coarse”) and high frequency
part (“fine”)
42 Atomistic Modeling of Materials Failure
A critical step in solving the dynamical equations (for instance, using the
Velocity Verlet scheme) is to consider the size of the time step. Figure 2.4
depicts a schematic of the atomic displacement field as a function of time.
As can be seen, the displacement history consists of low- and high-frequency
contributions, where the total displacements can be written as
with u (t) as the fine contribution and u(t) as the coarse part. To solve the
equations of motion, the fine part needs to be discretized, which results in
a significant computational burden as most systems require time steps of
approximately 1 fs or 10−15 s to discretize the rapid oscillations of u (t). Inter-
atomic bonds that involve relatively light hydrogen atoms sometimes require
even smaller time steps on the order of 0.1 fs.
There are also adaptive techniques that are based on the idea to dynami-
cally change the time step in a simulation, depending on the maximum atomic
velocities [18]. Such approaches may help to increase the efficiency of molecular
dynamics studies without adversely affecting the results.
Fig. 2.5 Example of harmonic oscillator with spring constant k = φ (r = r0 ), used
to extract information about the time step required for integration of the equations
of motion. The dashed line shows the (nonlinear) realistic potential function between
a pair of atoms, of which the harmonic oscillator is the second-order approximation.
To estimate the time step for a particular system, one can estimate the
oscillation frequency of a harmonic oscillator:
1 k
ν= (2.22)
2π m
2 Basic Atomistic Modeling 43
where k is the spring constant, given by the second derivative of the potential
function with respect to the diameter (k = φ (r = r0 )), evaluated at r = r0 .
The function φ describes how the energy of the bond changes as a function of
radius r. This is schematically shown in Fig. 2.5. The time step should then
be chosen
1
∆tmin (2.23)
ν
In summary, the time step ∆t needs to be small enough to model the vibra-
tions of atomic bonds correctly. The vibration frequencies may be extremely
high, in particular for light atoms as
√
∆tmin ∼ m (2.24)
The stiffer the bond is around its equilibrium position, the lower the critical
time step as
1
∆tmin ∼ √ . (2.25)
k
The fact that the time step is on the order of several femtoseconds has major
implications on the time scale molecular dynamics can reach. For example,
approximately 1,000,000 integration steps are needed to calculate a trajectory
that covers 1 ns, providing a severe computational burden. Further, the time
step typically cannot be varied during simulation. The total time scale that can
be reached by molecular dynamics is typically limited to a few nanoseconds.
Some exceptionally long simulations have been reported that cover up to a few
microseconds. However, such simulations typically take up to several months.
This aspect of molecular dynamics is sometimes referred to as the time
scale dilemma. Even though the number of atoms in a simulation can be
easily increased by adding more processors (e.g., using parallel computing),
time cannot easily be parallelized. As can be see in the updating scheme (2.13),
an atomistic system is generally not independent in time: The behavior at t0
influences the state at t1 > t0 , and the time stepping cannot be carried out
independently, on multiple processors.
Several researchers are currently developing techniques such as the tem-
perature accelerated dynamics method, parallel replica method, and many
others to overcome this limitation, and make use of massively parallelized
computing to expand the accessible time scales [89]. Some of these techniques
will be discussed later.
statics. Such approaches have been used to study dislocation nucleation from
crack tips or the deformation of carbon nanotubes. It mimics a quasistatic
experiment, albeit neglecting the effect of finite temperature.
A variety of algorithms exist to perform energy minimization, most notably
conjugate gradient methods or steepest descent methods. Figure 2.6 depicts an
example result of an energy minimization, showing how the potential energy
of the system decreases systematically with the number of iteration steps and
finally converges to a finite value.
Fig. 2.8 Summary of the Metropolis–Hastings Monte Carlo algorithm. Please see
Figure 2.7 for an illustration of how state B is generated based on a random pertur-
bation from state A. The procedure is repeated NA times, the number of iterations.
The number of steps is chosen so that convergence of the desired property is achieved
Fig. 2.9 Schematic of the typical characteristic of a chemical bond, showing repul-
sion at small distances below the equilibrium separation r0 , and attraction at larger
distances
featuring many different chemical bonds with varying strength, such as exem-
plified in materials including cement, proteins, or ceramics, or at interfaces
between metallic systems and organic components.
Despite the differences between different chemical bonds, many atom–atom
interactions show similar characteristic featyres. Figure 2.9 depicts the typical
characteristic of a chemical bond, showing repulsion at small distances below
the equilibrium separation r0 of a pair of atoms, and attraction at larger
distances.
In general, for any material we must consider the interplay of chemical
interactions that include, ordered by their approximate strength:
• Covalent bonds (due to overlap of electron orbitals, e.g., found in carbon
nanotubes, C–C bond, organic molecules such as proteins)
• Metallic bonds (found in all metals, e.g., copper, gold, nickel, silver)
• Electrostatic (ionic) interactions (Coulombic interactions, e.g., found in
ceramics such as Al2 O3 or in SiO2 )
• Hydrogen-bonds (e.g., found in polymers, proteins), as well as
• Weak or dispersive van der Waals (vdW) interactions (e.g., found in wax).
Electrostatic interactions can be significantly weakened by screening due
to electrolytes, which can lead to interactions that are weaker than vdW
interactions.
48 Atomistic Modeling of Materials Failure
Fig. 2.10 Atoms are composed of electrons, protons, and neutrons. Electrons and
protons are negative and positive charges of the same magnitude. In classical molec-
ular dynamics, the three-dimensional atom structure is replaced by a single mass
point
Fig. 2.11 Overview over different simulation tools and associated lengthscale and
timescale
There is no single approach that is suitable for all materials and for all
materials phenomena. The choice of the interatomic potential depends very
strongly on both the application and the material. Popular choices in par-
ticular for modeling mechanical properties of materials are semiempirical
or empirical methods, which typically allow one to simulate large systems
with many thousands to billions of particles. However, to address different
aspects of the mechanical behavior of a specific material typically requires the
application of a range of simulation approaches.
An overview over the most prominent materials simulation techniques
is shown in Fig. 2.11. In the plot we also indicate which lengthscale and
timescale can be reached with the various methods. The methods included
in the figure refer to quantum mechanics based methods, classical molecular
50 Atomistic Modeling of Materials Failure
1
N N
Utotal = φ(rij ), (2.27)
2 j=1
i=j=1
where rij is the distance between particles i and j. Note the factor 1/2 to
account for the double counting of atomic bonds. The procedure of summing
up the energies is shown in a schematic in Fig. 2.12.
The term φ(rij ) describes the potential energy of a bond formed between
two atoms, as a function of its distance rij . How can one obtain expressions
for φ? A possible approach is sketched in Fig. 2.13, showing how a full-electron
representation of the pair of atoms is reduced to a case where two point par-
ticles interact. The energy–distance relationship must be identical in both
cases. An approach often used is to carry out quantum mechanical calcula-
tions that provides the relationship between distance and energy of a pair of
2 Basic Atomistic Modeling 51
1
2 5
1
2 5
4
Fig. 2.12 Pair interaction approximation. The upper part shows all pair interactions
of atom 1 with its neighbors, atoms 2, 3, 4, and 5. When the bonds to atom 2 are
considered, the energy of the bond between atoms 1 and 2 is counted again (bond
marked with thicker line). This is accounted for by adding a factor 1/2 in (2.27)
atoms, which is then used to determine the parameters of the pair potential
expression. Pair potentials must capture the repulsion at short distances due
to the increasing overlap of electrons in the same orbitals, leading to high
energies due to the Pauli exclusion principle. At large distances, the poten-
tial must capture the effect that atoms attract each other to form a bond. In
many pair potentials, two separate terms are used to describe repulsion and
attraction, and the sum of these repulsive and attractive interactions yield
the total energy dependence on the radius:
φ = φRepulsion + φAttraction . (2.28)
A pair of atoms is in the equilibrium position at a balance between the
attractive and repulsive terms.
Pair potentials are the simplest choice for describing atomic interactions.
Even so, in some materials the interatomic interactions are best described
by pair potentials (because the underlying quantum mechanical governing
equations actually predict such a behavior). Prominent examples include the
noble gases (e.g., argon, neon) [83] as well as Coulomb interactions due to
partial charges. Pair potentials have also proven to be a reasonable model
for more complex materials such as SiO2 [102]. The potential energy of an
individual atom is given by
52 Atomistic Modeling of Materials Failure
Ni
Ui = φ(rij ) (2.29)
j=1
The LJ potential can be fitted to the elastic constants and lattice spacing
of metals (however, this model has some shortcomings with respect to the
stacking fault energy and the elasticity of metals). The term with power 12
represents atomic repulsion, and the term with power 6 represents attractive
interactions. The parameter σ scales the length and 0 the energy of atomic
bonds. Often, pair potentials are cutoff smoothly with a spline cutoff function
(see for instance [104] or [105]).
For the LJ potential, the equilibrium distance between atoms (denoted as
r0 ) is given by √
r0 = σ 6 2. (2.31)
The maximum force between two atoms is
2.3940
Fmax,LJ = . (2.32)
σ
Figure 2.14 shows a plot of the LJ potential (dotted line) and its deriva-
tive (continuous line, describing the interatomic forces), in a parametrization
2 Basic Atomistic Modeling 53
Fig. 2.14 Plot of the LJ potential and its derivative (for interatomic forces) in a
parametrization for copper as reported in [9]
1
φ(rij ) = a0 + k(rij − r0 )2 (2.34)
2
54 Atomistic Modeling of Materials Failure
The concept that the total energy of the system is simply a sum over the energy
contributions between all pairs of atoms in a system is a great simplification
that leads to great challenges. For example, at a surface of a crystal, the
atomic bonds may have different properties than in the bulk. Pair potentials
are not capable of capturing this effect. The limitation of pair potentials to
model more complex situation, in particular the dependence of the properties
of chemical bonds between pairs of atoms on the environment, is sketched
in Fig. 2.15. This behavior is particularly important for metals, because of
quantum mechanical effects that describe the influence of the electron gas.
Fig. 2.15 Difference in bond properties at a surface. Pair potentials (left panel) are
not able to distinguish bonds in different environments, as all bonds are equal. To
accurately represent the change in bond properties at a surface, one needs to adapt
a description that considers the environment of an atom to determine the bond
strength, as shown in the right panel. The bond energy between two particles is then
no longer simply a function of its distance, but instead a function of the positions
of all other particles in the vicinity (that way, changes in the bond strength, for
instance at surfaces, can be captured). Multibody potentials (e.g., EAM) provide
such a description
longer simply a function of its distance, but instead a function of the posi-
tions of all other particles in the vicinity. This behavior can be captured
in multibody potentials. The idea behind multibody potentials is to incor-
porate more specific information on the bonds between atoms than simply
the distance between two neighbors. In such potentials, the energy of bonds
therefore depends not only on the distance of atoms, but also on its local
environment, that is, on the positions of neighboring atoms. In the case of
metals, the interactions of atoms can be quite accurately described using
potentials based on the embedded atom method (EAM), or other so-called
n-body potentials (e.g., [94, 107]. Several variations of the classical EAM
potentials exist [108, 109]). Another, similar approach is based on effective
medium theory (EMT) [110, 111]. Particularly appropriate models have been
reported for metals such as copper and nickel. Other metals (e.g., aluminum)
are more difficult to model with such approaches [112, 113].
An EAM potential for metals is typically given in the form
Ni
Ui = φ(rij ) + f (ρi ), (2.35)
j=1
where ρi is the local electron density and f is the embedding function. The
electron density ρi depends on the local environment of the atom i, and
the embedding function f describes how the energy of an atom depends on
the local energy density. The electron density itself is typically calculated
based on a simple pair potential that maps the distance between atoms to
the corresponding contribution to the local electron density. The potential
features a contribution by a two-body term φ to capture the basic repulsion
and attraction of atoms (just like in LJ or Morse potentials), in conjunction
with a multibody term that accounts for the local electronic environment of
the atom.
Overall multibody potentials allow a much better reproduction of the
elastic properties of metals than pair potentials (e.g., [109]). For instance,
most real materials violate the Cauchy relation (that is, the condition that
c1122 = c1212 ). Any pair potential predicts an agreement with the Cauchy rela-
tion [109]. Multibody potentials are capable of reproducing the appropriate
elastic behavior. Figure 2.16 illustrates how an EAM-type multibody potential
can represent different effective pair interactions between bonds at a surface
and in the bulk.
However, most conventional multibody potentials are not capable of mod-
eling any effect of directional bonding. Whereas this is not important for
metals such as Ni or Cu, this becomes quite significant in materials with a
more covalent character of the interatomic bonding. To address these effects,
modified embedded atom potentials (MEAM) have been proposed that can
be parameterized, for instance, for silicon [114].
56 Atomistic Modeling of Materials Failure
Fig. 2.16 This plot illustrates how an EAM-type multibody potential can represent
different effective pair interactions between bonds at a surface and in the bulk
The bases for simulations of polymers, organic substances, or proteins are force
fields that describe the various chemical interactions based on a combination
of energy terms. This is required since in these materials, the set of char-
acteristic chemical bonds is much more diverse than in metals, for instance,
which requires the explicit consideration of ionic, covalent, and vdW interac-
tions. Figure 2.17 illustrates this chemical complexity, exemplified in a small
alpha-helical coiled coil protein domain.
A prominent example for this approach is the classical force field CHARMM
[115]. The CHARMM force field is widely used in the protein and biophysics
community, and provides a reasonable description of the behavior of proteins.
This force field is based on harmonic and anharmonic terms describing
covalent interactions, in addition to long-range contributions describing vdW
interactions, ionic (Coulomb) interactions, as well as hydrogen bonding. Since
the bonds between atoms are modeled by harmonic springs or its variations,
bonds (other than H-bonds) between atoms cannot be broken, and new bonds
cannot be formed. Also, the charges are fixed and cannot change, and the
equilibrium angles do not change depending on stretch. The CHARMM force
field belongs to a class of models with similar descriptions of the interatomic
forces; other models include the DREIDING force field [116], the UFF force
field [117], the GROMOS force field, or the AMBER model (see [118], for
instance, for a review of various force fields for biological systems).
In the CHARMM model, the mathematical formulation for the empirical
energy function that contains terms for both internal and external interactions
2 Basic Atomistic Modeling 57
Fig. 2.18 Schematic of the contributions of the different terms in the potential
expressions given in (2.36), illustrating the contributions of bond stretching, angle
bending, bond rotations, electrostatic interactions, and vdW interactions
58 Atomistic Modeling of Materials Failure
and
Utorsion = φtorsion . (2.41)
quadruples
Due to the harmonic approximation, these expressions are only valid for
small deformation from the equilibrium configuration of the bond. Large
deformation or fracture of these bonds cannot be described. Figure 2.18
schematically illustrates the energy contributions provided in (2.36).
The Coulomb energies are evaluated between pairs of atoms and are
described as
qi qj
φCoulomb = (2.42)
1 rij
where qi and qj are the partial charges of atoms i and j, and 1 is the effective
dielectric constant (1 = 1.602 × 10−19 C for vacuum). The total Coulomb
energy is then given by
2 Basic Atomistic Modeling 59
N N
qi qj
UCoulomb = (2.43)
r
j=1 1 ij
j=i=1
Table 2.2 Overview over force fields suitable for organic substances
atoms is not constant, but depends on the local environment, similar to the
EAM approach (please see [119] for a discussion). However, specific terms
are included to specify the directional dependence of the bonding. The idea
is based on the concept of mapping bond distances to bond orders, which
enables one to determine specific quantum chemical states of a molecular
structure. The concept of bond order was initially introduced by Pauling
[120]. Very well-known models in this class of potentials are the reactive
bond order potentials (REBO) [121, 122], the Tersoff potential [10, 123], the
Stillinger–Weber potential [124], Brenner’s force fields [125, 126], the Stu-
art reactive potential [127], and more recently, the ReaxFF reactive force
field [93].
Abell–Tersoff Approach
The basic concept of bond order potentials is simple to explain. The key idea
is to modulate the bond strength based on the atomic environment, taking
advantage of some theoretical chemistry principles.
Instead of expressing φ(rij ) as a harmonic function or an LJ function (see
above), in the Abell–Tersoff approach the interaction between two atoms is
expressed as
Fig. 2.19 The plot shows the cohesive energy per atom (upper plot, in eV) and
the bond length (lower plot, in Å), for several real and hypothetical polytypes of
carbon, comparing the predictions from the Tersoff potential [10] for C with experi-
mental and other computational results. The structures include a C2 dimer molecule,
graphite, diamond, simple cubic, BCC, and FCC structures. The squares correspond
to experimental values for these phases and calculations for hypothetical phases [11].
The circles are the results of Tersoff’s model [10]. The continuous lines are spline
fits to guide the eye. Reprinted from: J. Tersoff, Empirical interatomic potentials
for carbon, with applications to amorphous carbon, Physical Review Letters, Vol.
61(25), 1988, pp. 2879–2883. Copyright c 1988 by the American Physical Society
the Pauling relationship between bond length and bond order. Abell suggested
that
Mij ∼ Z −δ (2.45)
where δ depends on the particular system and Z is the coordination number of
atom i that depends on the bond radius. For pair potentials, Mij ∼ Z, which
is not true for many real materials. The Abell–Tersoff approach provides a
realistic model for these effects.
In the Tersoff potential [10, 123], the many-body term depends explicitly
on the angles of triplets of atoms, in addition to considering the effect of
coordination. Thus, this and related potentials are also referred to as three-
body potentials. The explicit angular dependence also illustrates the difference
to EAM potentials: Here, the multibody term solely depends on the elec-
tron density, and not on any directional information. Tersoff has successfully
parametrized his potential approach for carbon, silicon, and other semicon-
ductors. Figure 2.19 shows the cohesive energy per atom for several real and
62 Atomistic Modeling of Materials Failure
Fig. 2.20 An example to demonstrate the basic concept of the ReaxFF potential.
It has been developed to accurately describe transition states in addition to ground
states
2 Basic Atomistic Modeling 63
torsion are included in Utors . The term Uconj describes energetic contributions
of resonance effects. A maximum contribution of the conjugation energy is
obtained when successive bonds have bond order values of 1.5, as it is the
case in benzene, for instance. H-bonds are treated in the term UH-bonds , and
its interactions are calculated between groups X–H and Y, where X and Y
are atoms that can form H-bonds (for instance, N, O). Nonbonded two-body
interactions are included in UvdW and in UCoulomb . They are included for all
atom pairs whether they are bonded or nonbonded. This is important to avoid
energy discontinuities when chemical reactions occur. To enable the calcula-
tion for these interactions for all atom pairs, a shielded Coulomb potential of
the form
Ni Ni
qi qj
φi = 3 + γ −3 )1/3
(2.48)
i=1 j=1
1 (rij ij
This correction refers to the fact that carbon, for instance, cannot have more
than four bonds, or hydrogen cannot have more than one bond. This expres-
sion illustrates that the bonding term is a multibody expression, as it depends
on all j neighbors of an atom i.
Once the bond orders are known, the energy contributions can be calcu-
lated. For the term considered here,
φbond, ij = −De BOij exp pbe,1 1 − BOpbe,1 ij , (2.51)
where De and pbe,i are additional bond parameters. The total bond energy is
then given by a summation over all bonds in the system,
2 Basic Atomistic Modeling 65
Fig. 2.21 Illustration of basic concept of bond order potentials. Subplot (a) shows
how the bond order potential allows for a more general description of chemistry, since
all energy terms are expressed dependent on bond order. In contrast, conventional
potentials (such as LJ, Morse) express the energy directly as a function of the bond
distance as shown in subplot (b). Subplot (c) illustrates the concept for a C–C single,
double, and triple bond, showing how the bond distance is used to map to the bond
order, serving as the basis for all energy contributions in the potential formulation
defined in (2.47)
Ubond = φbond,ij (2.52)
all bonds ij
Equation (2.51) also illustrates that the energy contributions vanish when the
bond order goes to zero, which corresponds to a broken chemical bond.
All other terms are also expressed as a function of bond orders. For
instance, the angle contributions are given as
Fig. 2.22 Results of a ReaxFF study of water formation, comparing the production
rate with and without a Pt catalyst. The presence of the Pt catalyst significantly
increases the water production rate (results taken from [12])
An example simulation is shown in Fig. 2.22 [12]. This plot depicts the
results of a ReaxFF study of water formation, comparing the production rate
with and without a Pt catalyst (same pressure and temperature, the only
difference is the catalyst). It is evident that the presence of the Pt catalyst
significantly increases the water production rate.
2 Basic Atomistic Modeling 67
Fig. 2.23 Water production at varying temperature, for constant pressure. Subplot
(a) depicts the water production rate. Subplot (b) shows an Arrhenius analysis,
enabling us to extract the activation barrier for the elementary chemical process of
12 kcal/mol. This result is close to DFT level calculations of the energy barrier [12]
68 Atomistic Modeling of Materials Failure
Figure 2.23 depicts an analysis of the system with Pt catalyst, for vari-
ations of the temperature. These simulations involves thousands of reactive
atoms, a computational task that cannot be achieved using DFT or simi-
lar approaches. Thus, the ReaxFF approach provides a very useful bridge
between quantum mechanical methods and empirical potentials, as illus-
trated in Fig. 2.24. ReaxFF simulations have been reported with system sizes
approaching millions of atoms, as recently reported in the ReaxFF parallelized
algorithm [133].
Fig. 2.24 The ReaxFF force field fills a gap between quantum mechanical methods
(e.g., DFT) and empirical molecular dynamics
Fig. 2.25 Schematic of the numerical scheme in carrying out molecular dynamics
simulations
Figure 2.25 depicts the basic numerical scheme of carrying out a molecular
dynamics study. The basic steps of a molecular dynamics simulation are
• Define initial conditions and boundary conditions (including positions and
velocities at t = t0 ); typically the velocities of particles are drawn from a
Maxwell–Boltzmann distribution.
70 Atomistic Modeling of Materials Failure
Fig. 2.26 Schematic of the numerical scheme in carrying out molecular dynamics
simulations
We recall that at each integration step, forces are required to obtain acceler-
ations. Forces are calculated from the positions of all atoms, by considering
the atomic potential energy surface U (r) that depends on the positions of all
atoms. According to (2.9), the force vector is given by taking partial deriva-
tives of the potential energy surface with respect to the atomic coordinates of
the atom considered,
dU (r)
Fi = − . (2.54)
dri
Fig. 2.27 Schematic of force calculation scheme in molecular dynamics, for a pair
potential. To obtain the force vector F one takes projections of the magnitude of
the force vector F into the three axial directions xi (this is done for all atoms in the
system)
For the special case where the interatomic forces are described by a poten-
tial φ that depends only on the distance between pairs of atoms, here denoted
by r, the contributions to the total force vector due this particular interaction
can be obtained by taking projections of the magnitude of the force vector F
into the three axial projections ri of the vector between the pair of particles.
The magnitude of the force vector due to interactions of pairs of atoms is then
given by
dφ(r)
F = , (2.55)
dr
where F =| F |. The ith component of the force vector is then given by
ri
Fi = −F . (2.56)
r
Figure 2.27 depicts this approach schematically.
In principle, all atoms in the system interact with all other atoms, which
requires a nested loop for calculation of the force vectors of all atoms. This
renders the total computational time requirement to solve the problem second
order with respect to the number of particles N ,
72 Atomistic Modeling of Materials Failure
Fig. 2.28 Use of neighbor lists and bins to achieve linear scaling ∼N in molecular
dynamics. Panel (a): Example of how neighbor lists are used. The four neighbors
of the central atom (in the circle) are stored in a list so that force calculation can
be done directly based on this information. This changes the numerical problem to
a linear scaling effort. Panel (b): The computational domain is divided into bins
according to the physical position of atoms. Then, atomic interactions must only be
considered within the atom’s own bin and atoms in the neighboring bins
ttot ∼ N 2 . (2.57)
The first loop goes over all atoms, and the second loop goes through all
possible neighbors of each atom. The following pseudocode illustrates this
process:
for i=1 to N # loop over all atoms i
for j=1 to N (i not equal to j) # second loop over all atoms neighboring
atom i
r=distance(i,j) # calculate distance between atoms i and j
F=f(r) # calculate force depending on radius
that are the nearest, second nearest, and so forth neighbors of each particle.
This method is used to save time from checking every particle in the system
every time a force calculation is made. The list can be used for several time
steps before updating. Each update is expensive since it involves N × N oper-
ations for an N -particle system. In low-temperature solids where the particles
do not move very much, it is possible to perform an entire simulation without
or with only a few updating, whereas in simulation of liquids, updating every
5 or 10 steps is quite common. Figure 2.28a shows a schematic of how neighbor
lists are used. We note that neighbor lists can only be implemented if particles
interact only up to a certain cutoff radius; for very long range interactions,
the definition of neighbor lists may not be feasible.
An alternative to generation of neighbor lists is the decomposition of
the computational domain into bins. The size of the bin is chosen compa-
rable to the cutoff radius of the potential, so that atomic interactions must
only be considered within the atom’s own bin and the neighboring bins (see
Fig. 2.28b).
2 K
T = . (2.58)
3 N kB
Note that the numerical value for the Boltzmann constant kB = 1.3806503 ×
10−23 m2 kgs−2 K−1 , relating energy and temperature at the level of individual
atoms or particles (its units are energy per absolute temperature). Please see
also Table 4.7 for other units and their conversion to SI units.
74 Atomistic Modeling of Materials Failure
The first term stems from the kinetic contributions of particles hitting the
wall of the container, and the second term stems from the interatomic forces,
expressed as the force vector contribution multiplied by the distance vector
component between particles i and j.
r+ r r+ 2
r
2
N (r ± ∆r
2 )
g(r) = , (2.61)
Ω(r ± 2 )ρ0
∆r
where N (r ± ∆r
2 ) is the number of particles in the volume shell with volume
Ω(r ± ∆r
2 ).
2 Basic Atomistic Modeling 75
Fig. 2.30 Radial distribution function g(r) for various atomistic configurations,
including a solid (crystal), a liquid and a gas
1
N
∆r2 = (ri (t) − ri (t = 0))2 . (2.62)
N i=1
If averaged over all particles, the mean square displacement function pro-
vides the square distance that particles have moved during time t. The mean
square displacement function is zero at t = 0, and it grows with increasing
time. In a solid, it is expected that the mean square displacement function
grows to a characteristic value and then saturates at a constant value. In
a liquid, all atoms move continuously through the material, as in Brownian
motion. Diffusivity in liquids is related to the linear variation of the mean
square displacement function with time t. The diffusivity D can be calculated
as
1 d 2
D= lim ∆r , (2.63)
2d t→∞ dt
where d = 2 in 2D and d = 3 in 3D (the parameter d describes the
dimensionality).
76 Atomistic Modeling of Materials Failure
1 1
N N
v(0)v(t) = vi (t0 )vi (t0 + t), (2.64)
N i=1 N j=1
VAF
Ideal gas
1
Dense gas
Liquid
T
Solid
Fig. 2.31 Velocity autocorrelation function (VAF) for a gas, liquid, and solid
Fig. 2.32 Relating the continuum stress with the atomistic stress. The left shows
a continuum system in which σij (r) is defined at any point r. In contrast, in the
atomistic system the stress tensor is only defined at discrete points where atoms are
located
Note that the pressure P as discussed in Sect. 2.8.2 is a special case of the
stress tensor definition. The virial stress needs to be averaged over space and
time to converge to the Cauchy stress tensor. For further discussion on the
virial stress and other definitions of the Cauchy stress tensor (e.g., the Hardy
stress) we refer the reader to a review article [136].
F1
1 2 3
F2
r21 r23
Fig. 2.33 Example of how to calculate the stress tensor in a 1D system
Figure 2.33 depicts a schematic that shows how the stress tensor can be
calculated in a 1D system, for a nearest neighbor pair potential. Neglecting
the kinetic contribution (v ≈ 0) and with
dφ(r) ri
F =− (2.68)
dr r
as the force between two particles, the stress tensor coefficient σ11 is
11
σ11 = (F1 r21 + F23 r23 ) . (2.69)
V 2
huge challenges for data analysis and visualization, or just for data handling
and storage. However, for many properties of materials it is not necessary to
consider 1023 atoms. Many thermodynamical properties of solids or liquids
can be accurately described in systems of thousands of atoms or less, in some
instances. Many mechanical properties require larger system sizes, containing
at least tens of thousands to billions of atoms.
The need for military applications has strongly driven the development of
supercomputers. (This includes computers operated by the U.S. Department
of Energy, DoE, or the U.S. National Science Foundation, which are among the
most powerful supercomputers of the world. DoE computers are for instance
being used to maintain the U.S. stockpile of nuclear weapons.) Many other
large computational centers have been established in recent years in Europe
and Asia, including Japan’s Earth Simulator (which, for some time, was the
most powerful computer on Earth).
Early supercomputers, such as the Almaden Spark that was used in the
1960s were capable of simulating 100s of atoms. GFLOPS computing enabled
the simulation of millions of particles in the mid 1990s. TFLOPS systems led
to first simulations of systems that exceed one billion. The current state-of-the
art in atomistic simulation reaches system sizes of tens to hundreds of bil-
lions of atoms, corresponding to several micrometer-sized physical dimensions
80 Atomistic Modeling of Materials Failure
[133, 137–141] (see Fig. 2.34). However, only few computational groups or
scientists routinely carry out such large simulations.
Fig. 2.34 Increase in computer power over the last decades and possible system sizes
for classical molecular dynamics modeling. The availability of PFLOPS computers
is expected by the end of the current decade, which should enable simulations with
hundreds of billions of atoms
Fig. 2.35 Summary of top 10 of the TOP500 supercomputer list, as of Spring 2008
with a given particle, only the cell in which it is located and the next-nearest
neighbors have to considered. Since placing the particles in the correct cells
scales linearly with N , the problem originally scaling with N 2 can therefore be
reduced to N . With a parallel computer whose number of processors increases
with the number of cells (the number of particles per cell does not change),
the computational burden remains constant.
The speedup factor S is defined as the ratio of execution time on one
processor (Ts ) over the execution time on p processors (Tp ):
Ts
S= . (2.70)
Tp
The perfectly efficient parallel computer would exhibit linear speedup. This
would mean that the computation time for p processors is 1/p times the
execution time on one processor. However, the speedup depends strongly on
the fraction of the work done in parallel. We refer the reader to Plimpton’s
algorithms for molecular dynamics with short-range forces [143].
Figure 2.36 depicts a schematic and scaling results of a modern paralleliza-
tion scheme, referred to as the tunable hierarchical cellular decomposition
scheme (THCD) (see Fig. 2.36a) [133]. The THCD involves a more com-
plex domain decomposition that includes a hierarchy of parameterized cell
data/computation structures, and adaptive load balancing through wavelet-
based computational-space decomposition. The THCD was applied to paral-
lelize the ReaxFF reactive force field approach (referred to as F-ReaxFF). This
parallelization scheme enabled the simulation of 0.56 billion-atom F-ReaxFF
systems. Figure 2.36b depicts the total execution (circles) and communica-
tion (squares) times per molecular dynamics time step as a function of the
number of processors, illustrating that the parallel speedup is almost perfect
82 Atomistic Modeling of Materials Failure
Fig. 2.36 Modern parallelization scheme. Subplot (a) depicts the schematic of the
tunable hierarchical cellular decomposition scheme (THCD). The physical volume
is subdivided into process groups, PGγ , each of which is spatially decomposed into
processes, Pγπ . Each process consists of a number of computational cells (e.g., linked-
list cells in molecular dynamics). Subplot (b) shows the total execution (circles) and
communication (squares) times per molecular dynamics time step as a function of the
number of processors for the F-ReaxFF molecular dynamics algorithm with scaled
workloads (in a 36,288P atom RDX systems on P processors (P = 1, . . . , 1920)
of Columbia [Columbia is a supercomputer at NASA]). Reprinted from Computa-
tional Materials Science, Vol 38(4), A. Nakano, R. Kalia, K. Nomura, A. Sharma, P.
Vashishta, F. Shimojo, A. van Duin, W.A. Goddard III, R. Biswas and D. Srivastava,
A divide-and-conquer/cellular-decomposition framework for million-to-billion atom
simulations of chemical reactions, pp. 642–652, copyright c 2007, with permission
from Elsevier
2.9.3 Discussion
We emphasize, however, that the “size” of the simulations does not determine
how “useful” a simulation is by itself. Instead, the most important issue and
measure for a successful simulation is always the physics that can be extracted
from the simulation. This objective should dictate the system size. In many
cases, such as for dislocation–dislocation interaction, system sizes on the order
of micrometers are needed (dislocation interaction is associated with a char-
acteristic length scale of micrometers). This example illustrates that there is
still a need for the development of simulation techniques and more computer
power.
Future development using cheap off-the-shelve technology based on LINUX
clusters to build supercomputers (instead of using very expensive UNIX-based
supercomputers) is promising, as indicated by recent publications [68,69] and
the analysis of data in the TOP500 list.
2 Basic Atomistic Modeling 83
real-time interaction of users with particles in the simulation. For example, the
biophysics group around Klaus Schulten has set up a system where scientists
can interact with the simulation by using a tool to manipulate molecules [145].
The researchers implemented a system called Interactive Molecular Dynamics
(IMD). This system allows manipulation of molecules in molecular dynamics
simulations with real-time force feedback, as well as graphical display. Com-
munication is achieved through an efficient socket connection between the
visualization program (VMD) and a molecular dynamics program (NAMD)
running on single or multiple machines. In this method, a natural force
feedback interface for molecular steering is provided by a haptic device [145].
For the analysis of atomistic simulations focused on mechanical properties,
measures like strain, stress, or potential energy of atoms are important quan-
tities that can be used to analyze atomistic data, in particular with respect
to continuum mechanics theories. However, it is often advantageous to post-
process the data and derive new quantities providing more information of the
defect structure. Here we discuss a few examples for the analysis of crystal
defects in metals that will become particularly important in the third part of
this thesis.
Richard Hamming’s statement the purpose of scientific computing is
insight, not numbers underlines the importance of processing the raw
simulation data appearing as “useless” numbers to gain understanding.
2 Basic Atomistic Modeling 85
Fig. 2.38 Analysis of a dislocation network using the energy filtering method in
nickel with 150,000,000 atoms [13, 14]. Subplot (a) shows the whole simulation cell
with two cracks at the surfaces serving as sources for dislocations, and subplot (b)
shows a zoom into a small subvolume. Partial dislocations appear as wiggly lines,
and sessile defects appear as straight lines with slightly higher potential energy
To visualize crystal defects, the easiest approach is to use the energy method.
This method has frequently been used to “see” into the interior of the solid
(e.g., [138, 146]). In this method, only those atoms with potential energy
greater than or equal to a critical energy φcr above the bulk energy φb are
shown. The energy method has been very effective for displaying disloca-
tions, microcracks, and other imperfections in crystal packing. This method
reduces the number of atoms being displayed by approximately two orders of
magnitude in three dimensions [138].
An example of an analysis of a dislocation network using the energy
method is shown in Fig. 2.38. Figure 2.38a shows the whole simulation cell
with two cracks at the surfaces serving as sources for dislocations, and
Figure 2.38b shows a zoom into a small subvolume of the material revealing
a complex dislocation microstructure. The data are taken from a simula-
tion of work-hardening in nickel that comprises of approximately 150,000,000
atoms [13].
Assuming a crystal defect is identified as a dislocation, it can be studied
in more detail based on a geometric analysis of the lattice close to the dislo-
cation core allowing to determine the Burgers vector and the slip plane. For
that purpose, one can rotate the atomic lattice such that one is looking onto a
86 Atomistic Modeling of Materials Failure
{111}-plane, with the horizontal (x) axis oriented into a 110 direction, and
the vertical (y) axis aligned with a 111 direction. To help visualizing disloca-
tions, stretching the atomic lattice by a factor of 5 to 10 in the 110 direction
is helpful. A systematic rotation of the atomic lattice to investigate all possible
Burgers vectors is then necessary. Instead of analyzing a part of the atomic
lattice containing many dislocations, one can choose a domain of the atomic
lattice which contains only one dislocation. This approach requires a very
detailed understanding of the lattice and dislocations [38, 60]. This method
of analysis is similar to the analysis of TEM images from “real” laboratory
experiments.
Fig. 2.40 The figure shows a close view on the defect structure in a simulation of
work-hardening in nickel analyzed using the centrosymmetry technique [13,14]. The
plot shows the same subvolume as in Figure 2.38b
its nearest neighbors. During deformation, bonds will change direction and/or
length, but they remain equal and opposite within the same pair. This rule
breaks down when a defect is close to an atom under consideration. The
centrosymmetry method is particularly helpful to separate different types of
defects from one another, and to display stacking faults (in contrast, using the
energy method it is difficult to observe stacking faults). The centrosymmetry
parameter for an atom is defined as [36]
3
6
ci = | rk,j + rk,j+6 |2 , (2.71)
j=1 k=1
where rk,j is the kth component of the bond vector (here, k = 1, 2, 3 corre-
sponding to the directions x, y, and z) of atom i with its neighbor atom j,
and rk,j+6 is the same quantity with respect to the opposite neighbor in a
FCC crystal. We summarize the interpretation of ci in Table 2.4 (assuming
that the nearest neighbor distance does not change near a defect). For the
analysis, it is reasonable to display ranges of these parameters. The method
can also be applied at elevated temperature, which is not possible using the
energy method due to the thermal fluctuation of atoms.
Fig. 2.41 Analysis of a dislocation using the slip vector approach. From the result of
the numerical analysis, direct information about the Burgers vector can be obtained.
The slip vector s is drawn at each atom as a small arrow. The Burgers vector b is
drawn at the dislocation (its actual length is exaggerated to make it better visible).
The dislocation line is approximated by discrete, straight dislocation segments. A
line element between “a” and “b” is considered
from an energy analysis, and the line direction of a segment between point
“a” and “b” of the dislocation line is indicated by the vector l. The Burgers
vector b is given by the slip vector s directly. The analysis reveals that the
dislocation has Burgers vector b = 16 [112]. The unit vector of line direction of
the segment is l ≈ [−0.3618 0.8148 − 0.5530]. The length of the line segment
is approximately 9 nearest neighbor distances in the [110] direction. The slip
plane normal is given by the cross product ns = l × b ∼ [111], and the
dislocation thus glides in the (111) plane.
Table 2.5 Distinguishing modeling and simulation, for tasks associated with
classical molecular dynamics
Modeling Simulation
Mathematical model of physical system ×
Numerical solution of problem ×
Choice of model geometry ×
Choice of numerical integrator (e.g., Verlet) ×
Choice of interatomic potential and parameters ×
Force calculation ×
Choice of boundary conditions ×
Implementation of boundary conditions ×
Choice of system size ×
Choice of thermodynamical ensemble ×
Algorithm to specify the thermodynamic ensemble
(e.g., temperature control, pressure control) ×
Fig. 2.43 Simulation method of domain decomposition via the method of virtual
atom types. The atoms in region 2 do not move according to the physical equations
of motion, but are displaced according to a prescribed displacement history. An
initial velocity gradient as shown in the right half of the plot is used to provide
smooth initical conditions
By monitoring the applied force (F ) and the position of the atoms that are
pulled over the simulation time, it is possible to obtain force-vs.-displacement
data that can be used to derive the mechanical properties such as bending stiff-
ness or the Young’s modulus (or other mechanical properties). SMD studies
are typically carried out with a spring constant k = 10 kcal mol−1 Å−2 , albeit
this value can be varied depending on the particular situation considered.
Figure 2.44 depicts a schematic that illustrates how load is applied with
SMD, comparing an AFM experiment with the numerical scheme. One of
the first applications of the SMD technique was to study protein unfolding.
Fig. 2.45 shows steered molecular dynamics simulations of I27 extensibility
under constant force, illustrating the molecular geometry under different force
levels (compare with Fig. 1.16a) as well as the force–extension curve (compare
with Fig. 1.16b).
2 Basic Atomistic Modeling 93
Fig. 2.44 Schematic to illustrate the use of steered molecular dynamics to apply
mechanical load to small-scale structures (subplot (a): AFM experiment; subplot
(b) Steered Molecular Dynamics model)
2.13 Summary
We summarize the main points presented in this section. We discussed analy-
sis techniques, to extract useful information from molecular dynamics results,
including the velocity autocorrelation function, the atomic stress, and the
radial distribution function. These are useful analysis methods since they pro-
vide quantitative information about molecular structure in the simulation, for
example during phase transformations, to study how atoms diffuse, for elastic
(mechanical) properties, and others. We introduced several interatomic poten-
tials that describe the atomic interactions. The basic approach in developing
such models is to condense out electronic degrees of freedom and to model
atoms as point particles.
Properties accessible to molecular dynamics can be classified into these
broad categories [86]:
• Structural - crystal structure, g(r), defects such as vacancies and intersti-
tials, dislocations, grain boundaries, precipitates
• Thermodynamic – equation of state, heat capacities, thermal expansion,
free energies
• Mechanical – elastic constants, cohesive and shear strength, elastic and
plastic deformation, fracture toughness
• Vibrational – phonon dispersion curves, vibrational frequency spectrum,
molecular spectroscopy
• Transport – diffusion, viscous flow, thermal conduction
94 Atomistic Modeling of Materials Failure
Fig. 2.45 Steered molecular dynamics simulations of I27 extensibility under con-
stant force. Subplot (a) shows snapshots of the structure of the I27 module simulated
at a force of 50 pN (I, at 1 ns) and 150 pN (II, at 1 ns). At 50 pN, the hydrogen
bonds between strands A and B are maintained, whereas at 150 pN they are broken.
Subplot (b) displays the corresponding force–extension relationship obtained from
the simulations. The discontinuity observed between 50 and 100 pN corresponds
to an abrupt extension of the module by 4–7 Å caused by the rupture of the AB
hydrogen bonds, and the subsequent extension of the partially freed polypeptide
segment. Reprinted with permission from Macmillan Publishers Ltd., Nature [6] c
1999
• Third law: To every action there is always a reaction with opposed direc-
tion. In other words, the mutual interaction of two bodies are always equal
in magnitude, and directed contrary to each other. For two interacting par-
ticles i and j, this implies that the mangitude of interaction Fij = Fji ,
and for the vectors
Fij = −Fji . (3.5)
In the following sections we will demonstrate how these simple laws can
be used to post and solve mechanical problems. Due to the scope of this
book, this discussion is brief and we refer the reader to further literature for
additional details.
Fig. 3.1 Axial tensile loading of a beam and schematic force–extension response.
Reversible deformation denotes the elastic regime; upon unloading of the sample
the displacement returns to the initial point. Irreversible deformation denotes the
plastic regime; upon unloading (indicated in the graph) the displacement does not
return to the initial point. (It is noted that the specific shape of the force-extension
curve may vary significantly depending on the type of material)
3 Basic Continuum Mechanics 97
∆u ∼ F. (3.6)
This relationship was first noted by Hooke (1635–1703), who stated ‘ut tensio,
sic vis’, which means translated “extension is directly proportional to force.”
This discovery is an important foundation for modern elasticity theory.
The observation that ∆u ∼ F can be used to propose an expression that
links the force to the extension,
F = ks ∆u, (3.7)
σ = Eε, (3.11)
where
Lks
E= (3.12)
A
is defined as Young’s modulus. Equation (3.12) is typically referred to as
Hooke’s law.
98 Atomistic Modeling of Materials Failure
Deformation in the first regime is elastic, that is, the original shape of the
beam is assumed after forces are released. However, for large forces, deforma-
tion is irreversible and permanent deformation remains even after the load
has been released. Permanent deformation is accompanied by dissipative pro-
cesses, that is, work done to the system by external forces (e.g., F in the
example) cannot be recovered.
As we shall explain in the forthcoming sections, Newton’s laws can be used
to solve more complex mechanical problems, as for instance, a typical beam
bending problem as shown schematically in Fig. 3.2.
Fig. 3.2 Example for deformation of a beam due to mechanical loading of a dis-
tributed force qz . The structure responds to the mechanical forces by a change in
shape. Continuum mechanical theory enables us to derive a relationship between
applied forces and displacements, strains, and stresses.
Fig. 3.3 The beam problem as multiscale problem. The goal of solving this problem
is to connect the global scale (scale on the order of L where boundary conditions
are applied, for instance, load P , N , prescribed displacements) with the local scale
(section of the beam, e.g., the stress variation σxx , across the section)
Fig. 3.4 Cross-sectional view of a body. Subplot (a) free body with representative
internal forces. Subplot (b) enlarged view with components of the force vector split
up
Fig. 3.5 The most general state of stress acting on a infinitesimal material element.
All stresses shown in the figure have positive sense
The idea of normalizing the forces by the area they act on can be gener-
alized from the 1D case to 3D case. Here we explicitly consider the different
spatial directions of forces and the plane orientation. The plane cut area and
its orientation is characterized by Ai (by the normal vectors ni ), where the
index i relates to the normal orientation of described surface). The stress
tensor is then defined as
∆Pj
σij = lim . (3.13)
∆A→0 ∆Ai
The components σii are called normal stresses, and the components σij where
i = j are called shear stresses. The stress tensor is a second-order tensor with
generally nine unknowns.
The most general state of stress acting on a infinitesimal material ele-
ment is illustrated in Fig. 3.5. This figure provides a visual illustration of the
definition introduced in (3.13).
The expressions for the stress tensor enable us to write differential equilibrium
conditions based on Newton’s first law. The integration of the differential
equilibrium conditions, by application of boundary conditions to solve for
unknown integration constants, will then enable us to predict the behavior of
an entire structure.
3 Basic Continuum Mechanics 101
Fig. 3.6 Infinitesimal element with stresses and body forces fi acting as volume
forces
The force balance on the infinitesimal element is shown in Fig. 3.6, for a
two-dimensional example (with an out-of-plane thickness of “1”). This sim-
ple schematic is now used to derive the differential equilibrium equations for
continuum mechanics problems.
Fig. 3.7 This schematic explains the condition that σij = σji so that there is no
moment on the infinitesimal element, since it cannot rotate
div σ + f = 0. (3.22)
If inertia terms are included (that is, the time dependence of the solution is
considered), the dynamic equilibrium equations are
∂σij ∂ 2 ui
+ fi = . (3.23)
∂xj ∂t2
In index notation, the equations are written as
Figure 3.8 illustrates the concept of the deformation tensor. The deformation
tensor relates the deformed and undeformed configurations of a material or
structure. We consider an undeformed and deformed beam with two points A
and B, at a distance of dx. The displacement of point A is given by u. Noting
that the term du/dx describes the change of displacement in the x-direction;
assuming a linear relationship we can estimate the additional displacement at
x + dx (at point B) given by
du
u+ dx. (3.25)
dx
The initial distance between points A and B is dx; the deformed distance
between the two points is
du
dx + dx. (3.26)
dx
According to the definition of the strain above (ε = ∆L/L), the strain is
defined as
du
dx + dx − dx du
ε= dx = . (3.27)
dx dx
The displacement of any point can be expressed as
du
u= x, (3.28)
dx
where
du
e= . (3.29)
dx
104 Atomistic Modeling of Materials Failure
∂ui
eij = . (3.30)
∂xj
With the deformation tensor, the displacements can be written as
ui = eij xj . (3.31)
The deformation tensor includes both rigid body rotations and shear
deformations. The deformation tensor can be rewritten so that the parts
describing shear deformation and those describing rotation are separated. The
deformation tensor is given by
1 1
eij = (eij + eji ) + (eij − eji ) , (3.32)
2 2
where
1
(eij + eji )
εij = (3.33)
2
is defined as the strain tensor describing shear deformation, and the second
term
1
(eij − eji ) (3.34)
2
describes the rotational component of deformation that does not contribute
to a shape change of the material. This is also illustrated in Fig. 3.9.
Fig. 3.9 Schematic to illustrate the difference between rotational and deformation
part of the deformation tensor
3 Basic Continuum Mechanics 105
∂S δQ
≥ , (3.36)
∂t T
where δQ/T is the change in entropy supplied to the system in the form of
heat. This second law basically states that the change in entropy is always
greater or equal than the entropy supplied to the system in the form of heat.
The difference between the left and right side in (3.36) corresponds to an
internal heat source due to dissipation. The dissipation rate is therefore given
by
∂D ∂S
=T − δQ ≥ 0. (3.37)
∂t ∂t
Recalling that δQ = dU/dt − δW (the first law of thermodynamics), we can
write
∂D ∂
= δW − (U − T S) . (3.38)
∂t ∂t
The quantity F = U − T S is defined as free energy, or Helmholtz energy,
describing the maximum internal capacity of the system to do work.
During elastic deformation, no dissipation exists by definition (all mechan-
ical work has to be completely recoverable), and therefore
∂D
=0 (3.39)
∂t
and
∂F
δW − =0 (3.40)
∂t
The part δW is known and can be written as δW = ẋFe , and therefore
106 Atomistic Modeling of Materials Failure
∂F ∂x
ẋFe − =0 (3.41)
∂x ∂t
or
∂F
ẋ Fe − = 0. (3.42)
∂x
Since this expression has to be valid for any deformation rate ẋ, the deforma-
tion force Fe can be calculated directly from the change of free energy with
respect to deformation x,
∂F
Fe = . (3.43)
∂x
Assuming that only the internal energy U changes during deformation,
∂U
Fe = . (3.44)
∂x
The change of the force Fe with respect to deformation x yields the spring
constant
∂2U
ks = (3.45)
∂x2
and to satisfy the condition ks > 0 the function U needs to be convex.
By normalization of the force with respect to the area it acts on, (3.44)
can be rewritten as
∂Ψ
σ= , (3.46)
∂ε
where Ψ = F/V is the free energy density. Equation (3.46) is easily obtained
by dividing (3.44) by AL, noting that dx = Ldε.
Further, (3.46) can be generalized to
∂Ψ
σij = . (3.47)
∂εij
∂2Ψ
cijkl = . (3.48)
∂εij dεkl
The elasticity tensor cijkl describes the elastic material properties in the most
general form. This equation is also referred to as the constitutive law. In tensor
notation, (3.49) is written as
σ = cε, (3.50)
with the stress tensor σ = σij ei ⊗ ej , ε = εkl ek ⊗ el (both second-order
tensors) and the fourth-order elasticity tensor, c = cijkl ei ⊗ ej ⊗ ek ⊗ el .
The elasticity tensor has 81 entries. However, due to the symmetries of the
stress and strain tensor and the independence of the order of differentiation
in (3.48), the number of independent entries reduces to 21. These properties
are referred to as minor and major symmetries of cijkl . The minor symmetries
reflect the fact that
cijkl = cijlk = cjikl , (3.51)
which is a direct consequence of σij = σji and εlk = εkl . The minor symmetries
reduce the number of elastic coefficients to 36. The major symmetry is
There are several ways of writing the relationship between stress and strain
for an isotropic solid. A common method is to use Lame coefficients λ and µ,
so that the elasticity tensor coefficients can be written as:
cijkl = µ(δik δjl + δil δjk ) + λδij δkl . (3.53)
From µ and λ, other important quantities can be calculated, including
3 1
K= = (3.54)
3λ + 2µ B
E= 2µ(1 + ν) (3.55)
E − 2µ
ν= (3.56)
2µ
E
µ= (3.57)
2(1 + ν)
3νµ
λ= . (3.58)
1 − 2ν
The parameter K is the bulk modulus, ν, the Poisson’s ratio, and µ the shear
modulus. The Lame parameters are related to cijkl as
1
µ = c1212 = (c1111 − c1122 ) (3.59)
2
λ = c1122 (3.60)
λ + 2µ = c1111 . (3.61)
Expressions for stress and strain relations can be derived from these equations:
σxx = (λ + 2µ) εxx + λεyy + λεzz (3.62)
σyy = λεxx + (λ + 2µ) εyy + λεzz (3.63)
σzz = λεxx + λεyy + (λ + 2µ) εzz (3.64)
σyz = 2µεyz (3.65)
σzx = 2µεzx (3.66)
σxy = 2µεxy . (3.67)
The strain energy density Ψ is given by
1
Ψ= (λ + 2µ)(ε11 + ε22 + ε33 )2 + 2µ(ε223 + ε231 + ε212 − ε11 ε22 − ε11 ε33 − ε22 ε33 ).
2
(3.68)
is, the phenomenon that the material’s stiffness changes under deforma-
tion. Figure 3.10 shows schematic stress–strain plots and the corresponding
stiffness–strain plots, for a linear elastic reference case, hyperelastic stiffen-
ing and hyperelastic softening. Linear elasticity is based on the assumption
that the modulus is independent of strain. The framework of hyperelasticity
enables one to describe the behavior of real materials that do not show a
linear elastic behavior. Many materials show a stiffening effect (e.g., rubber,
polymers, biopolymers) or a softening effect (e.g., metals, ceramics) once the
applied strain approaches large magnitudes.
Here we present one of the simplest systems for which we can develop
a closed form, analytical solution using the tools discussed in the last few
sections. The purpose of this discussion is to illustrate a simple approach in
utilizing the differential equilibrium equations to solve real problems.
We consider a beam with length L and cross-sectional dimensions b and h,
where b/L 1 and h/L 1, as shown schematically in Fig. 3.11. Due to this
particular slender geometry, the beam problem is a quasi-one dimensional
problem. The key feature of a beam problem is the fact that it is a geo-
metrically simplified, special case for which the governing partial differential
equations take a simpler form.
The goal of solving beam problems is to develop solutions that yield expres-
sions for the displacements and stress inside the beam as a function of the
applied load.
We begin with an explanation of the simplifications that can be done to
the governing equations. First, by considering the ratios b/L and h/L one can
show that most stress tensor coefficients are zero. This is a direct consequence
of the particular geometry of the beam, in which L b, h.
The only nonzero components are σxx , σyx , and σzx . Among these, σxx
σyx , σzx so that the only remaining stress tensor coefficient is σxx in a first
approximation. Thus the entire stress tensor field in a beam is approximated
by only one entry, σxx .
FS = Nx ex + Qy ey + Qz ez . (3.72)
dNx
+ qx = 0 (3.76)
dx
dQy
+ qy = 0 (3.77)
dx
dQz
+ qz = 0 (3.78)
dx
and for the moments,
dMx
=0 (3.79)
dx
dMy
− Qz = 0 (3.80)
dx
dMz
+ Qy = 0 (3.81)
dx
112 Atomistic Modeling of Materials Failure
Here we consider a simply supported beam subject to its own dead weight,
q = −ρgAez , as shown in Fig. 3.12.
Fig. 3.12 Solution field for a simply supported beam under dead load ρg, show-
ing the shear force Qy , bending moment My , rotation ωy , and the beam axis
displacement uz
3 Basic Continuum Mechanics 113
Even though the distribution of forces and moments are now known, dis-
placements and stresses remain unspecified. To find these quantities, it is
necessary to introduce material parameters that provide a link between force
and deformation.
The strategy to achieve this is to express the section forces as a function of
section strains and material properties as well as stresses. Then, by knowing
the section force or moment we can easily calculate the corresponding stress
or strain state.
For the beam geometry, we can express the section displacements and
section strains as a function of the variable x:
∂uy,0
ωz,0 = , (3.93)
∂x
∂uz,0
ωy,0 =− . (3.94)
∂x
This leads to the following expressions for the strains
We note that the parameters θz,0 and θy,0 are defined as the change of the
rotations or curvatures:
∂ 2 uy,0
θz,0 = , (3.96)
∂x2
∂ 2 uz,0
θy,0 =− . (3.97)
∂x2
We reiterate that due to the slender geometry of the beam, the shear stresses
are much smaller than the normal stresses, thus σxz = σyx = 0, and therefore
we only have one nonzero stress component, σxx .
The typical solution of a beam problem results in section quantities Nx
(section normal), My , and Mz (both section moments). We recall that these
quantities can be obtained by the reduction formulas:
Nx = σxx dS, (3.98)
S
The terms that appear in the matrix of this equation can be identified as
purely
# geometric properties of the cross-section of the beam.
# The quantity
#
S
dS = A is the zero-order area moment, the quantities
# S ydS and S zdS
are the first-order area moments, and the quantities Iij = S xi xj dS is referred
to as the second-order area moment.
The normal force can be expressed as
Nx = Eεxx,0 A. (3.107)
z 1
σxx (x = 0; y, z) = ρgAL. (3.115)
Izz 2
The entire distribution as a function of all variables x, y, and z is given by
z 1
σxx (x; y, z) = ρgA(L − x)2 . (3.116)
Izz 2
Note that due to the particular boundary condition, there is no variation of
stresses in the y-direction. Figure 3.3 illustrates schematically the distribution
of σxx inside the section.
d2 Mx
= 0, (3.117)
dx2
2
d My
+ qz = 0, (3.118)
dx2
d2 Mz
− qy = 0. (3.119)
dx2
Now we take advantage of the result from (3.112), considering (3.97). There-
fore,
3 Basic Continuum Mechanics 117
∂ 4 uz,0
−EIzz + qz = 0, (3.120)
∂x4
∂ 4 uy,0
EIyy − qy = 0. (3.121)
∂x4
These are the fourth-order differential equations that govern beam elasticity.
These are the equations most frequently used to solve beam problems.
Integrating these equations four times, while considering the boundary
conditions, leads to solutions for the displacement field in the beam. We note
that for a two-dimensional beam geometry (that is displacements only in the
z direction) the shear forces are given by
∂ 3 uz,0
Qz (x) = −EIzz , (3.122)
∂x3
the moment is given by
∂ 2 uz,0
My (x) = −EIzz , (3.123)
∂x2
the rotations are given by
∂uz,0
ωy (x) = − , (3.124)
∂x
and the displacements are uz,0 .
Once the solutions for the deformation of a beam are known, the stored
elastic strain energy can be calculated by integrating over the strain and
curvature fields:
1
Ψ= EAε2xx,0 + EIzz θy,0
2 2
+ EIyy θz,0 dx. (3.125)
2 x
and
1
My (x) = ρgA(x − L)2 . (3.127)
2
This solution is identical to the one presented in (3.89) and (3.89) and is shown
in Fig. 3.12b and c. Further integration of the equations leads to expressions
for the rotations and the displacements:
118 Atomistic Modeling of Materials Failure
ρgA 3
ωy (x) = x (3.128)
6EIzz
and
ρgA 4
uz (x) = − x . (3.129)
24EIzz
These solutions are sketched in Fig. 3.12d and e. Note that whereas forces
and moments have a maximum at x = 0, the rotations and displacements are
largest at x = L. The largest displacement at the tip of the beam is given by
ρgA 4
uz (L) = − L . (3.130)
24EIzz
This maximum tip displacement is proportional to the load ρgA,
the moment
the rotation
P x2
ωy (x) = − − xL , (3.135)
EIzz 2
Fig. 3.14 Solution field for a simply supported beam under a point load P applied
at the end of the beam, showing the shear force Qy , bending moment My , rotation
ωy , and the beam axis displacement uz
This chapter is dedicated to provide a link between the atomistic models and
continuum theoretical concepts of stress and strain. We review the basic ther-
modynamical concepts that enable a rigorous link between atomistic systems
and corresponding continuum theoretical concepts. The discussion includes a
review of entropic and energetic contributions to elasticity, free energy mod-
els, entropic effects in rubber-like materials as well as the Cauchy–Born rule.
We discuss elastic properties of a variety of crystal structures for different
interatomic potentials.
F = U − T S, (4.1)
This finding is quite significant: If one can calculate the free energy density
of a atomistic system for various deformation states, then one can estimate the
stress as well as Young’s modulus! Provided that an accurate model for the
atomic interactions is used, this provides a theoretical strategy to predict
the elastic properties of materials, addressing the point discussed in Sect. 3.7.
Equation (4.3) can be written as a tensor equation, considering all 21
independent elastic coefficients cijkl and all independent entries in the stress
and strain tensor. In more general terms, the stress tensor σij and the elasticity
tensor cijkl can be written as:
∂2Ψ
σij = (4.5)
∂εij
and
∂2Ψ
cijkl = . (4.6)
∂εij ∂εkl
These equations inform us that the change of the energy of a system
controls the elastic properties, and thus the elastic properties are strongly
influenced by the actual contributions to the free energy F . For instance, the
free energy density of crystalline materials (e.g., metals, semiconductors) at
moderate temperatures is primarily controlled by energetic changes of the
internal energy. In natural and biological materials, the free energy density is
dominated by entropic contributions. Therefore in many crystalline materials,
the entropic term can be neglected, so that in (4.3) and (4.5) can be directly
substituted by U/V , the internal energy, and therefore
U
Ψ≈ . (4.7)
V
However, in biopolymers, entropic contributions can dominate the elasticity,
in particular for small deformation, and therefore Ψ can be substituted by
−T S/V , so that
TS
Ψ ≈− . (4.8)
V
The dominance of entropic behavior is a well-known and well-studied phe-
nomenon in many polymers. The contributions to the entropic term to
elasticity can be described in several ways, including classical descriptions
such as the WLC or the freely jointed Gaussian chain model. These models
will be discussed shortly.
were ∆xkl
i = xli − xki
and ∆xj = xlj − xkj . The quantity N refers to the number
of nearest neighbors considered. The left Cauchy–Green strain tensor is given
by
1 ∆xkl
N kl
l N l i ∆xj
bij = qij = , (4.10)
λ λ r02
k=1
where λ is a prefactor depending on the lattice considered. For a two-
dimensional triangular lattice with nearest neighbor interaction, λ = 3; λ = 2
for a square lattice with nearest neighbor interaction; and λ = 4/3 for a
face-centered cubic lattice.
This definition provides an expression for a measure of deformation defined
using continuum mechanics and in terms of atomic positions.
The Eulerian
strain tensor of atom l is obtained from (4.10), elij = 12 δij − blij . One can
√
calculate the engineering strain ε = b − 1. Unlike the virial stress, the
atomic strain is valid instantaneously in space and time, since it is a purely
geometric definition. However, the expression is only strictly applicable away
from surfaces and interfaces.
continuum strain field. Thereby it is assumed that the continuum fields can
immediately be mapped to the atomic scale. This is only possible if the strain
field is slowly varying at the atomistic scale. In other words, at the microscopic
atomic scale, it is assumed that no gradients in the strain field occur.
In this approach, the atomic displacements are calculated directly from the
continuum strain fields by imposing the appropriate displacements directly at
the atomistic lattice. This provides the positions of all atoms in the atomic
system as a function of the deformation field, assuming that a continuous
interpolation function can be used.
The knowledge of the atomic positions then enables one to calculate the
strain energy density (SED) Ψ of an atomic system as a function of the con-
tinuum strain field. The first derivatives of the SED with respect to the strain
yield the stress, the second derivatives of the SED with respect to the strain
give the modulus, as given in (4.5) and (4.6).
In general form, the SED can be expressed as follows:
1
Ψ (εij ) = U (r)dΩ, (4.11)
Ω Ω
where Ω is the volume of the atomic system and U (r) is the potential energy
as a function of all atomic coordinates, denoted by r. In order to obtain the
desired relationship of the SED as a function of the strain tensor, the atomic
coordinates must be expressed as a function of the strain tensor by using a
mapping function
r = DΩ (εij ). (4.12)
The mapping function DΩ maps the strain tensor field to atomic positions,
thereby providing a direct relationship of r(ε). By integrating over the volume
of the atomic system considered, the total energy is determined directly as a
function of the strain tensor field:
1
Ψ (εij ) = U (DΩ (εij ))dΩ. (4.13)
Ω Ω
1
Ψ (εij ) = φ(rα (εij )). (4.14)
Ω α
φ(r) = a0 + a1 (r − r0 ) + a2 (r − r0 )2 + · · · . (4.18)
Since the force vs. bond stretch is zero at the equilibrium bond length, a1 =
φ (r0 ) = 0. The parameter a2 = φ (r0 ) corresponds to the force–bond distance
slope around r0 and is now denoted as k. The parameter a0 is a constant, which
can be set to zero since we are only interested in the derivatives of Ψ (ε).
Thus, we can approximate
k 2
φ(r) ≈ (r − r0 ) . (4.19)
2
We can then rewrite (4.17) as
1 k 2 r0
Ψ (ε) = ((1 + ε − 1)r0 ) = kε. (4.20)
r0 ΩD 2 2D
The first derivative of Ψ (ε) yields the stress for a given strain ε,
∂Ψ (ε) r0
= σ(ε) = kε. (4.21)
∂ε D
The second derivative yields the modulus,
∂ 2 Ψ 2 (ε) r0
2
= E = k. (4.22)
∂ε D
Note that since we only considered a second-order expansion of the interatomic
potential, the modulus is independent of strain, as assumed in linear elasticity.
Equations (4.21) and (4.22) provide relationships between stress and strain
and define Young’s modulus for this system.
We can use this expression to calculate other material properties, such
as the wave speed. For the one-dimensional system, the (only) wave speed is
given by
E
c= , (4.23)
ρ
where ρ is the atomic density, given by
m
ρ= , (4.24)
r0
with m denoting the atomic mass.
These relationships have been established by only considering (1) the
atomic microstructure, (2) the interatomic potential, and (3) by assuming
that the Cauchy–Born rule can be applied (that is, that one can map the con-
tinuum displacement field to the atomic lattice). No other phenomenological
assumptions have been made.
128 Atomistic Modeling of Materials Failure
The Cauchy–Born rule is not applicable to all atomic structures. In the case
of molecular crystals, or more complex lattice structures (e.g., in graphene,
carbon nanotubes), condition (3) may not be satisfied. This is because in
such systems, atomic displacements within the unit cell do not necessarily
correspond to the continuum displacement field. Other systems where this is
relevant are amorphous materials where the Cauchy–Born rule can only be
applied in a statistical sense.
Fig. 4.2 Subplot (a) rectangular cell in a uniformly deformed triangular lattice;
subplot (b) the geometrical parameters used to calculate the continuum properties
of the lattice
Here we review a case study that illustrates the methods that can be used to
assess the elastic properties of a simple two-dimensional triangular lattice (see
[27] for further details). We note that here we use large-deformation elasticity
theory, a concept that was briefly rigorously introduced in Sect. 3.5, discussing
the stress–strain relations in softening and stiffening material behavior (see
Fig. 3.10). Large-deformation elasticity theory allows one to describe how the
elastic response changes under increasing deformation, as it is for instance the
case during stiffening of rubber, or in softening of metals. We refer the reader
to the literature for additional details in regards to this approach, in particular
to find a more rigorous introduction into this topic (see for instance [163]).
Classical hyperelastic continuum theory is based on the existence of a
strain energy function [163]. Using the Cauchy–Born rule, applied to the tri-
angular lattice considered in this section (see Fig. 4.2 for the geometry of
the unit cell), the strain energy density per unit undeformed area is given
by [27, 160, 161]
4 Atomistic Elasticity: Linking Atoms and Continuum 129
2
Ψ = √ (φ(l1 ) + φ(l2 ) + φ(l3 )) , (4.25)
3
where the values of li are determined from geometric relations of the trian-
gular lattice as shown in Fig. 4.2. The function φ(r) refers to the interatomic
potential. The unknowns Eij are the Green–Lagrangian strain components
[163, 164], and these can be determined to be
Exx = Λ21 − 1 /2, Eyy = Λ22 − 1 /2, Exy = Eyx = Λ1 Λ2 cos (Θ/2) .
(4.26)
Here, Θ is the shear angle, while Λi describe the elongation of the sides of a
lattice unit cell as indicated in Fig. 4.2. From geometric relations, it is found
that
l1 = 1 + 2Exx , (4.27)
' √
l2 = 1 + 1/2Exx + 3/2Eyy − 3/2(Exy + Eyx ), (4.28)
and ' √
l3 = 1 + 1/2Exx + 3/2Eyy + 3/2(Exy + Eyx ). (4.29)
The symmetric second Piola–Kirchhoff stress tensor is given by
∂Ψ
Sij = . (4.30)
∂Eij
The “slope” of the S−E relationship is often called the material tangent
modulus
∂2Ψ
Cijkl = . (4.31)
∂Eij ∂Ekl
For infinitesimal strains, the Green–Lagrangian strain reduces to the stress
tensor of linear elasticity Eij → εij . The same argument can be used for the
stresses, and the second symmetric Piola–Kirchoff stress tensor reduces to the
linear elasticity stress tensor Sij → σij , as well as Cijkl → cijkl . This scheme is
universally applicable, as long as the interatomic potential and thus the strain
energy function Ψ is known (it can for instance be applied to pair potentials
or the embedded atom method (EAM) for metals).
Poisson’s ratio ν is defined as the ratio of transverse strain to longitudinal
strain in the direction of stretching force. Poisson’s ratio can be found by
choosing a Green–Lagrangian strain Exx and finding a value for Eyy such
that Syy assumes zero (and vice versa). We can define
Eyy
ν=− (4.32)
Exx
as Poisson’s ratio valid also for large strain.
To obtain a linear elasticity formulation with first-order stress–strain law,
the strain energy given in (4.25) is expanded up to second-order terms. After
some lengthy calculations it can be shown that
130 Atomistic Modeling of Materials Failure
√
3
Ψ= φ (r0 ) 3ε2xx + 2εxx εyy + ε2yy + (εxy + εyx )2 , (4.33)
8
where r0 is the nearest neighbor distance. Using
∂Ψ
σij = , (4.34)
∂εij
one can derive expressions for stress–strain relations, like for instance
√
∂Ψ 3
σyy = = φ (r0 ) (εxx + 3εyy ) . (4.35)
∂εyy 4
As in the case of the one-dimensional example, this equation provides a rig-
orous link between the elastic properties and the interatomic potential. In
the following sections, we review the application of this approach to different
interatomic potentials.
Lennard-Jones Potential
We begin with the elastic properties of a solid in which atoms interact accord-
ing to a LJ potential as defined in (2.30) (we choose σ = 0 = 1). LJ type
potentials have frequently been used in simulating fracture using molecu-
lar dynamics [146]. Solids defined by this potential behave as a very brittle
material in a two-dimensional triangular lattice. Figure 4.3 shows numeri-
cal estimates of the elastic properties of a LJ solid. The systems are loaded
uniaxially in the two symmetry directions of the triangular lattice.
The plot of the LJ system shows that the y-direction requires a higher
breaking strain than in the x-direction (about 18% vs. 12%). The tangent
Young’s modulus drops significantly from around 66 for small strain until it
reaches zero when the solid fails [163]. Poisson ratio remains around 1/3, but
increases slightly when loaded in the x-direction and decreases slightly when
loaded in the orthogonal direction.
The objective is to obtain a solid with the property that its tangent moduli
stiffen with strain, in contrast to the LJ potential described above [163]. In
addition, the small-strain elastic properties should be the same as in the LJ
potential. The nonlinear tethered LJ potential is obtained by modifying the
well-known LJ 12:6 potential: The potential is mirrored at r = r0 = 21/6 ≈
1.12246, leading to a strong stiffening effect instead of the normal softening
associated with atomic separation.
⎧ 6
⎪
⎪
12
⎨40 1
rij − 1
rij if rij < r0 ,
φ(rij ) = 12 6 (4.36)
⎪
⎪
⎩40 1
(2r0 −rij ) − 1
(2r0 −rij ) if rij ≥ r0 ,
4 Atomistic Elasticity: Linking Atoms and Continuum 131
Fig. 4.3 Elastic properties of the Lennard-Jones solid (continuous line) and elastic
properties associated with the harmonic potential (dashed line). The dash-dotted
lines in the upper plots show Poisson’s ratio. The lower plots show the tangent
modulus for this case. This plot is an actual material law representing the schematic
shown in Fig. 3.10
Fig. 4.4 Elastic properties associated with the tethered LJ potential, and in compar-
ison, elastic properties associated with the harmonic potential (dashed line). Unlike
in the softening case, where Young’s modulus softens with strain (Fig. 4.3), here
Young’s modulus stiffens with strain
Harmonic Potential
cR ≈ βcs . (4.40)
Table 4.1 Elastic properties and wave speeds associated with the harmonic potential
(see (2.34)) in a two-dimensional solid for different choices of the spring constant k
Results for elastic properties and wave speeds are summarized in Table 4.1
for two different choices of the spring constant.
To check if the predictions by (6.67) hold even for large strains, we inves-
tigate the elastic properties numerically. The numerically estimated elastic
properties for uniaxial tension are shown in Fig. 4.5 for the two different crystal
orientations in a triangular lattice and k ≈ 28.57. We find reasonable agree-
ment, which could be verified by comparing the values reported in Table 4.1
with the results shown in Fig. 4.5.
134 Atomistic Modeling of Materials Failure
Fig. 4.5 Elastic properties of the triangular lattice with harmonic interactions, stress
vs. strain (left) and tangent moduli Ex and Ey (right). The stress state is uniaxial
tension, that is the stress in the direction orthogonal to the loading is relaxed and
zero
Young’s moduli agree well with the continuum mechanics prediction for
small strains. However, we observe a slight stiffening effect for large strains,
that is, E is increasing with strain. As predicted, the lattice is found to
be isotropic for small deformations, but the results show there exists an
anisotropy effect for large deformations. The values of Poisson’s ratio match
the linear approximation for small strains, but deviate slightly for large strains.
This suggests that even if harmonic potentials are introduced between atoms,
the triangular lattice structure yields a slightly nonlinear stress–strain law.
We note that the values for Young’s modulus associated with the LJ poten-
tial at small strains are in consistency with the results using the harmonic
potential with k ≈ 57.14 (see Fig. 4.3). The small-strain elastic properties
also agree in the case of the tethered LJ potential (see Fig. 4.4). The compari-
son with the harmonic potential nicely illustrates the softening and stiffening
effect of the LJ and tethered LJ potential. Since the small-strain elastic prop-
erties agree with the harmonic potential in both cases, the small-strain wave
speed is also identical and thus given by (4.38)–(4.40). Note that there is no
unique definition of the wave speed for large strains in the nonlinear potentials.
Fig. 4.6 Illustration of the shape of the harmonic potential, comparing the one
defined in (2.34) (panel (a)) and the one defined in (4.43) with the bond snapping
parameter rbreak (panel (b))
Figure 4.6 shows a graph that displays the shape of the harmonic potential,
comparing the potential defined in (2.34) and the one defined in (4.43) with
the bond snapping parameter rbreak .
Fig. 4.7 The figure shows the stretching of the triangular lattice in two different
directions
The elastic properties for rij < rbreak are identical to those discussed in
Sect. 4.4.3 and shown in Fig. 4.5, but for large strains close to the failure of
the solid there are strong differences. We focus on the differences in elastic
properties due to stretching in the x-direction vs. the y-direction. Figure 4.7
136 Atomistic Modeling of Materials Failure
shows the crystal orientation for stretching of the triangular lattice in the two
different directions. We define two different bond types r1 and r2 : The bonds
denoted by r1 have a component only in the x-direction, whereas bonds r2
have a component in the x as well as in the y-direction.
Fig. 4.8 The figure plots the elastic properties under uniaxial loading with Poisson
relaxation for the harmonic potential. In the plot, stress vs. strain, Poisson’s ratio
as well as the number of nearest neighbors are shown. The lower two subplots show
Young’s modulus
Fig. 4.9 The figure plots the elastic properties under uniaxial loading without
Poisson relaxation for the harmonic potential. In the plot, stress vs. strain, as well
as the number of nearest neighbors are shown. The lower two subplots show Young’s
modulus
the x-direction and smaller in the y-direction (see Fig. 4.9). Cracks are there-
fore expected to propagate stable along the direction of lower fracture surface
energy (crack extension along x-direction). We will verify this prediction with
molecular dynamics results in a chapter 6.
Biharmonic Potential
Thus far, we have studied a LJ potential that yields elastic properties that
soften strongly with strain, a tethered LJ potential that yields a solid strongly
stiffening with strain and a harmonic potential. To be able to smoothly inter-
polate between harmonic potentials and strongly nonlinear potentials, we
adopt a biharmonic, interatomic potential composed of two spring constants
k0 and k1 similar to that discussed in Sect. 6.4.5 for the one-dimensional case
(all quantities given are in dimensionless units).
We consider two “model materials,” one with elastic stiffening and the
other with elastic softening behavior. In the elastic stiffening system, the
spring constant k0 is associated with small perturbations from the equilib-
rium distance r0 , and the second spring constant k1 is associated with large
4 Atomistic Elasticity: Linking Atoms and Continuum 139
bond stretching for atomic separation r > ron . The role of √ k0 and k1 is reversed
in the elastic softening system (k0 = 2k1 and k1 = 36/ 3 2).
Purely harmonic systems are obtained if ron is chosen to be larger than
rbreak . Poisson’s ratio ν is found to be approximately independent of strain
and around ν ≈ 0.33 for all potentials. In the stiffening system, the small
deformation (up to about 0.5% of strain) Young’s modulus is E ≈ 33 with
shear modulus µ ≈ 12.4, and the large deformations tangent Young’s modulus
is E ≈ 66 with shear modulus µ ≈ 28.8. The values are reversed for the
softening system where the small deformation Young’s modulus is E ≈ 66,
and the large deformation tangent Young’s modulus is E ≈ 33.
The biharmonic potential is defined as
a0 + 12 k0 (rij − r0 )2 if rij < ron ,
φij (rij ) = (4.44)
a1 + 12 k1 (rij − r1 )2 if rij ≥ ron ,
where ron is the critical atomic separation for onset of the hyperelastic effect,
and
1 1
a1 = a0 + k0 (ron − r0 )2 − k1 (ron − r1 )2 (4.45)
2 2
as well as
1
r1 = (ron + r0 ) (4.46)
2
are found by continuity conditions of the potential at r = ron (note that the
expressions (4.45) and (4.46), derived from energy continuity, are only valid
for the ratio 2 of the large to small spring constant; similar expressions can be
developed for other ratios). The values of k0 and k1 refer to the small-strain
and large-strain spring constants. Figure 4.10 shows a graphical explanation
of the parameters used in the biharmonic potential defined in (4.44).
Fig. 4.10 Illustration of the parameters used in the biharmonic potential defined
in (4.44). The plot defines r, k0 , k1 , ron , rbreak , as well as the “atomic” strain
140 Atomistic Modeling of Materials Failure
The elastic properties associated with the biharmonic potential are shown
in Fig. 4.11. The wave speeds for small and large strains are given by the
values of the corresponding harmonic potentials. Therefore, the wave speeds
strains are given by cl,1 = κcl,0 , cs,1 = κcs,0 and cR,1 =
associated with large
κcR,0 where κ = k1 /k0 .
Similarly as described in the previous section, a critical bond breaking
distance rbreak can be introduced allowing for snapping bonds.
Fig. 4.11 Elastic properties of the triangular lattice with biharmonic interactions,
stress vs. strain in the x-direction (a) and in the y-direction (b). The stress state
is uniaxial tension, that is the stress in the direction orthogonal to the loading is
relaxed and zero
depicted in Fig. 4.12a for cracks propagating along the direction with highest
fracture
√ surface energy. In this case, four bonds break while the crack proceeds
d = 3r0 . Figure 4.12b shows the bond breaking process for crack orientation
along the direction of lowest fracture surface energy. In this direction, two
bonds break while the crack advances d = r0 .
Fig. 4.12 Bond breaking process along the fracture plane and calculation of fracture
surface energy for (a) direction of high fracture surface energy and (b) direction of
low fracture surface energy
For the case considered in the simulations, the fracture surface energy
is determined assuming that bonds between nearest neighbors snap during
crack propagation. Unlike the wave velocity, the fracture surface energy is
well defined for both linear and nonlinear cases. We summarize the results for
different potentials described in the earlier sections of this chapter.
The fracture surface energy for the harmonic bond snapping model for
crack propagation along the direction of high fracture surface energy (as shown
in Fig. 4.12) is given by
k(rbreak − r0 )2 E(rbreak − r0 )2
γsbs,h = √ = , (4.48)
3r0 2r0
which yields γbs = 0.0332 for rbreak = 1.17. For the direction of low fracture
surface energy,
k(rbreak − r0 )2
γsbs,l = , (4.49)
2r0
which yields γbs = 0.0288 for rbreak = 1.17 and is about 15% smaller than in
the other direction.
The surface energy for the biharmonic bond snapping model along the
direction of high fracture surface energy is given by
, -
2a1 + k1 (r1 − rbreak )2 E0 (ron − r0 )2 − E1 (ron − r1 )2 − (r1 − rbreak − r0 )2
γsbi,h = √ = .
3r0 2r0
(4.50)
For the purely harmonic case, the fracture surface energy reduces to the
expression given by (4.48). In the direction of lower fracture surface energy
142 Atomistic Modeling of Materials Failure
, -
2a1 + k1 (r1 − rbreak )2 E0 (ron − r0 )2 − E1 (ron − r1 )2 − (r1 − rbreak − r0 )2
γsbi,l = = √ .
2r0 4/ 3r0
(4.51)
Harmonic Potential
The fact that c1122 = c1212 shows that the Cauchy relation holds. For a cubical
crystal orientation (that is, x = [100] and y = [010] and z = [001]), the nonzero
factors bijkl are given by
The second derivative of the potential φ = k, where k is the spring constant
associated with the harmonic potential. The shear modulus can be expressed
in terms of the spring constant and the nearest neighbor distance as
r02
µ= k. (4.54)
2
This analysis is based on the material discussed in [109].
For k0 = 28.57 this leads to numerical values c1111 ≈ 36 and µ = c1122 =
c1212 ≈ 18. Note that λ = 2µ, and therefore Young’s modulus is
4 Atomistic Elasticity: Linking Atoms and Continuum 143
µ (3λ + 2µ) 8
E= = µ ≈ 48 (4.55)
λ+µ 3
and the shear modulus is µ = c1122 = c1212 ≈ 18. Poisson’s ratio is determined
to be
λ
ν= = 1/3. (4.56)
2(λ + µ)
Fig. 4.13 Elastic properties associated with the harmonic potential, [100] crystal
orientation, with Poisson relaxation. Poisson ratio is ν ≈ 0.33 and is approximately
independent of the applied strain. The plot shows the elastic properties as a function
of strain
Figure 4.16b plots the results for uniaxial tension with Poisson relaxation
in the [111] direction. Young’s modulus is approximately E ≈ 100. Poisson’s
ratio is identical in the y- and z-direction and is found to be ν ≈ 0.2. As for the
loading in [110], this result can also be obtained from continuum mechanics
theories. The elastic properties in the [110] and [111] direction are summarized
in Table 4.4.
Loading direction k E νy νz
√
[110] 36 2 ≈ 28.57
3
72 0 0.5
√
[111] 36 3 2 ≈ 28.57 100 0.2 0.2
Table 4.4 Elastic properties associated with the harmonic potential (2.34) in a
three-dimensional solid for different choices of the spring constant k and [110] and
[111] crystal orientation
4 Atomistic Elasticity: Linking Atoms and Continuum 145
Fig. 4.14 Elastic properties associated with the harmonic potential, [100] crystal
orientation, without Poisson relaxation. The plot shows the elastic properties as a
function of strain
Fig. 4.15 Elastic properties associated with the harmonic potential, [100] crystal
orientation, triaxial loading. The plot shows the elastic properties as a function of
strain
√
2a0
σ= √ . (4.60)
262
These relationships can be used, for instance, to determine the potential
parameters for a specific metal. For copper with a bulk modulus K = 140 GPa,
one can determine ε0 = 0.161 eV and σ = 2.277 Å. These parameters can be
obtained by first determining σ so that the equilibrium lattice constant of
copper is reproduced using (4.60), and then by using (4.59) to determine the
other unknown to fit the bulk modulus. These potential parameters for LJ
copper are similar to those reported in [9].
Fig. 4.16 Elastic properties associated with the harmonic potential, (a) [110] and
(b) [111] crystal orientation, uniaxial loading with Poisson relaxation. The plot
shows the elastic properties as a function of strain
For other potentials, we do not give the analytical expression but sum-
marize the results in Table 4.6. The results for harmonic and biharmonic
potentials are also included.
148 Atomistic Modeling of Materials Failure
Fig. 4.17 Elastic properties associated with (a) LJ potential, and (b) an EAM
potential for nickel [15], uniaxial loading in [100], [110] and [111] with Poisson
relaxation
2γ = Nb ρA ∆φ, (4.61)
where ρA = 2/a20 ≈ 0.794 is the density of surface atoms along the fracture
plane and ∆φ denotes the potential energy per bond. The factor Nb = 4 since
each atom has 4 bonds across the [100] plane (thus (010) crack faces). The
potential energy per bond is given by
1
∆φ = k0 (rbreak − r0 )2 (4.62)
2
and ∆φ ≈ 2.26 × 10−3 for rbreak = 1.17 and k0 ≈ 57.32.
Therefore, the fracture surface energy is 2γ ≈ 0.21. As in the two-
dimensional case, note that γ ∼ k0 and therefore γ ∼ E.
4 Atomistic Elasticity: Linking Atoms and Continuum 149
Summary
The results reviewed in this chapter provide analytical expressions for the
elastic properties of three-dimensional solids with harmonic interatomic poten-
tials. The analytical predictions were verified by numerical calculations of the
elastic properties. We also report results of elastic properties of FCC solids
with LJ interatomic potential and EAM potentials. An interesting observa-
tion was that when pulling in the [110] direction, the solid fails at a very low
strain (and at very low stress) compared to other pulling directions ([100] and
[111]). This phenomenon is likely due to the strong softening of the bonds in
the LJ and EAM potential. In contrast, such phenomenon does not appear
in the harmonic potential since the bonds do not weaken with stretching (see
Figs. 4.13 and Fig. 4.16). In fact, Young’s modulus in the [110] direction is in
between the values of the [100] and [111] direction. Similar observations have
been made in earlier studies for LJ pair potentials [156]. The results show
that this also applies to EAM potentials. Therefore, it is expected that this
phenomenon should occur in metals.
The observation of this “weak” crystal orientation could potentially have
impact on the design of nanowires or electronic interconnects in integrated
circuits.
descriptions such as the worm-like chain model or the freely jointed Gaussian
chain model (for a good introduction into these concepts, please see [167,
168]).
Such descriptions are similar to constitutive models in continuum elastic-
ity, and require input parameters that are typically determined empirically.
In contrast to these models, molecular dynamics modeling can provide a first
principles-based description of entropic elasticity – without any additional
fitting parameters beyond the atomic interactions.
Fig. 4.18 This plot depicts a series of snapshots of a single molecule with increasing
length L, at constant temperature. The longer the molecule, the more wiggly the
geometrical shape
EI
ξp = . (4.63)
kB T
When the length of molecules, denoted by L is beyond the persistence
length, that is, L ξp , thermal energy can bend the molecule, and entropic
elasticity typically plays a role. On the other hand, when L ξp , entropic
effects play a minor role, and energetic elasticity governs.
Fig. 4.19 Entropy controlled molecular elasticity. Subplot (a) Coiled, entangled
state of a molecule with contour length much larger than the persistence length.
The end-to-end distance is measured by the variable x. Subplot (b) Response of
the molecule to mechanical loading. As the applied force is increased, the end-to-
end distance x increases until the molecule is fully entangled. Clearly, the continued
disentanglement leads to a reduction of entropy in the system, which induces a force
that can be measured as an elastic spring. Once the molecule is fully extended, the
change in entropy due to increased force approaches zero, and the elastic response is
controlled by changes in the internal potential energy of the system, corresponding
to the energetic elasticity
stiffness. If these properties are known from atomistic calculations, the WLC
model provides a quantitative estimate of entropic elasticity.
Figure 4.20 depicts the entropic response (F < 14 pN) of a single tropocol-
lagen molecule, illustrating that the WLC model is a good approximation for
experimental and numerical results of stretching biopolymers [17]. The plot
also shows a comparison with experimental results [16].
Fig. 4.20 This plot depicts the entropic response (F < 14 pN) of a single tropocolla-
gen molecule, obtained by direct molecular dynamics simulation using a multi-scale
model [17]. This plot also depicts experimental results [16] obtained for TC molecules
with similar contour lengths, as well as the prediction of the WLC model with persis-
tence length of approximately 16 nm [17]. The force-extension curve shows a strong
hyperelastic stiffening effect (see also Fig. 3.10)
Fig. 4.21 The concepts of entropic elasticity of single molecules can be immediately
applied to understand two-dimensional and three-dimensional networks of molecules
in a polymer. This figure demonstrates how a change in state of deformation poses
constraints on the end-to-end distances of molecules, influencing the entropy of the
system. Such considerations enable to link the properties of single molecules (their
entropy) with the overall macroscopic elastic behavior of the material
S = c − kB b2 r2 , (4.65)
where
3
b2 = . (4.66)
2nl2
The change in entropy due to deformation is given by
∆S = −kB b2 (λ21 − 1)x2 + (λ22 − 1)y 2 + (λ23 − 1)z 2 . (4.67)
Nb
and finally at
−kB Nb 2
∆S = (λ1 − 1) + (λ22 − 1) + (λ23 − 1) . (4.72)
2
The change in free energy is then given by
1
A = −T ∆S = kB N T (λ21 − 1) + (λ22 − 1) + (λ23 − 1) . (4.73)
2
The free energy density is given by
A
Ψ= . (4.74)
V
This expression of the free energy density function can be used to derive
expression for Young’s modulus, leading to
E = 3N ∗ kB T, (4.75)
4.6 Discussion
In this chapter, we have summarized the main concepts of how to link
atomistic models with continuum concepts, via the utilization of a language
common to both approaches: Thermodynamics. Several case studies have
been provided, including model materials for two-dimensional and three-
dimensional crystal lattices. It is noted that these cases will be revisited later
when the propagation of fractures in these lattices is discussed.
We will provide a few concluding remarks regarding the use of units in
molecular dynamics simulations. To interpret the resulting quantities from
MD simulation properly, it is important to consider the particular reference
4 Atomistic Elasticity: Linking Atoms and Continuum 155
units used in the code. For instance, many EAM codes use the following ref-
erence units: Energies are expressed in E ∗ = 1 eV, distances are expressed in
L∗ = 1 Å, and the mass is expressed in M ∗ = 1 amu. In order to concert these
units to SI units, appropriate unit versions must be carried out. For example,
resulting pressure or stress tensor components will be in the reference stress
unit σ ∗ = 160.2 GPa and temperature results will be expressed in 11, 600 K
(energy divided by the Boltzmann constant). The reference time unit corre-
sponds to T ∗ = 1.0181 × 10−14 s. Similar conversions can be carried out for
other physical quantities such as the density or for other reference units (for
example, many chemistry codes use kcal mol−1 as the energy reference unit).
Table 4.7 displays a summary of conversions of frequently used units to SI
units. The table also shows frequently used conversions between units often
found in molecular dynamics codes.
5.1 Introduction
The motivation for multiscale simulation methods is that it is not always
necessary to calculate the full atomistic information in the whole simulation
domain. Based on this insight, several researchers have articulated the need for
multiscale methods [20,171–175] by combining atomistic simulations with con-
tinuum mechanics methods (for instance, finite element methods). A variety
of different methods have to be developed to achieve this.
An important motivation for this is to save computational time and by
doing that, to extend the lengthscale or timescale accessible to the simula-
tions. It is common to distinguish between hierarchical multiscale methods and
on-the-fly concurrent multiscale methods. In hierarchical multiscale methods,
a set of different computational tools are used sequentially. First, the most
accurate method (e.g., quantum mechanics) is used to determine parameters
for the next computational approach (e.g., via force field fitting to gener-
ate interatomic potentials). Molecular dynamics simulations with interatomic
potentials are then used to determine constitutive equations, or criteria for
plasticity, which are utilized as parameters in finite element approaches. These
approaches can be carried out for a variety of computational methods. In on-
the-fly concurrent multiscale methods, the computational domain is divided
into different regions where different simulation methods are applied. A crit-
ical issue in both methods is the correct mechanical and thermodynamical
158 Atomistic Modeling of Materials Failure
Fig. 5.1 A summary of a hierarchical multiscale scheme that can be used to develop
an understanding of the behavior of materials across scales in length and time
Fig. 5.2 Overview over the process of predictive multiscale modeling. Quantitative
predictions are enabled via the validation of key properties, which then enables
to extrapolate and predict the behavior of systems not included in the initial
training set
important insight into the physical basis of the simulated phenomena. The
use of differential multiscale methods is particularly fruitful in the analysis of
very complex processes or phenomena, in which there is no a priori known
theoretical guidance about the behavior. It can also be very helpful in devel-
oping physics-based theories to describe phenomena, as it is possible to center
the attention on the core properties at distinct lengthscales.
A combination of differential and predictive tools can be used for the study
of problems, providing an advantageous method in the analysis in particular
of very complex mechanisms. For example, quantitative methods can be used
to help establish an atomistic model in the regime of interest, thereby forming
a reference system with proper variables and parameters. Then, differential
methods can be used to explore the behavior in the vicinity of this refer-
ence system. It is noted that the reference system is sometimes also referred
to as the control system in the spirit of the analysis of biological labora-
tory experiments. Notably, these laboratory methods were developed based
on the difficulty of obtaining quantitative results. However, one is able to
effectively control the boundary conditions and thereby being able to study
the correlations between microscopic parameters and the system behavior.
Fig. 5.4 Hierarchical modeling of Cybersteel [18]. Subplot (a) shows quantum
mechanical calculations that provide the traction-separation law. Subplot (b) depicts
concurrent modeling of the submicron cell based on the traction-separation law.
Subplot (c) illustrates concurrent modeling of the microcell with the embedded
constitutive law of the submicron cell. Subplot (d) shows results of modeling the
fracture of the Cybersteel with embedded constitutive law of the microcell. Subplot
(e) depicts the fracture toughness and the yield strength of the Cybersteel as a
function of decohesion energy, determined by geometry of the nanostructures. Sub-
plot (f) shows snap-shots of the localization induced debonding process. Subplot (g)
summarizes experimental observations. Reprinted from [18], Computer Methods in
Applied Mechanics and Engineering, Vol. 193, pp. 1529–1578, W.K. Liu, E.G. Kar-
pov, S. Zhang, and H.S. Park, An introduction to computational nanomechanics
and materials, copyright c 2004, with permission from Elsevier
forming covalent bonds with the strength and length of bulk Si bonds. Thus,
at the perimeter of the MD/TB region, there are silogens sitting directly on
top of the atoms of the MD region. The MD atoms of the interface, on the
other hand, have a full complement of neighbors, including neighbors whose
positions are determined by the dynamics of atoms in the TB region.
Major successes in the description of silicon fracture through this method
have been the correct depiction of brittle fracture at low temperatures with
experimental agreement with crack propagation speeds, and elucidation of
possible mechanisms of brittle-to-ductile transition in fracture at higher
temperatures.
A general problem with such interface coupling methods is the spurious
reflection of elastic waves (phonons) as the boundaries due to changes in sys-
tem description. In a subsequent paper, the authors reported that there was
no visible reflection of phonons at the FE/MD interface and no obvious dis-
continuities at the MD/TB interface. However, this particular scheme is very
confined to covalently bonded crystalline materials. One of the reasons for this
limitation is that transition regions for other systems (e.g., metals) are more
difficult to implement. More recent ongoing efforts are exploring the possibil-
ity of applying it to metals and metallic alloys, where the MD/TB region has
to be coupled very differently owing to nondirectional bonding in metals.
Also, the use of the TB method at the crack tip to describe bond breaking
and making, limits the size of the crack affected zone computationally, since
TB is a much costlier method than empirical potentials and FE methods.
Thus, in materials where cracks branch off and/or have large plastic zones
or voids around them, the computational requirements for the problem can
escalate drastically.
Other methods in this area coupled DFT level methods with empirical
potentials, as for instance done for the case of metals [185,186]. These methods
also included an incorporation of the quasicontinuum method (see Sect. 5.4.3
for details).
A new development in this field is the bridging scale technique, which
enables a seamless integration of atomistic and continuum formulations
throughout the entire computational domain [18, 19, 187]. A core feature of
this method is that it is assumed that the continuum and atomistic-scale
solutions exist simultaneously in the entire computational domain. Molecular
dynamics calculations are only performed in the parts of the domain where
this level of accuracy is required. This is possible by decomposing the dis-
placement functions into a slowly varying and rapidly varying part (see also
the schematic of the displacement time history as shown in Fig. 2.4). By sub-
tracting the bridging scale from the total solution, the authors arrive at a
coarse-fine decomposition that decouples the kinetic energy of the two sim-
ulations. A major advantage is two different time-step sizes can be used for
the two different scales. This method is particularly suitable for finite tem-
perature applications, enabling to connect atomistic and continuum domains
seamlessly, even at finite temperature. This is achieved by the definition of
5 Multiscale Modeling and Simulation Methods 165
Fig. 5.5 This plot shows a multiscale analysis of a 15-walled CNT by a bridging
scale method. Subplot (a) illustrates the multiscale simulation model. It consists
of ten rings of carbon atoms (with 49,400 atoms each) and a meshfree continuum
approximation of the 15-walled CNT by 27,450 nodes. Subplot (b) shows the global
buckling pattern captured by meshfree method, whereas the detailed local buckling
of the ten rings of atoms are captured by a concurrent bridging scale molecular
dynamic simulation. Reprinted from [18], Computer Methods in Applied Mechan-
ics and Engineering, Vol. 193, pp. 1529–1578, W.K. Liu, E.G. Karpov, S. Zhang,
and H.S. Park, An introduction to computational nanomechanics and materials,
copyright c 2004, with permission from Elsevier
where ni is the quadrature weight signifying how many atoms a given repre-
sentative atom stands for and Ui is the energy of the ith representative atom.
The selection of representative atoms is based on the local variations of the
deformation field. For example, near dislocation cores and on planes under-
going slip, full atomistic resolution is attained with adaptive meshing. Far
away from defects, the density of representative atoms sharply decreases, and
the collective motion of very large number of atoms is described by a small
number of representative atoms.
The quasicontinuum method has been extended to complex Bravais crys-
tals and polycrystalline materials. It has been applied to many problems,
including dislocation structures, interaction of cracks with grain boundaries,
dislocation junctions, and other crystal defects. One drawback is that because
of the particular expressions for energy in the quasicontinuum method, the
actual atomistic methods that can be implemented are limited to ones that can
easily be expressed in a suitable form. Also, finite temperature applications
remain challenging, although some new developments have been proposed
that combine coarse-grained dynamical simulations with the quasicontinuum
method (see, for instance [190]). Further, a fully nonlocal three-dimensional
version of the method has been introduced and applied to the study of nanoin-
dentation. A recent thrust area in the quasicontinuum field has been on
incorporating ab initio methods, such as orbital free DFT, for example, instead
of relying entirely on empirical potentials.
One potential pitfall of the quasicontinuum method is the so-called “ghost
force” at the interface between the coarse-grained representative atoms and
the atomically resolved regions [35]. The error arises because of the discontinu-
ity between neighboring cells where the cell sizes are less than the range of the
atomistic potential. Care must be taken to correct these “ghost forces.” Qua-
sicontinuum approach also shares some features with hierarchical methods as
the constitutive equations for FE nodes are drawn from atomistic calculations,
and hence there is a message passing across scales.
To exemplify the approach and to illustrate how simulation domains
appear in these methods, we review two examples. The quasicontinuum
method finds particularly useful applications in studies of fracture and defor-
mation, as it is illustrated here in a simple example of a thin films constrained
5 Multiscale Modeling and Simulation Methods 167
by a substrate. A set of results for this case is shown in Fig. 5.6. Here we inves-
tigate a thin copper film with a (111) surface on a rigid substrate (the film
thickness is hf ≈ 30 nm). The interatomic interactions are modeled by Voter
and Chen’s EAM potential for copper [34,35]. We consider a crack orthogonal
to the surface. Such a crack could for instance be created by grain boundary
cracking or constrained grain boundary diffusion [46]. Figure 5.6a shows dif-
ferent snapshots as the lateral mode I opening loading of the film is increased
(the black line indicates the interface of substrate and thin film). The atomic
region adapts and expands, as dislocations gliding on glide planes parallel
to the film surface are nucleated and flow into the film material. Figure 5.6b
shows a zoom into the crack tip region.
Figure 5.7 shows another example application of the quasicontinuum
method, here illustrating the simulation of a nanoindentation experiment
[191].
A coupled atomistic and discrete dislocations (CADD) method has been
developed [192], exemplifying a multiscale approach aimed at coupling a fully
atomistic region to a “defected” dislocation dynamics region. In the CADD
method, dislocations in the continuum region are treated with a standard dis-
crete dislocation method, and the atomistic region can have any kind of atomic
scale defects. Key strengths are automatic detection and smooth passing of
dislocations back and forth in the atomistic and continuum regions. So far, this
approach is restricted to two-dimensional, quasistatic problems. Good agree-
ment to fully atomistic simulations has been shown in atomic scale void growth
and two-dimensional indentation problems. A major challenge to extension
168 Atomistic Modeling of Materials Failure
Basic Concepts
The first principles based reactive force field ReaxFF (see also discussion
in Sect. 2.6.4) has been shown to provide the versatility required to predict
catalytic processes in complex systems nearly as accurate as QM at compu-
tational costs closer to that of simple force fields, including the capability
to describe charge transfer during chemical reactions. There have been other
attempts of modeling charge transfer in metal/metal-oxide systems, such as
the modified charge transfer-embedded atom method potentials [194, 195].
170 Atomistic Modeling of Materials Failure
The force and energy contribution from different simulation engines is weighted
as shown in Fig. 5.8. Every computational engine i has a specific weight wi
associated with it that describe how much the energy of this particular com-
putational engine contributes to the total energy. Thus, for two computational
engines, the total Hamiltonian of the system is written as (here done for a
combination of ReaxFF and EAM, but a similar approach can be used to
couple ReaxFF):
where
Fig. 5.8 The interpolation method for defining a mixed Hamiltonian in the tran-
sition region between two different paradigms. As an alternative to the linear
interpolation we have also implemented smooth interpolation function based on
a sinusoidal function. This enables using slightly smaller handshake regions thus
increasing the computational efficiency
relative contribution from the two force fields that are being connected in the
transition region.
In this equation, the parameter wReaxFF (x) is the weight of the reactive
force field in the handshaking region. Forces on individual atoms given by the
negative of the partial derivatives of Htot with respect to the atom’s coordi-
nates (see the discussion in Sect. 2.7.2 and (2.54), noting also that the kinetic
part of the Hamiltonian does not contribute to the forces as the velocities do
not depend on the atomic coordinates).
Using (5.3) to calculate the forces in the domains in which only a single
force field is used (that is, either the ReaxFF or the EAM domain in this
example) is straightforward to implement. The forces are calculated the same
way as for the individual potentials, either ReaxFF, Tersoff, or EAM. In the
handshaking region, however, due to a gradual change of weights with position,
one obtains
FReaxFF−EAM = wReaxFF (x)FReaxFF + (1 − wReaxFF (x))FEAM
∂wReaxFF
− (HReaxFF − HEAM ) , (5.5)
∂x
which reduces to
FReaxFF−EAM = wReaxFF (x)FReaxFF + (1 − wReaxFF (x))FEAM
∂wReaxFF
− (UReaxFF − UEAM ) , (5.6)
∂x
since the difference in the kinetic part of the different force fields in the
transition region is identical (it only depends on the particle linear momenta).
172 Atomistic Modeling of Materials Failure
Fig. 5.9 Example of the energy landscape of two force fields, a ReaxFF reactive
force field and a nonreactive force field. The plot illustrates that the two models
yield a similar energy landscape for small deviations from the minimum potential
well, the equilibrium position. An exemplification of this effect specifically for silicon
is shown in Fig. 6.108
This equation can be simplified quite a bit based on two conditions. First,
if wReaxFF (x) varies slowly in the spatial domain from zero to one at the edges
of the handshaking regions (that is, if the gradient ∂wReaxFF /∂x is small), the
last term in the equation can be neglected. Further, if the potential energies
for the reactive force field and the EAM method are almost the same (that is,
if the difference UReaxFF −UEAM is negligible in the transition region), the last
term in (5.6) can be neglected. Both conditions lead to a simplified expression
that only involves the weighting of forces in the transition region and not the
considerations of the potential energies.
NC
Htot = wi Hi , (5.8)
i=1
NC
wi = 1. (5.9)
i=0
The forces and energies are weighted accordingly, and the force vector of an
atom j is calculated as
Fj = Fj,i wi , (5.10)
i=0,...,N
Since all simulation tools and engines can be called from a Python scripting
level, the scale agnostic combinations of various modeling approaches can
easily be realized. This strategy enables complex simulations to be simplified
to a series of calls to various modules and packages, whereas communication
between the packages is realized through the CMDF central data structures
that are of no concern to the applications scientist. An excerpt of a CMDF
script is shown in Fig. 5.10. CMDF is designed to:
• Provide a general, extensible approach of a simulation environment uti-
lizing a library of a variety of computational tools spanning scales from
quantum mechanics to continuum theories.
• Establish a reusable library of highly complex computational tools that
can be used as black boxes for most applications, while being initialized
with standard parameters for easy usage in standard cases.
• Enable atomistic applications to be used by engineers and experimen-
tal scientists, while retaining the possibility of building highly complex
simulations and models.
• Close the gap in coupling fundamental, quantum mechanical methods
such as DFT to the ReaxFF reactive force field, to nonreactive force field
descriptions (e.g., DREIDING, UFF).
• Provide a test bed for developing new model and algorithms, making it
simple to develop new communication channels between computational
engines (e.g., developing a new force fields combining distinct methods as
QEq, Morse potentials, ReaxFF, or M/EAM).
The CMDF approach reviewed here is only one out of many other
approaches. Many other software suites such as Konrad Hinsen’s Molecular
Modeling Toolkit (MMTK) or the CAMD Open Software project (CAMPOS)
of the Center for Atomic-Scale Materials Design at Danmarks Tekniske Uni-
versitet provide similar approaches. In addition, codes like NAMD can also be
driven by a Python script, providing further opportunity to integrate other
codes.
Fig. 5.10 Example CMDF script (upper part) and schematic of the structure of
CMDF (lower part)
Not only multiscale methods are developed to bridge spatial dimensions, but
also other methods are focused on bridging across vast timescales. In classical
molecular dynamics schemes it is in principle possible to simulate arbitrarily
large systems, provided sufficiently large computers are available. However,
the timescale remains confined to several nanoseconds. Surprisingly, this is also
true for very small systems (independent of how large computers we use). The
reason is that very small systems cannot be effectively parallelized. Also, time
cannot easily be parallelized. Therefore, surprisingly there exists little trade off
176 Atomistic Modeling of Materials Failure
Fig. 5.11 Study of a nanoscale elliptical penny-shaped crack in nickel, filled with
O2 , illustrating the hybrid ReaxFF-EAM approach (crystal is loaded in tension, in
the horizontal direction)
between the desired simulation time and desired simulation size. This problem
is referred to as the time-scale dilemma of molecular dynamics [89, 134, 198].
Many systems of interest spend a lot of time in local free energy minima
before a transition to another state occurs. In such cases, the free energy
surface has several local minima separated by large barriers. This is compu-
tationally highly inefficient for simulations with classical molecular dynamics
methods.
An alternative to classical molecular dynamics schemes is using Monte-
Carlo techniques such as the Metropolis algorithm. In such schemes, all events
and their associated energy must be known in advance. Note that in kinetic
Monte-Carlo schemes all events and associated activation energy that take
place during the simulation should be known in advance. For that purpose,
the state space for the atoms has to be discretized on a lattice. Besides having
to know all events, another drawback of such methods is that no real dynamics
is obtained.
To overcome the time-scale dilemma and still obtain real dynamics while
not knowing the events prior to the simulation, a number of different advanced
simulation techniques have been developed in recent years (for a more
5 Multiscale Modeling and Simulation Methods 177
extensive list of references see [199]). They are based on a variety of ideas, such
as flattening the free energy surface, parallel sampling for state transitions,
and finding the saddle points or trajectory-based schemes. Such techniques
could find useful applications in problems in nanodimensions. Time spans of
microseconds, seconds, or even years may be possible with these methods.
Examples of such techniques are the parallel-replica (PR) method [200, 201],
the hyperdynamics method [202], and the temperature-accelerated dynam-
ics (TAD) method [203]. These methods have been developed by the group
around Voter [89] (further references could be found therein) and allow cal-
culating the real time-trajectory of atomistic systems over long time spans.
Other methods have been proposed by the group around Parrinello, who for
instance developed a Non-Markovian coarse grain dynamics method [199].
The method finds fast ways out of local free energy minima by adding a bias
potential wherever the system has been previously, thus quickly “filling up”
local minima.
The methods discussed in these paragraphs could be useful for modeling
deformation of nanosized structures and materials over long time spans, such
as biological structures (e.g., mechanical deformation of proteins and proper-
ties at surfaces). A drawback in many of these methods is that schemes to
detect state transitions need to be known. Also, the methods are often only
effective for a particular class of problems and conditions.
We give an example of using the TAD method in calculating the sur-
face diffusivity of copper (modeled by an EAM potential [34]). We briefly
review the method. The simulation is speed up by simulating the system at
a temperature higher than the actual temperature of interest. Therefore, in
this method two temperatures are critical: The low temperature at which the
dynamics of the system is studied, and a high temperature where the system
is sampled for state transitions during a critical sampling time. This critical
sampling time can be estimated based on theoretical considerations in transi-
tion state theory [89]. For every state transition, the time at low temperature
is estimated based on the activation energy of the event. Among all state
transitions detected during the critical sampling time, only the state transi-
tion that would have occurred at low temperature is selected to evolve the
system and the process is repeated.
To calculate the surface diffusivity of copper, we consider a single atom on
top of a flat [100] surface as shown in Fig. 5.12. The atom is constrained to
move at the surface. The total simulation time approaches ∆t = 3 × 10−4 s.
This is a very long timescale compared to classical molecular dynamics
timescales (see Fig. 2.11). The surface diffusivity is calculated according to
| xi (t) − xi (t0 ) |2
Ds = lim . (5.11)
t→∞ 6(t − t0 )
Fig. 5.12 Atomistic model to study surface diffusion of a single adatom on a flat
[100] copper surface
Fig. 5.13 Study of atomic mechanisms near a surface step at a [100] copper surface.
The living time (or temporal stability) of states A (perfect step) and B (single atom
hopped away from step) as a function of temperature. The higher the temperature,
the closer the living times of states A and B get
Fig. 5.14 Snapshots of states A (perfect step) and B (single atom hopped away
from step)
stable up to several hundred seconds. Figure 5.14 shows the two states in a
three-dimensional atomic plot. Table 5.5 summarizes the different activation
energies is higher than that of the reverse process. The activation energy to
get from state (A) to state (B) is higher than that of the reverse process. This
immediately explains why state (B) is not as stable as state (A).
Such methods have recently also been applied to better understand rate
dependence effects in the deformation of metals [21]. In this work, the authors
illustrated the rate dependence of twinning of metals across a large range
of timescales, showing how simulation and experimental results can be con-
nected. The authors used a combination of the CADD method to reduce
the number of atomic degrees of freedom, together with the parallel replica
method. This method enabled them to study dynamical materials failure
mechanisms over many orders of magnitudes of timescales (see Fig. 5.15).
The brief examples reviewed here illustrate the great appeal of these
advanced simulation techniques. Experimental techniques are currently not
able to provide the resolution in space and time to track the motion of sin-
gle atoms. On the other hands, advanced molecular dynamics simulation
techniques can track the motion of atoms on a surface on a relatively long
timescale, with a very high resolution of time.
180 Atomistic Modeling of Materials Failure
5.6 Discussion
Multiscale simulation methods have developed quite significantly over the past
decades. In particular in the past 10 years, many new methods have been
developed that contributed to an extensive database of available methods.
5 Multiscale Modeling and Simulation Methods 181
The new methods have contributed both to extend the accessible lengthscales
as well as timescales. The methods find useful applications for both scientific
applications (see for instance Fig. 5.15) and also for new methods in the design
of novel materials (see for instance Fig. 5.4).
Part III
Figure 6.1 shows a picture of shattered glass. Glass is a very brittle material
that typically shatters into many pieces upon fracture.
186 Atomistic Modeling of Materials Failure
Fig. 6.2 Characteristic length scales associated with dynamic fracture. Relevant
length scales reach from the atomic scale of several Ångstrom to the macroscopic
scale of micrometers and more
It will be shown that the predicted limiting speed of cracks agrees with the
simulation result and the harmonic atomistic model can be used as a reference
system.
Section 6.6 focuses on the role that material nonlinearities play on the
limiting speed of cracks propagating along a prescribed straight fracture path.
It will be shown that hyperelasticity can govern dynamic fracture when the
size of the nonlinear region around the crack tip approaches a newly discovered
length scale associated with energy flux to the crack tip. The characteristic
energy length scale helps to explain many experimental and computational
results. The analysis illustrates that under certain conditions, cracks can break
through the sound barrier and move supersonically through materials. An
important aspect of the analysis is the prediction of intersonic mode I cracks.
The preceding section focuses on the dynamics of constrained cracks,
Sect. 6.7 focuses on the dynamics of unconstrained cracks and the effect of
hyperelasticity. The main focus is an investigation of the critical crack speed
when straight crack motion becomes unstable. By a systematic study with dif-
ferent model materials representing weak and strong hyperelastic effects we
review evidence that suggests that hyperelasticity governs the critical speed
of crack tip instabilities.
In Sect. 6.8 we discuss inertia properties of cracks by investigating the
dynamics of suddenly stopping cracks. We will show good agreement of sud-
denly stopping mode I cracks with theory and experiment [22,23], and discuss
the dynamics of suddenly stopping mode II cracks with respect to continuum
mechanics theories [208]. We also address the role of material nonlinearities,
and report a Griffith analysis for crack initiation for different interatomic
potentials.
Section 6.9 discusses several aspects of dynamic fracture along interfaces
of dissimilar materials. We will show that mother–daughter mechanisms, for-
merly believed to exist only under mode II loading, also exist in the dynamics
of mode I cracks along interfaces of elastically dissimilar materials. Further, we
illustrate that mode II cracks moving along interfaces of dissimilar materials
feature a mother–daughter–granddaughter mechanism.
The final two sections are devoted to the dynamics of mode III cracks.
Since mode III cracks can only be modeled with three-dimensional models, we
utilize the results of mechanical and physical properties of three-dimensional
solids (as those presented in Sect. 4.4.4).
Section 6.10 contains a discussion of the dynamics of mode III cracks. We
will study a crack in a stiff material layer embedded in a soft matrix, and
confirm the existence of the characteristic length scale for energy flux also
for mode III cracks. The results of atomistic simulations are quantitatively
compared with continuum mechanics theory.
Section 6.11 is dedicated to a case study of studies of brittle fracture of
silicon using a multiparadigm approach coupling ReaxFF force fields with
empirical Tersoff models.
6 Deformation and Dynamical Failure of Brittle Materials 189
form of energy – it was used to create new surface area of the material. This
energy is characterized by γs A = Ws , where A is the newly created surface
area (note that when A = 2aB, Ws = 2γs aB, where a is the crack length and
B the thickness of a specimen, assuming that the crack extends through the
entire thickness).
Griffith first postulated that a crack starts to extend when the decrease
in potential energy due to crack growth equals the energy necessary to create
new material surface. This model can be illustrated in a few simple equations.
The total energy in the system is given by
U = Wp + Ws , (6.1)
where Wp is the potential energy of the cracked material. Thus, crack growth
occurs when
dU dWp dWs
= + = 0, (6.2)
dA dA dA
where
dWp
G := − (6.3)
dA
is defined as the energy release rate, typically denoted by the symbol G. This
quantity was first introduced by Irwin [63]. The criterion to describe the onset
of fracture is then given by
G = 2γs . (6.4)
This condition for fracture initiation can be understood from the first law
of thermodynamics: When a system goes from nonequilibrium to an equilib-
rium state there is a net decrease in energy. Before the crack is nucleated,
the potential energy Wp (corresponding to the elastic energy) increases, since
the applied load increases. Thus, the elastic energy stored in the material
increases. The surface energy Ws remains constant, since the crack has not
yet nucleated. At the critical point, the change in surface energy and the
change in elastic energy with respect to an infinitesimal increase in crack
length increases – the crack starts to nucleate.
To apply the idea Griffith put forward, we must have an expression of
Wp as a function of crack length, load applied, and geometry of the cracked
specimen. The quantity Ws is typically found based on the crack geometry
and the surface energy (here we will only consider cases in which the crack
extends through the entire thickness of a material).
This concept can be illustrated by considering the geometry shown in
Fig. 6.3. To develop an expression of the potential energy as a function of
crack length we consider the beam case shown in Fig. 3.14, noting that the
length of each of the two beams is l. Using beam theory, the potential energy
Wp is given by
P 2 l3
Ws = − (6.5)
3EI
6 Deformation and Dynamical Failure of Brittle Materials 191
Fig. 6.3 Using the solution to the beam problem to predict the critical force P at
which fracture initiates. Subplot (a) shows the geometry of a crack in a beam-like
structure. Subplot (b) shows the representation of the upper and lower part as two
cantilever beams
by utilizing the solution of the beam problem in (3.125) to calculate the poten-
tial energy of the beam under load and by considering that there are two beam
structures.
From this expression the energy release rate is given by
2 3
∂ P l
G= , (6.6)
∂A 3EI
where dA = Bdl with B being the out-of-plane thickness of the beam, and
therefore
P 2 l2
G= , (6.7)
BEI
and with the Griffith condition G = 2γs ,
1 2γs EI
Pf = . (6.8)
B l2
A similar calculation can be carried out for a thin strip geometry with
a semi-infinite crack, as shown in Fig. 6.4. The energy release rate is easily
obtained by considering a representative material element ahead of the crack
and one far behind the tip of the crack, of width ã, thickness B, and the height
of the strip ξ. The elastic energy in such a representative element far ahead
of the crack is given by
σ 2 (1 − ν 2 )
Wp(1) = ξãB (6.9)
2E
where the term ãB corresponds to the volume of this control element, E is
Young’s modulus, ν is Poisson ratio, and σ0 is the applied stress. Behind the
crack, the elastic energy is completely relaxed:
Wp(2) = 0. (6.10)
192 Atomistic Modeling of Materials Failure
Fig. 6.4 Thin strip geometry. The gray arrows indicate the mode I (tensile loading),
by a stress σ0
For the case of an elliptical crack in an infinite medium (see Fig. 1.3 for
the geometry) – a solution developed by Inglis – the expression for Wp is
πa2 σ 2 B
Wp (a, σ) = − , (6.14)
2E
where E = E/(1 − ν 2 ) for plane strain and E = E for plane stress. The
energy release rate is then given by
2 2
∂ πa σ B
G= , (6.15)
∂A 2E
6 Deformation and Dynamical Failure of Brittle Materials 193
and therefore
πaσ 2
G= . (6.16)
E
At the onset of crack growth, G = 2γs , which leads to a condition for the
failure stress as a function of material parameters and crack size a:
2γs E
σf = . (6.17)
πa
Fig. 6.5 Summary of the basic physical processes involved in brittle fracture, that
is, the process of dissipating stored elastic energy toward breaking of chemical bonds
Finally, Fig. 6.5 summarizes the basic physical processes involved in brit-
tle fracture, that is, the process of dissipating stored elastic energy toward
breaking of chemical bonds.
Fig. 6.6 Schematic of cracks under mode I, mode II, and mode III crack loading
Fig. 6.7 Closing a crack by negating the tractions at the tip, as used in the derivation
of the relation between the stress intensity factor and the energy release rate
There exists a unique relationship between the energy release rate G and
the stress intensity factor KI,II,III . By considering the scenarios of closing a
crack with length a + δa to length a by applying proper negating tractions
at the tip of the crack, we can calculate the amount of energy necessary (see
Fig. 6.7):
0
1
GδaB = δΦ = −2B σ22 (x1 )u2 (x1 )dx1 (6.23)
−δa 2
so that the singularity of the crack tip is at x1 − δa. The displacements in the
y-direction are given by
KI −2x1
u2 = 1 − ν2 , (6.28)
E π
noting that the negative sign stems from the fact that for the displacements
we integrate from the left to the right, opposing the direction of the polar
coordinate system. The displacements are obtained by considering the dis-
placement field behind a crack with tip at a + δa, whereas the tractions are
obtained by considering the region ahead of a crack with tip at a.
Evaluating the integral (6.23) with the expressions for stress and displace-
ments leads to the relationship between G and KI :
KI2
G= . (6.29)
E/(1 − ν 2 )
This relationship can be used to obtain the stress intensity factor for a specific
geometry, such as the thin strip case. For this case,
KI = σ0 ξ/2. (6.30)
Fig. 6.8 Mode II loading experimental setup for studies of dynamic fracture in
Homalite-100. Subplot (a) depicts the geometry of the experiment, indicating the
location of projectile impact to generate rapid mode II loading along a weak plane.
The dashed circle displays the view of the circular polariscope for the analysis of the
stress field. Subplot (b) displays the evolution of crack speed as the shear crack prop-
agates along the weak plane. The crack tip speed was obtained from crack length
history (squares) and from shock wave angles (circles) for a field of view around
the notch tip (solid symbols), and for a field of view ahead of the notch (open sym-
bols). The analysis confirms intersonic and supersonic regimes of crack propagation.
Reprinted from Science, Vol. 284, A.J. Rosakis, O. Samudrala, D. Coker, Cracks
Faster than the Shear Wave Speed, copyright c 1999, with permission from AAAS
Fig. 6.9 Enlarged view of the isochromatic fringe pattern around a steady-state
mode II intersonic crack along a weak plane in Homalite-100. Subplot (a) shows
the experimental pattern, and subplot (b) the theoretical prediction [22]. For both
cases, β = 53o and v = 1.47cs . Reprinted from Science, Vol. 284, A.J. Rosakis,
O. Samudrala, D. Coker, Cracks Faster than the Shear Wave Speed, copyright c
1999, with permission from AAAS
stress and deformation fields around the crack tip suggest good agreement of
the atomistic model with the continuum theory.
In the second part of the section, the one-dimensional model is used to
study crack dynamics in nonlinear materials. On the basis of the concept of
local elastic properties [27], an analytical model is proposed for the dynamics
of the crack and for the prediction of the deformation field. An important
prediction of this model is the possibility of supersonic crack propagation
if there is a local elastically stiff region close to the crack tip. By atomistic
simulations, it is shown that this hypothesis is true and that an elastically stiff
zone at the crack tip allows for supersonic crack propagation. This suggests
that local elasticity at the crack tip is crucial for the dynamics of fracture. In
most classical theories of fracture it is believed that there is a unique definition
of how fast waves propagate in solids. Our results prove that this concept
cannot capture all phenomena in dynamic fracture, and instead should be
replaced by the concept of local wave speeds.
6.4.1 Introduction
Most of the theoretical modeling and most computer simulations have been
carried out in two or more dimensions (e.g., [22, 61, 155, 219]). One of the
important objectives in understanding hyperelasticity in dynamic fracture is
to obtain analytical models. However, finding analytical solutions for dynamic
fracture in nonlinear materials seems extremely difficult, if not impossible
in many cases [230]. To investigate the nonlinear dynamics of fracture at a
simple level, we propose a one-dimensional (1D) model of dynamic fracture,
as originally reported by Hellan [231] for linear elastic material behavior.
The model can be described as a straight, homogeneous bar under lateral
loading σ0 . Part of the bar is attached to a rigid substrate, and this attachment
can be broken, so that a crack-like front of debonding moves along the bar (in
the following, we refer to the front of debonding as crack tip). The model is
depicted in Fig. 6.10. A complete analytical solution of this problem is avail-
able based on linear elastic continuum mechanics theory [231]. Theory predicts
6 Deformation and Dynamical Failure of Brittle Materials 203
agreement at the two scales. The results provide evidence that the predictions
of the new continuum model for a bilinear stress–strain law are reasonable.
We will show that the crack limiting velocity is indeed associated with the
elastic properties localized to the crack tip.
∂σ ∂2u
=ρ 2, (6.35)
∂x ∂t
where ρ is the material density (this is a simplified version of (3.24)). This
equation can be combined with Hooke’s law, given by
∂u
σ = Eε = E , (6.36)
∂x
with E as Young’s modulus and ε as strain. This leads to a partial differential
equation to be solved for u(x, t)
∂u2 ∂2u
c20 = (6.37)
∂x2 ∂t2
where c0 is the wave velocity. It can be shown that (6.37) has solutions of the
form u = f (x ∓ c0 t) = f (ξ), because
x-direction. When the crack front has moved by the length da = dtȧ, a point
which has formerly been situated at the crack tip is displaced backward by
du = −εda, because the detached part of the string has attained the axial
strain ε. A crack represents a signal constrained to be travelling at a lower
velocity than ȧ ≤ c0 . According to (6.40), this corresponds to the particle
velocity
ȧ
u̇ = −εȧ = − σt , (6.41)
E
where σt is the local stress to the left to the crack tip. Furthermore, we assume
that the stress behind the crack tip can be expressed as the sum of the initial
stress σ0 , and an emitted stress wave to the separation, σe , so that
σt = σ0 + σe . (6.42)
σ02
= R0 (6.51)
2E
where R0 is fracture surface energy defined as the energy required to break
atomic bonds per unit crack advance. Since R(ȧ) is generally not known, it has
to be determined from experiments or numerical calculations. To determine
the curve R(α), one may apply a stress σ0 , measure the crack limiting speed
α, and calculate the value of R(α) as
σ02 1 − α
R(α) = . (6.52)
2E 1 + α
If this curve is known, the crack equation of motion can be solved completely.
A simplification to make the one-dimensional problem solvable in closed form
is to assume a constant dynamic fracture toughness, thus R(α) = R0 g(α).
This assumption is usually a good approximation for low propagation veloc-
ities. For higher velocities close to the crack limiting speed ȧ → c0 , however,
it is expected that even though the stress is increased significantly, the crack
speed will not change much [22, 231].
Equation (6.52) states that the dynamics of the crack responds imme-
diately to a change in loading or fracture energy, implying that the crack
carries no inertia. However, the information about the change in loading or
fracture resistance is transmitted with the sound velocity, as indicated by
(6.39). When the crack suddenly stops from a high propagation velocity, the
local strain immediately changes from the magnitude at high propagation
velocity to εt = ε0 (static field solution). This can be verified using (6.46).
The crack carries no inertia since the crack immediately responds to a change
in the boundary conditions. The crack tip velocity responds instantaneously
to a change in fracture energy.
We summarize the predictions of the continuum model. A critical loading
is necessary to initiate fracture (that is, to break the first bond), similar to
6 Deformation and Dynamical Failure of Brittle Materials 207
the Griffith condition. While the crack propagates, it sends out a stress wave
with a magnitude depending on the crack propagation velocity. For α = 0, no
stress wave is emitted and the local stress σt = σ0 . For α → 1, the local stress
wave has magnitude σt = σ0 /2. For intermediate values of α, the stress wave
magnitude decreases monotonically from σ0 to σ0 /2, as α increases from zero
to one. The theory predicts that the largest velocity the crack may achieve
is the sound velocity c0 , hence αmax = 1. As higher-dimensional cracks, it is
predicted that the 1D crack carries no inertia.
zone of large-deformation is found in regions far away from the crack tip, but
the zone of small deformation close to the crack tip is associated with large
deformation gradients. Considering the stress field in the vicinity of a moving
crack based on the continuum model, this can be verified straightforwardly,
since 1/2σ0 ≤ σt ≤ σ0 , and the stress ahead of the crack is zero, while it is
σ0 far behind the crack tip. Therefore, if the stress–strain law shows softening
with increasing strain, the elastic properties at the crack tip tend to be stiffer
than in the far-field. Even though there exists this qualitative difference of the
elastic fields near a 1D and a higher-dimensional crack, the dynamics of these
systems can be compared immediately if proper interpretation of the features
of the deformation fields is done. A stiffening potential in higher-dimensional
models tends to yield an elastically stiff zone at the crack tip. In the 1D model,
a potential softening with strain is required to provide an elastically stiff zone
at the crack tip.
Here we focus on the case when El > Eg , which implies that there exists
a region close to the crack tip where the material is elastically stiffer than in
regions far away. In a string of atoms, the stress σ due to strain ε is given by
El ε if ε < εon ,
σ= (6.53)
Eg (ε − εon ) + El εon if ε ≥ εon ,
where εon is the critical onset strain for hyperelasticity. Therefore, the initial
equilibrium strain due to an applied stress σ0 is given by
σ0 /El if ε0 < εon ,
ε0 = (6.54)
σ0 /Eg − εon El /Eg + εon if ε0 ≥ εon .
The concept of local and global elastic properties enables us to define two
reduced crack speeds αg = v/cg and αl = v/cl . We note that
El
αg = · αl , (6.56)
Eg
Fig. 6.13 Continuum model for local strain near a supersonic crack. The plot shows
a schematic of the two cases 1 (subplot (a)) and case 2 (subplot (b))
the local elastic properties in this case, and due to the signal travelling to
the left with lower strain, the hyperelastic stiff region expands to the left
of the crack (see Fig. 6.13a).
• Case 2: The local strain near the crack tip εt is larger than the onset
strain of the hyperelastic effect εon . Therefore, the region of hyperelastic
material response remains confined to the vicinity of the moving crack tip
(see Fig. 6.13b).
We assume that the crack advance and material detachment occurs in a region
with local elastic properties (associated with El ), as shown schematically
in Fig. 6.13a. The local emitted strain in the hyperelastic case is therefore
predicted to be
1 σ0
εt = . (6.57)
1 + αl El
This equation is only valid if εt < εon , that is, the local strain wave lays
completely within the zone of local (stiff) elastic properties. An important
implication of the assumption is that the limiting speed of the crack is deter-
mined by the local elastic wave speed. Since αmax
l = 1, and αg = 2αl , the crack
can propagate supersonically with respect to the global elastic properties. The
210 Atomistic Modeling of Materials Failure
ratio of local strain to initial strain is given by combining (6.57) and (6.54)
⎧ 1
⎪
⎨ if ε0 < εon ,
1 + αl
εt /ε0 = 1/(1 + αl )σ0 /El (6.58)
⎪
⎩ if ε0 ≥ εon .
σ0 /Eg − εon El /Eg + εon
Here we consider the case when εt as given by (6.57) is larger than the onset
strain of the hyperelastic effect, that is εt ≥ εon . The emitted strain wave
cannot lay within the soft material since crack motion is supersonic with
respect to the global soft elastic properties and no signal faster than the sound
speed can be transported through the material. Therefore, a shock wave will
be induced when the elastic properties change from stiff to soft. The signal
of stress relief is transported through the soft material as a secondary wave
and represents a wave travelling at cg , the wave speed of the soft material,
independent of how fast the crack propagates.
In summary, there are two waves propagating behind the crack tip. The
first wave features a magnitude
(1)
εt = εon , (6.60)
representing a signal travelling in the soft material at the wave speed of the
soft material, also independent on the crack speed. The model is schematically
summarized in Fig. 6.13b.
The particle velocity u̇ is given by
1 σ0
u̇(2) = − − εon c0,g . (6.62)
2 Eg
strain near the moving crack is smaller than the onset strain of the hyperelastic
effect and case 2 when it is larger.
In case 1, the local elastic properties completely govern the dynamics of
the crack. As a consequence, the model predicts that the crack can propagate
supersonically. The upper limit of the propagation speed is given by the wave
speed associated with the local elastic properties.
In case 2, detachment of the material occurs completely in the hyperelastic
region and remains confined during crack growth. In this case, two waves with
(1) (2)
magnitude εt and εt are moving behind the crack tip, one is a shock front
associated with the change in elastic properties and the other represents a
signal travelling in the elastically soft material carrying the stress relief due
to crack propagation at the wave speed of the soft material. As the size of the
hyperelastic zone shrinks with decreasing εon , the limiting crack speed is also
expected to decrease and approach the limiting value of αg → 1 for εon → a0 .
This is because the material detachment eventually occurs completely within
the zone of soft elastic properties. On the other hand, if εon is chosen larger,
the stiff zone expands and eventually the situation corresponding to case 1 is
attained when εt < εon and the dynamics is completely governed by the local
elastic properties.
In any case, when the crack propagates supersonically, a dramatic reduc-
tion in the ratio εt /ε0 is possible due to the local stiffening effect.
According to Fig. 6.11, the atoms are numbered from left to right with increas-
ing index, with a total number of atoms Nt . We assume that atoms with index
i > Nf are attached to the substrate, and atoms with i ≤ Nf are free and
only interact with other nearest neighbor atoms. The state of an atom i is
uniquely defined by a position xi and its velocity ẋi . The mass of each par-
ticle is m = 1. Only nearest neighbor interaction is considered. The systems
contain up to 20,000 atoms, which equals a string of atoms of length of about
20 µm in physical dimensions. To study one-dimensional fracture, we have
developed a specific molecular dynamics code optimized for one-dimensional
analyses.
The basis for the atomic interactions is the Lennard-Jones interatomic
potential defined in (2.30). We express all quantities in reduced units, so
lengths are scaled by the LJ parameter σ which is assumed to be unity in
this study, and energies are scaled by the parameter 0 = 1/2, the depth of
the minimum of the LJ potential. The reduced temperature is kB T /0 with
kB being the Boltzmann constant. To study a harmonic system, we expand
the LJ potential around its equilibrium position a0 = 21/6 ≈ 1.12246, and
consider only first-order terms yielding harmonic atomic interactions.
The mass of each atoms in the models is assumed to be unity, relative
to the
reference mass m∗ . The reference time unit is then given by t∗ = mσ 2 /. For
212 Atomistic Modeling of Materials Failure
where k is the spring constant for interatomic interaction, and kp is the spring
constant of the pinning potential. The variable
and the variable xi is the current position of the atom i. The variable x0,i
stands for the initial position of atom i. We integrate the equations of motion
using a velocity verlet algorithm,
and choose a time step ∆t = 0.000, 036 in
reduced atomic units of σ m/.
When all strain is equilibrated in the free standing part of the string, we
begin the fracture simulation phase where the bonds to substrate have finite
energy. The total energy of the system is then given by
1
1
Utot = k(rij − a0 ) +
2
H(i − Nf )H(rbreak − r̂i )kp r̂i
2
(6.65)
i,j
2 i
2
where rbreak is the snapping bond distance for the pinning potential. The
fracture energy R0 in (6.51) is given by
1 kp r̂2
R0 = . (6.66)
2 a0
Assuming a stress–strain law as given by (6.36), we define a Young’s modulus
for a one-dimensional string of atoms [114]
E = k a0 . (6.67)
6 Deformation and Dynamical Failure of Brittle Materials 213
Fig. 6.14 Magnitude of the local stress wave for different crack propagation
velocities from atomistic simulations, in comparison with the theory prediction
larger velocities, the curve deviates significantly from a constant and increases
dramatically. This behavior is expected from theory [232] (and also for higher
dimensions, as discussed, for instance, in [22]).
If α < 1, the crack front propagates slower than the local wave front behind
the crack. If the material left to the crack is of finite length, the reflected wave
from the left end will eventually hit the crack tip at a time
2L + ∆a
δt = , (6.70)
c0
where L denotes the initial free length of the bar, and ∆a is the distance the
crack has travelled until it is hit. Once the reflected wave front impinges the
crack, the stress will suddenly increase causing a jump in crack propagation
speed [232]. In the atomistic simulations, we observe this effect, but note that
the crack does not reach a steady-state as predicted by the theory. Instead, the
6 Deformation and Dynamical Failure of Brittle Materials 215
Fig. 6.15 Dynamic fracture toughness for different crack propagation velocities
crack speed seemed to decrease continuously, much below the value predicted
by the theory. During this process, the temperature in the system increased
continuously and energy seems to be dissipated into heat (“thermalization”
process).
To investigate the dynamics of a suddenly stopping crack, we let the crack
propagate at a high velocity α ≈ 0.9, and then force the crack to stop. This is
achieved by setting r̂ to a large number r̂∞ r̂ for all atoms with identifica-
tion number greater than istop > Nf . This forces the crack to suddenly stop
once the crack tip reaches the atom with index equal istop :
r̂0 if i < istop ,
r̂(i) = (6.71)
r̂∞ if i > istop .
The simulation results illustrate that the theory prediction is satisfied, and
the local strain immediately attains the magnitude ε0 as soon as the crack is
stopped. The static field spreads out with the wave velocity. The results are
plotted in Fig. 6.16.
The discussion of the suddenly stopping one-dimensional crack proves that
a one-dimensional crack carries no inertia. According to this observation, the
crack tip velocity should immediately respond to a change in the fracture
energy. For example, if the crack senses a higher fracture surface energy,
the velocity should instantaneously decrease, and if the crack senses a lower
fracture surface energy, vice versa. We test this statement by introducing a
periodically varying fracture surface energy as the crack propagates along x.
The velocity should change in antiphase with the change of fracture surface
216 Atomistic Modeling of Materials Failure
Fig. 6.16 Strain field near a suddenly stopping one-dimensional crack. The crack is
forced to stop at x ≈ 790. As soon as the crack stops (at x = 550), the strain field
of the static solution is spread out with the wave speed
where r̂0 is the value around the snapping distance. The bond snapping
distance oscillates with amplitude ∆r̂ and period factor p.
In Figs. 6.17 and 6.18, results are plotted for r̂0 = 0.008, ∆r̂ = 0.003, and
p = 30. The velocity oscillates around α ≈ 0.6, which is in agreement with
the velocity of a crack under loading σ0 = 0.02 and a fracture toughness of
r̂ = 0.008. The same observation applies to the upper and lower limit of the
propagation velocity, which correspond to the limiting velocity of the crack
if it would be propagating along a path with constant fracture energy of the
corresponding magnitude. Therefore,
6 Deformation and Dynamical Failure of Brittle Materials 217
Fig. 6.17 Prescribed fracture toughness and measured crack velocity as the crack
proceeds along x
where v0 is the velocity associated with r̂0 , and ∆v ≈ 1.3 can be approximated
by the difference of the propagation velocity associated with r̂0 + ∆r̂.
i,j
2 2
1 (6.74)
+ H(i − Nf )H(rbreak − r̂i )kp r̂i ,
2
i
2
where ron is a potential parameter allowing for different onset points of the
hyperelastic effect (thus controlling the strength of the hyperelastic effect),
and
ron − a0
εon = . (6.75)
a0
The choice of β allows for different types of nonlinearities. If −1 < β ≤ 0, the
potential softens with strain, and if β = 0, the model reduces to harmonic
interactions. The small-perturbation spring constant is always given by k0 =
k, and the large-strain spring constant is given by k1 = (1 + β)k. Elastic
properties for β = −3/4 are shown in Fig. 6.19, which plots the atomic stress
σ vs. atomic separation and the tangent modulus E. The local sound velocity
218 Atomistic Modeling of Materials Failure
Fig. 6.18 Strain field of a crack travelling in a material with periodically varying
fracture toughness
c0,l is readily obtained from E. The figure shows that the tangent modulus
softens with strain. We reiterate that if the stress–strain law softens with
strain, the elastic properties at the crack tip are stiffer than in the far-field.
The simulation procedure when using the bilinear stress–strain law is iden-
tical to the previously described procedure. However, the dynamics of the
crack with the bilinear stress–strain law is significantly different from the har-
monic case. It is observed that the crack can propagate supersonically with
respect to the global elastic properties. Figure 6.20a plots the limiting velocity
of the crack for different values of the potential parameter ron . For large val-
ues of ron , the local hyperelastic zone becomes larger and the limiting velocity
approaches Mach 2, or αg ≈ 2. For ron → a0 , the hyperelastic zone shrinks
and the velocity of the crack approaches αg ≈ 1. This plot proves that the
limiting velocity of the crack is very sensitive with respect to the potential
parameter ron . A small change in ron affects the extension of the hyperelastic
area and has impact on the limiting velocity. The simulation results prove
that supersonic crack propagation is possible even if the hyperelastic zone is
very small.
6 Deformation and Dynamical Failure of Brittle Materials 219
Fig. 6.19 Elastic properties associated with the biharmonic interatomic potential,
for ron = 1.125 and Eg = 8 = 1/4El
When the crack is propagating at αg > 1, the local strain wave has a
magnitude of less than 50% of the equilibrium strain. This is in disagreement
with the classical theory stating that the local strain wave is always equal
or larger than 50% of the equilibrium strain for a crack propagating at the
limiting speed (sound velocity). However, these observations can be explained
well by the new continuum model proposed based on the concept of local
elastic properties. Figure 6.20b plots a comparison of the continuum model
with molecular dynamics simulation results of supersonic crack propagation.
The agreement is reasonable. The regimes where case 1 and case 2 are valid is
also indicated. Figure 6.21 depicts the strain field in the vicinity of a supersonic
crack for ron = 1.124.
Figure 6.22 depicts the particle velocity field near the moving supersonic
crack.
Fig. 6.20 Subplot (a) Velocity of the crack for different values of the potential
parameter ron . The larger ron , the larger the stiff area around the crack tip. As the
hyperelastic area becomes sufficiently large, the crack speed approaches the local
wave speed αl = 1 corresponding to αg = 2. Subplot (b) shows a quantitative com-
parison between theory and computation of the strain field near a supersonic crack
as a function of the potential parameter ron . The different regimes corresponding to
case 1 and case 2 are indicated. The loading is chosen σ0 = 0.1, with kp /k = 0.1
and r̂ = 0.001
Fig. 6.21 Sequence of strain field near a rapidly propagating supersonic 1D crack
moving with Mach 1.85 for ron = 1.124. The primary (1) and secondary wave (2)
are indicated in the plot. The wave front (1) propagates supersonically through the
material
is true even if the hyperelastic region is highly confined to the crack tip. The
observation of supersonic crack motion has been found by other researchers
as well in 2D and 3D studies [219]. The finding that the dynamics of the
crack is governed by the local elastic properties (the local wave speed) has
been predicted theoretically [27] and observed previously [155, 219]. The case
of stiffening material response corresponds to materials such as polymers,
showing a hyperelastic stiffening effect. Due to the large deformation in the
vicinity of the crack, the elastic properties in such materials are stiffer close
to the crack than in regions far away from the crack. Laboratory experiments
of dynamic fracture in such materials could provide further insight into the
nature of hyperelastic stiffening dynamic fracture and associated supersonic
crack propagation.
In this section, we have concentrated on the case when local elastic prop-
erties are stiffer than in the far-field elastic properties, crack propagation is
supersonic. In the same sense, if the local elastic properties are softer, crack
propagation must be subsonic on a local scale. We have also performed similar
222 Atomistic Modeling of Materials Failure
Fig. 6.22 Particle velocity field near a supersonic crack, comparison between theory
and simulation
Further,
γl = 1 − (v sin(Θl /cl )2 ), (6.80)
tan(Θl ) = αl tan Θ, (6.81)
γs = 1 − (v sin(Θs /cs )2 ), (6.82)
and
tan(Θs ) = αs tan Θ. (6.83)
The two factors αs and αl are defined as
αs = 1 − v 2 /c2s (6.84)
and '
αl = 1 − v 2 /c2l . (6.85)
The asymptotic stress field in the vicinity of a dynamic crack depends only on
the ratio of crack speed to the wave velocities in the solid. Similar expressions
for the asymptotic field have also been derived for mode II cracks [22].
224 Atomistic Modeling of Materials Failure
The asymptotic field strongly depends on the crack velocity, and has uni-
versal character because it is independent of the details of applied loading.
(1)
The values of σij and the first-order contribution O(1) are determined from
the boundary conditions, and neglected in the remainder of this work since
the first term dominates very close to the crack tip.
In the following sections, we review a systematic comparison of atomistic
simulations and linear elastic continuum theory of the stress and deforma-
tion field near rapidly propagating cracks. Harmonic interatomic potentials
are used to model a linear elastic plane-stress sheet. To compare the results
for different crack velocities, we report atomistic simulations with different
loading rates driving the crack to different terminal velocities.
Figure 6.23 shows the slab geometry used in the simulations. The slab
size is given by lx and ly . The crack propagates in the y-direction, and its
extension is denoted by a. The crack propagates in a triangular hexagonal
lattice with nearest neighbor distance along the crystal orientation shown in
Fig. 6.23. A weak fracture layer is introduced to avoid crack branching by
assuming harmonic bonding in the bulk but an LJ potential across the weak
layer (see also [165]).
Fig. 6.23 Simulation geometry and coordinate system for studies of rapidly
propagating mode I cracks in harmonic lattices
All simulations presented here are two dimensional. Previous studies have
provided evidence that 2D molecular dynamics is a good framework to inves-
tigate the dynamics of fracture [165, 219]. This is because the atomistic
simulations of a two-dimensional solid and a three-dimensional solid show
no difference in the details of the dynamics of the crack. The 2D model cap-
tures important features of dynamic fracture such as surface roughening and
crack tip instabilities [146, 155].
The outline of this section is as follows. It will be shown that in molecu-
lar dynamics simulations of cracks traveling in perfect harmonic lattices the
6 Deformation and Dynamical Failure of Brittle Materials 225
Fig. 6.24 Comparison between σxx from molecular dynamics simulation with har-
monic potential and the prediction of the continuum mechanics theory for different
reduced crack speeds v/cR
Fig. 6.25 Comparison between σyy from molecular dynamics simulation with har-
monic potential and the prediction of the continuum mechanics theory for different
reduced crack speeds v/cR
The principal strain field is shown in Fig. 6.29 for different velocities of
v/cR ≈ 0, v/cR ≈ 0.5, and v/cR ≈ 1. The upper plot is the simulation result,
while the lower part is the prediction by continuum mechanics. We note that
the principal stress field is in good agreement with the continuum theory. The
6 Deformation and Dynamical Failure of Brittle Materials 227
Fig. 6.26 Comparison between σxy from molecular dynamics simulation with har-
monic potential and the prediction of the continuum mechanics theory for different
reduced crack speeds v/cR
Here we discuss the energy flow near a crack tip in molecular dynamics sim-
ulations compared with the continuum theory [22]. A similar study has been
reported in [235]. In contrast to the treatment of the dynamic Poynting vector
for steady-state cracks at high velocities in analogy to the discussion in [22],
the authors in [235] only considered the static Poynting vector to study the
energy radiation of rapidly moving cracks.
The dynamic Poynting vector for a crack moving at velocity v in the
y-direction can be expressed as
228 Atomistic Modeling of Materials Failure
Fig. 6.27 Comparison between hoop stress from molecular dynamics simulation
with harmonic potential and the prediction of the continuum mechanics theory for
different reduced crack speeds v/cR
Fig. 6.28 Comparison between the maximum principal stress σ1 from molecular
dynamics simulation with harmonic potential and the prediction of the continuum
mechanics theory for different reduced crack speeds v/cR
where δij is the Kronecker delta function. The kinetic energy is given by T =
1
2 ρu̇i u̇i , and the strain energy density for an isotropic medium is given by [236]
6 Deformation and Dynamical Failure of Brittle Materials 229
Fig. 6.29 Principal strain field at various crack velocities (a) v/cR ≈ 0, (b) v/cR ≈
0.5, (c) v/cR ≈ 1. In each of the plots (a)–(c), the upper plot is the simulation result
and the lower part is the prediction by continuum mechanics
1 , 2 -
Ψ= 2
σ11 + σ22 − 2νσ11 σ22 + 2(1 + ν)σ12
2
. (6.87)
2E
The
magnitude of the dynamic Poynting vector is calculated as P =
P12 + P22 , and can be identified as a measure for the local energy flow.
Figure 6.32a shows the strain energy field near the crack tip predicted by
both the continuum theory prediction and the molecular dynamics simulation
result. Figure 6.32b shows the magnitude of the dynamic Poynting vector field.
Figure 6.33 shows in panel (a) the continuum mechanics prediction, and in
panel (b) the molecular dynamics simulation result of the dynamic Poynting
vector field in the vicinity of the crack tip, for a crack propagating close to
the Rayleigh speed.
Fig. 6.30 Stress fields close to the crack tip for a crack propagating close to the
Rayleigh velocity v/cR ≈ 1. Plots (a), (b), and (c) show σxx , σyy , and σxy . In each
of the plots (a)–(c), the upper plot is the simulation result and the lower part is the
prediction by continuum mechanics
For mode I cracks considered in this section, the limiting speed is given by
the Rayleigh wave speed [22].
Figure 6.34 shows the crack tip position history a(t) as well as the crack
speed history ȧ(t) for a soft as well as a stiff harmonic material, for a mode
I crack. The results are in consistency with the predicted limiting speed (see
data in Table 4.1).
Similar studies have been carried out for mode II cracks, as is discussed
in [237]. Additional results for mode III cracks will be discussed in Sect. 6.10.
6.5.4 Summary
Fig. 6.31 Particle velocity field close to the crack tip for a crack propagating close
to the Rayleigh velocity, v/cR ≈ 1. Plots (a) shows u̇x and plot (b) shows u̇y . In
each of the plots (a) and (b), the upper plot is the simulation result and the lower
part is the prediction by continuum mechanics
Fig. 6.32 Potential energy field and magnitude of the dynamic Poynting vector.
(a) Potential energy field near a crack close to the Rayleigh speed. (b) Energy flow
near a rapidly propagating crack. This plot shows the magnitude of the dynamic
Poynting vector in the vicinity of a crack propagating at a velocity close to the
Rayleigh speed
Fig. 6.33 Energy flow near a rapidly propagating crack. This plot shows (a) the
continuum mechanics prediction, and (b) the molecular dynamics simulation result
of the dynamic Poynting vector field in the vicinity of the crack tip, for a crack
propagating close to the Rayleigh wave speed
Fig. 6.34 Crack tip history as well as the crack speed history for a soft as well
as a stiff harmonic material (two different choices of spring constants as given in
Table 4.1)
and minima are shifted to slightly smaller values compared to the theory
prediction.
Figure 6.25 illustrates that the shift of the maximum in the σyy (Θ) curve
from about 60◦ to about 80◦ is reproduced only qualitatively. For low velocities
the maximum is found at lower angles around 40◦ , but it approaches the
value of the continuum theory at higher velocities. At 87% of Rayleigh speed,
the difference is only a few degrees. The shear stress σxy shown in Fig. 6.26
also agrees qualitatively with the continuum theory. As in the previous cases,
the angles of local minima and maxima are shifted to lower values in the
6 Deformation and Dynamical Failure of Brittle Materials 233
simulation, but the agreement gets better when the crack velocity is faster.
Even though we see small deviations in σxx , σyy , and σxy , the hoop stress σΘ
agrees quantitatively with the continuum theory as shown in Fig. 6.27.
The angles of the maxima and minima during crack acceleration compare
well with theory. However, the angles of the maxima and minima of the maxi-
mum principle stress shown in Fig. 6.28 are also shifted to slightly lower values.
However, it is observed that two local maxima and one local minima develop
at a velocity of about 73% in quantitative agreement with continuum theory
(“trimodal structure”). The magnitude of the local maxima and minima also
agree quantitatively.
The analysis of the potential energy field near a crack close to the Rayleigh
speed agrees qualitatively with the prediction by the continuum mechanics
theory. As Fig. 6.32a shows, in both theory and computation the field clearly
shows three local maxima with respect to the angular variation (“trimodal
structure”), similar to the principal stress field. At larger distances away from
the crack tip, it is observed that other stress terms begin to dominate in
the simulation, so the distribution of the potential energy deviates from the
prediction by theory. As is expected since only the first term of (6.76) is
considered, these contributions are missing in the continuum solution.
Similar observations also hold for the magnitude of the dynamic Poynt-
ing vector, as it can be verified in Fig. 6.32b. The dynamic Poynting vector
field calculated by molecular dynamics is also in reasonable agreement with
the continuum mechanics prediction. This could be verified in Fig. 6.33. In
both theory (a) and molecular dynamics calculation (b), the orientation
of the dynamic Poynting vector is dominated by the direction opposite to
crack motion. The vector field seems to bow out around the crack tip, an
effect that is more pronounced in the simulation than predicted by theory.
Also, the flow ahead of the crack is larger in simulation than predicted by
theory. At the free surface of the crack, the measurement from the simu-
lation and the prediction by theory show differences. This could be based
on the fact that the continuum theory does not treat surface effects prop-
erly, in particular short-wave length Rayleigh waves (see also discussion in
[235]).
The calculation of the virial stress as shown here does not include the par-
ticle velocity contribution, following the suggestion put forward in [238] on
the linkage between virial stress and Cauchy stress of continuum mechan-
ics. Finally, it is noted that the virial expression of the stress tensor is
classically thought to be only valid under equilibrium conditions [239]. The
results reviewed in this section show that it is approximately valid even under
dynamic fracture conditions which are far from equilibrium.
234 Atomistic Modeling of Materials Failure
Fig. 6.35 The concept of hyperelasticity in dynamic fracture. Subplot (a) shows
the region of large deformation near a moving crack, due to the nonlinear elastic
behavior of solids (subplot (b)). The linear elastic approximation is only valid for
small deformation. Close to crack tips, material deformation is extremely large,
leading to significant changes of local elasticity, referred to as “hyperelasticity” (see
also Fig. 3.10 and related discussion)
6.6.1 Modeling
Fig. 6.36 This figure shows a continuously increasing hyperelastic stiffening effect,
as observed by measuring the elastic properties of a material (subplot (a)). The
increasingly strong hyperelastic effect is modeled by using biharmonic potentials,
thereby capturing the essential physics: A small-strain spring constant k0 and a
large-strain spring constant k1 (subplot (b)), where the ratio of the two is defined
as kratio = k1 /k0 . The bilinear or biharmonic model allows to tune the size of the
hyperelastic region near a moving crack, as indicated in subplots (c) and (d). The
local increase of elastic modulus and thus wave speeds can be tuned by changing
the slope of the large-strain stress–strain curve (“local modulus”)
This ensures that the crack achieves a steady-state during propagation through
the slab. The slab is initialized at zero temperature prior to simulation.
The length ly is several times larger than lx , with the ratio ranging from
two to five. The slab width lx considered ranges from 1,150 (smallest) up to
4,597 (largest, corresponding to micrometer length scale in physical dimen-
sions). The largest model contains over 70 million atoms. All quantities in this
section are given in reduced units. The condition for small-scale yielding is
satisfied in all cases (with harmonic, stiffening and softening potentials), since
breaking of atomic bonds occurs over a region involving only a few atoms along
the weak layer (that is, very small fracture process zone with a size on the
order of a few atomic distances). There is no dislocation processes and the
system is perfectly brittle. The slab is loaded with a maximum of a few per-
cent strain, according to the crack loading mode. The loading is significantly
lower than other studies [114]. A slit of length a is cut midway through the
slab as an initial, atomically sharp crack.
Accurate determination of crack tip velocity is important because we need
to be able to measure even smallest changes in the propagation speed. The
crack tip position is determined by finding the surface atom with maximum
y position in the interior of a search region inside the slab. This quantity is
averaged over a small time interval to eliminate very high frequency fluctua-
tions. To obtain the steady-state velocity of the crack, the measurements are
taken within a region of constant stress intensity factor [114]. In addition to
checking the velocity history, steady state is verified by path-independency of
the energy flux integral [22].
238 Atomistic Modeling of Materials Failure
Fig. 6.37 Hyperelastic region in a (a) softening and (b) stiffening system
the ratio S/γ. To achieve this objective, we keep the ratio S/γ constant in all
of the simulations. In the harmonic systems, as S ∼ E and γ ∼ E, we choose
the parameter rbreak to be identical in all cases. In the biharmonic systems,
we adjust the parameter rbreak , at given values of ron , E0 , and E1 , to always
keep S/γ constant.
We choose rbreak = 1.17 for the harmonic systems. The failure strain at
the crack tip can reach a magnitude of several percent, which is comparable
to many “real materials.” In the harmonic systems (with Young’s modulus
equal to E0 or E1 ), the crack achieves the same propagation velocity around
80% of the Rayleigh wave speed. This is consistent with the linear theory.
For the biharmonic systems, we choose ron = 1.1275 and rbreak = 1.1558
in the stiffening system and rbreak = 1.1919 in the softening system to keep
S/γ constant. In contrast to the linear theory prediction, we find that the
crack propagation velocity is about 20% larger in the stiffening system and
30% smaller in the softening system. These deviations cannot be explained
by the linear theory. The fact that we change the large-strain elasticity while
keeping the small-strain elasticity constant indicates that hyperelasticity is
affecting crack dynamics.
A measure for the direction and magnitude of energy flow in the vicinity of
the crack tip is the dynamic Poynting vector [22, 235]. The magnitude of the
dynamic Poynting vector '
P = P12 + P22 (6.93)
240 Atomistic Modeling of Materials Failure
Fig. 6.38 Hyperelastic region and enhancement of energy flow in the (a) softening
and (b) stiffening system
J-Integral Analysis
Fig. 6.40 Change of the crack speed as a function of εon . The smaller εon , the larger
is the hyperelastic region and the larger is the crack speed
It has been shown that a local hyperelastic zone around the crack tip can
have significant effect on the velocity of the crack. For a mode I tensile crack,
linear theory predicts that the energy release rate vanishes for all velocities in
excess of the Rayleigh wave speed [233], implying that a mode I crack cannot
move faster than the Rayleigh wave speed.
This prediction is indeed confirmed in systems with the harmonic potential
where crack velocity approaches the Rayleigh wave speed independent of the
slab size, provided that the applied strain is larger than 1.08% and the slab
width is sufficiently large (lx > 1,000). The systems are loaded dynamically
in this case. The strain levels are about ten times lower than in many other
studies [114].
We consider hyperelastic effects of different strengths by using a bihar-
monic potential with different onset strains governed by the parameter ron .
The parameter governs the onset strain of the hyperelastic effect
ron − r0
εon = . (6.96)
r0
The simulations reveal crack propagation at super-Rayleigh velocities in
steady state with a local stiffening zone around the crack tip.
Fig. 6.41 Shape of the hyperelastic regions for different choices of εon (the hyper-
elastic regions are symmetric with respect to the crack propagation direction). The
smaller εon , the larger is the hyperelastic region. The hyperelastic region takes a
complex butterfly shape
Figure 6.40 shows that the hyperelastic effect is very sensitive to the poten-
tial parameter and the extension of the local hyperelastic zone. Mode I cracks
can travel at steady-state intersonic velocities if there exists a locally stiffening
hyperelastic zone.
For example, when the large-strain spring constant is chosen to be k1 =
4k0 , with ron = 1.1375 and rbreak = 1.1483 (that is, “stronger” stiffening and
thus larger local wave velocity than before), the mode I crack propagates 21%
faster than the Rayleigh speed of the soft material, and becomes intersonic,
as shown by the Mach cone of shear wave front depicted in Fig. 6.42.
Fig. 6.42 Intersonic mode I crack. The plot shows a mode I crack in a strongly
stiffening material (k1 = 4k0 ) propagating faster than the shear wave speed
σ2 h
G= f (v, c0 , c1 ) (6.97)
E
6 Deformation and Dynamical Failure of Brittle Materials 245
Fig. 6.43 Supersonic mode II crack. Cracks under mode II loading can propagate
faster than all wave speeds in the material if there exists a local stiffening zone near
the crack tip
where σ is the applied stress, E the local Young’s modulus of the strip mate-
rial, h is the half width of the stiff layer, and f is a nondimensional function of
crack velocity v and wave speeds in the strip (c0 ) and the surrounding matrix
(c1 ). The dynamic Griffith energy balance requires G = 2γ, indicating that
crack propagation velocity is a function of the ratio h/χ where
γE
χ∼ (6.98)
σ2
can be defined as a characteristic length scale for local energy flux. By dimen-
sional analysis, the energy release rate of our hyperelastic stiffening material
is expected to have similar features except that Broberg’s strip width h should
be replaced by a characteristic size of the hyperelastic region rH (note that
rH could,
√ for instance, be defined as the square root of the hyperelastic area,
rH = AH ). Therefore, we introduce the concept of a characteristic length
γE
χ=β (6.99)
σ2
for local energy flux near a crack tip. The coefficient β may depend on the ratio
between hyperelastic and linear elastic properties as well as on the dynamic
246 Atomistic Modeling of Materials Failure
Fig. 6.44 The plot shows a temporal sequence of supersonic mode II crack
propagation. The field is colored according to the σxx stress component
Fig. 6.45 Geometry of the Broberg problem of a crack propagating in a thin stiff
layer embedded in soft matrix
energy balance. The characteristic energy length scale is defined such that h/χ
equals 1 when the increase in crack speed is 50% of the difference between the
shear wave speed of soft and stiff material.
We have simulated the Broberg problem and found that the mode I crack
speed approaches the local Rayleigh wave speed as soon as h/χ reaches val-
ues around 20. Numerous simulations verify that the scaling law in (6.99)
holds when γ, E, and σ are changed independently. The results are shown in
Fig. 6.46. From the simulations, we estimate numerically β ≈ 4.4 and therefore
χ ≈ 38. The potential energy field near a crack propagating at an intersonic
speed is shown in Fig. 6.47.
6 Deformation and Dynamical Failure of Brittle Materials 247
The transition from the limiting speed of the soft material to the limiting
speed of the stiff material depicted in Fig. 6.46 is reminiscent of the observa-
tions in the one-dimensional model of dynamic fracture (see Fig. 6.20a showing
the dependence of the crack speed as a function of the potential parameter
ron ).
Fig. 6.46 Calculation results of the Broberg problem. The plot shows results of
different calculations where the applied stress, elastic properties, and fracture surface
energy are independently varied. In accordance with the concept of the characteristic
energy length scale, all points fall onto the same curve and the velocity depends only
on the ratio h/χ
The existence of a characteristic length χ for local energy flux near the
crack tip has not been discussed in the literature and plays the central role in
understanding the effect of hyperelasticity. Under a particular experimental or
simulation condition, the relative importance of hyperelasticity is determined
by the ratio rH /χ. For small rH /χ, the crack dynamics is dominated by the
global linear elastic properties since much of the energy transport necessary to
sustain crack motion occurs in the linear elastic region. However, when rH /χ
approaches unity, as is the case in some of the molecular dynamics simulations,
the dynamics of the crack is dominated by local elastic properties because the
energy transport required for crack motion occurs within the hyperelastic
region. The concept of energy characteristic length χ immediately provides
an explanation how the classical barrier for transport of energy over large
distances can be undone by rapid transport near the tip.
248 Atomistic Modeling of Materials Failure
Fig. 6.47 The plot shows the potential energy field during intersonic mode I crack
propagation in the Broberg problem. Since crack motion is intersonic, there is one
Mach cone associated with the shear wave speed of the solid
6.6.6 Summary
It has been shown that local hyperelasticity can have a significant effect on the
dynamics of brittle crack speeds and have discovered a characteristic length
associated with energy transport near a crack tip. The assumption of lin-
ear elasticity fails if there is a hyperelastic zone in the vicinity of the crack
tip comparable to the energy characteristic length. Therefore, we conclude
that hyperelasticity is crucial for understanding and predicting the dynamics
of brittle fracture. The simulations prove that even if the hyperelastic zone
extends only a small area around the crack tip, there may be crucial effects
on the limiting speed and the energy flow toward the crack tip, as illustrated
in Fig. 6.40. If there is a local softening effect, we find that the limiting crack
speed is lower than in the case of harmonic solid.
The study has shown that hyperelasticity dominates the energy trans-
port process when the size of hyperelastic zone becomes comparable to the
characteristic length
χ ∼ γE/σ 2 . (6.100)
Under typical experimental conditions (that is, relatively small stresses), the
magnitude of stress may be one or two orders smaller than that under molecu-
lar dynamics simulations. In such cases, the characteristic length χ is relatively
large and the effect of hyperelasticity on effective velocity of energy transport
is relatively small. However, χ decreases with the square of the applied stress.
At about 1% of elastic strain as in the simulations, this zone is already on the
order of a few hundred atomic spacings and significant hyperelastic effects are
observed.
6 Deformation and Dynamical Failure of Brittle Materials 249
The simulations indicate that the universal function A(v/cR ) in the clas-
sical theory of dynamic fracture is no longer valid once the hyperelastic zone
size rH becomes comparable to the energy characteristic length χ. Linear elas-
tic fracture mechanics predicts that the energy release rate of a mode I crack
vanishes for all velocities in excess of the Rayleigh wave speed. However, this
is only true if rH /χ 1. A hyperelastic theory of dynamic fracture should
incorporate this ratio into the universal function so that the function should
be generalized as
A(v/cR , rH /χ). (6.101)
The local hyperelastic zone changes not only the near-tip stress field within
the hyperelastic region, but also induces a finite change in the integral of
energy flux around the crack tip.
We find that the dynamic J-integral around a super-Rayleigh mode I crack
is still path-independent but no longer vanishes in the presence of hyperelas-
ticity. Similarly, the supersonic mode II crack motion as shown in Fig. 6.44
can only be understood from the point of view of hyperelasticity. A single set
of global wave speeds is not capable of capturing all phenomena observed in
dynamic fracture.
We believe that the length scale χ, heretofore missing in the existing the-
ories of dynamic fracture, will prove to be helpful in forming a comprehensive
picture of crack dynamics. In most engineering and geological applications,
typical values of stress are much smaller than those in molecular dynamics
simulations. In such cases, the ratio rH /χ is small and effective speed of energy
transport is close to predictions by linear elastic theory. However, the effect
of hyperelasticity will be important for nanoscale materials, such as highly
strained thin films or nanostructured materials, as well as high speed impact
phenomena.
The prediction of intersonic mode I cracks (see Fig. 6.42) has also been
verified in experimental studies of dynamic fracture of rubber [250].
Fig. 6.48 Crack propagation in an LJ system. Subplot (a) shows the σxx -field and
indicates the mirror-mist-hackle transition. The crack velocity history (normalized
by the Rayleigh wave speed) is shown in subplot (b)
6.7.1 Introduction
There are several models for the instability problem proposed in the liter-
ature. Some theories assume that the stress distribution ahead of the crack
determines the onset of instability [22, 155, 210], while others are based on
energy flow in the vicinity of the crack tip [27, 248].
In the classical literature based on linear elastic fracture mechanics, the
instability was explained by the fact that the circumferential or hoop stress
σθ [22, 210] has a maximum straight ahead of the crack at low speeds, but
features two maxima in directions inclined to the crack at high crack speeds
(see Fig. 6.27). According to this criterion, the instability should occur at
speeds around 73% of Rayleigh wave speed. Other suggestions were based
on a perturbation analysis of the asymptotic stress field [251] that predicted
unstable crack motion at 65% of Rayleigh wave speed, thus at a comparable
speed as given by the Yoffe criterion. Both criteria predict that the crack
changes to another cleavage plane inclined about 60◦ to the initial crack plane.
There are two experimental and computational observations that disagree
with the Yoffe criterion. Firstly, in most experimental and computational
investigations, the instability establishes as wiggly crack path with crack
branches inclined 30◦ to the initial crack plane. This is in contrast to Yoffe’s
prediction of an angle of 60◦ relative to the initial crack plane. In addi-
tion, in many experiments [207] as well as in computer simulations [146, 155]
the crack tip instability was observed at speeds as low as 30% of the the-
oretical limiting speed thus much lower than the theoretical prediction of
73%.
In the literature it has been suggested that this lower critical speed
for the instability may be due to hyperelastic softening around the crack
tip [27, 146, 155, 248]. One attempt of explanation was a nonlinear continuum
analysis carried out by Gao [27,248] focusing on the energy transport near the
252 Atomistic Modeling of Materials Failure
moving crack. The model, for the first time, allowed quantitative estimates for
the instability speed. The main idea of the hyperelastic continuum mechanics
analysis was that once the crack speed exceeds the speed of local energy trans-
port near the crack tip (the local wave speed), the crack becomes unstable.
Due to the strong softening of many materials, the speed of energy transport
is significantly reduced in the vicinity of the crack tip. The theoretical analysis
of the critical instability speed [27] was in consistency with the value observed
in molecular dynamics simulation for cracks propagating in LJ solids.
In contrast to Gao’s analysis, Abraham and coworkers [146, 155] proposed
that due to the local softening around the crack tip, the hoop stress becomes
flattened at much lower speed than predicted by the linear elastic continuum
theories. It was also suggested that the instability could be a consequence of
lower lattice vibration frequencies in the soft region near the crack tip. It was
argued that once the crack starts to see local fluctuations of the atoms ahead
of the crack tip, the crack becomes unstable. This was assumed to occur when
the time for the crack to traverse one lattice distance becomes comparable to
the lattice vibration period.
By systematically changing the large-strain elastic properties while keep-
ing the small-strain elastic properties constant and thus tuning the strength
of the hyperelastic effect, we will show that the elasticity of large strains
governs the instability dynamics of cracks. Linear elastic materials serve as
reference systems for the studies, where we find that the instability speed
agrees well with the predicted value from Yoffe’s linear analysis [210]. Chang-
ing the strength and type of hyperelastic effect (stiffening vs. softening) allows
tuning the instability speed.
Fig. 6.50 Force vs. atomic separation for various choices of the parameters Ξ and
rbreak (these parameters are independent from each other). Whereas rbreak is used to
tune the cohesive stress in the material, Ξ is used to control the amount of softening
close to bond breaking
curve with respect to changes of rbreak , and the lower part shows the variation
in shape when Ξ is varied. For small values of Ξ (around 50), the softening
effect is quite large. For large values of Ξ (beyond 1,000), the amount of
softening close to bond breaking becomes very small, and the solid behaves
like one with snapping bonds. The parameter rbreak allows the cohesive stress
σcoh to be varied independently.
This model potential also describes the limiting cases of material behavior
corresponding to Yoffe’s model (linear elasticity with snapping bonds) and
Gao’s model (strongly nonlinear behavior near the crack tip).
Yoffe’s model predicts that the instability speed only depends on the small-
strain elasticity. Therefore, the instability speed should remain constant at
73% of the Rayleigh wave speed, regardless of the choices of the parameters
rbreak and Ξ. On the other hand, Gao’s model predicts that the instability
speed is only dependent on the cohesive stress σcoh (and thus rbreak ):
6 Deformation and Dynamical Failure of Brittle Materials 255
σcoh rbreak
vinst = ∼ (6.104)
ρ ρ
(note that ρ denotes the mass density, as defined above). According to Gao’s
model, variations in the softening parameter Ξ should not influence the crack
instability speed.
Fig. 6.51 Crack propagation in a homogeneous harmonic solid. When the crack
reaches a velocity of about 73% of Rayleigh wave speed, the crack becomes unstable
in the forward direction and starts to branch (the dotted line indicates the 60◦ plane
of maximum hoop stress)
Fig. 6.52 Comparison between hoop stresses calculated from molecular dynamics
simulation with harmonic potential and those predicted by linear elastic theory
for different reduced crack speeds v/cR . The plot clearly reveals development of
a maximum hoop stress at an inclined angle at crack speeds beyond 73% of the
Rayleigh wave speed
Fig. 6.53 The critical instability speed as a function of the parameter rbreak , for
different choices of Ξ. The results show that the instability speed varies with rbreak
and thus with the cohesive stress as suggested in Gao’s model, but the Yoffe speed
seems to provide an upper limit for the instability speed. The critical instability
speeds are normalized with respect to the local Rayleigh wave speed, accounting for
a slight stiffening effect of the moduli as shown in Fig. 4.5
Fig. 6.54 Subplot (a) schematic of stiffening materials behavior, illustrating the
ratio kratio = k1 /k0 . Subplot (b) extension of the hyperelastic stiffening region.
Despite the fact that the stiffening hyperelastic region is highly localized to the
crack tip and extends only a few atomic spacings, the crack instability speed is
larger than the Rayleigh wave speed
critical atomic separation ron , the spring constant of the harmonic potential
is changed and switched to a new “local” large-strain value.
The knowledge of ron allows for a clear definition of the extension of the
hyperelastic zone near the crack tip, as can be verified in Fig. 6.54b. Here
the F–D function is also used to smoothly cut off the potential at rbreak .
We discuss two different choices of the ratio k1 /k0 = 2 and k1 /k0 = 4. It is
observed that if there exists a hyperelastic stiffening zone, the Yoffe speed is
no longer a barrier for the instability speed and stable crack motion beyond
the Yoffe speed can be observed. This behavior is shown in Fig. 6.55. We find
that the stronger the stiffening effect, the more rapid the increase of instability
speed with increasing value of the ratio of rbreak .
Note that all points fall together once rbreak becomes comparable to ron ,
corresponding to the case when no hyperelastic zone is present and the poten-
tial is harmonic with smooth F–D bond breaking (here the potential shape
is identical because bonds rupture before onset of the hyperelastic effect). As
can be clearly seen in Fig. 6.55 for k1 /k0 = 4, the instability speed can even be
super-Rayleigh and approach intersonic speeds. This is inconsistent with the
classical, linear elastic theories but can be understood from a hyperelasticity
point of view. This observation suggests that the stiffening materials behavior
tends to have a stabilizing effect on straight crack motion.
Fig. 6.55 Molecular dynamics simulation results of instability speed for stiffening
materials behavior, showing stable super-Rayleigh crack motion as observed in recent
experiment. Such observation is in contrast to any existing theories, but can be
explained based on the hyperelastic viewpoint
the instability speed. The most important result of this section is that hyper-
elasticity, the elasticity of large strains (extending only a small domain from
the crack tip) governs the instability speed. By keeping the small-strain elas-
tic properties constant and systematically changing the large-strain elasticity,
we demonstrated that the instability speed can be tuned to higher and lower
values.
An important consequence of the results is that linear elastic theory cannot
be applied to describe the instability dynamics in nonlinear materials. Since
most real materials show nonlinearities at large strains, linear elastic theory
cannot be applied to describe crack dynamics in real materials.
The results conform that the local wave speed near the crack tip governs the
dynamics [27]. This also explains experimental [207] and other computational
results [146, 155]. We summarize the main findings.
• Cracks in purely harmonic lattices with the harmonic bond snapping
potential move straight as long as the propagation speed is below about
73% of the Rayleigh wave speed. This finding is in agreement with the
classical theory proposed by Yoffe [210] (since the hoop stress develops a
bimodal structure).
• Large-strain elastic properties and therefore the local wave speeds can dom-
inate the instability dynamics. This result was verified in Fig. 6.53 where
a correlation of the instability speed to the local wave speed is shown.
• In softening materials, the instability is reduced to much lower values than
the 73% mark of the Rayleigh wave speed.
260 Atomistic Modeling of Materials Failure
6.8.1 Introduction
v = v ∗ − v ∗ H (t − tstop ) , (6.105)
where t denotes the time and H(s) is the unit step function. The variable v ∗
stands for the constant propagation velocity which corresponds to the lim-
iting velocity of cracks in our simulations. As shown by (6.105), we study a
crack that propagates at its limiting speed up to time tstop and then sud-
denly stops. The reason for crack stopping could, for example, be that the
resistance of the material to fracture increases dramatically. This problem is
6 Deformation and Dynamical Failure of Brittle Materials 261
c s < v ≤ cl (6.107)
c R < v ≤ cs (6.108)
Fig. 6.56 Allowed velocities for mode I and mode II crack propagation, linear and
nonlinear stiffening case
(“prospective crack plane”) [234], while in other areas around the crack tip
this is achieved only in the long-time limit (after a number of elastic eaves
have been emitted) [22, 23]. Although we do not study a mode III crack in
this chapter, we would like to note that the static field is radiated out behind
a circular wave front and the region of instantaneous “switch” to the static
field is not confined to the prospective crack line [252]. First experimental
observations on nonuniform crack growth in the sub-Rayleigh regime were
published in [256].
Here we review molecular dynamics model suitable to simulate the dynam-
ical processes of a suddenly stopping crack. Explaining this phenomenon at
the atomistic scale will help forming a more complete picture of dynamic
fracture. We will review a study of suddenly stopping cracks using a combina-
tion of continuum theory, laboratory experiments, and computer simulations.
Important references for the study will be [22] and [23], where analytical and
experimental results for the mode I case are described. For the mode II case,
we will compare the findings to the analytical work in [208, 217], where the
fundamental solution for an intersonic mode II crack and the solution for a
suddenly stopping crack were derived. No laboratory experiments are available
up to date for the suddenly stopping intersonic mode II crack.
The outline of this section is as follows: For both mode I and mode II
loading conditions, linear system solutions are established by assuming the
interaction between atoms to be a central pair potential similar to a har-
monic ball-spring model. It will be shown that these simulations reproduce
the continuum mechanics solution for the plane stress case. Subsequently, we
use the linear study as a reference to probe crack dynamics in nonlinear mate-
rials characterized by an “anharmonic” tethered LJ potential [165, 219, 237].
The harmonic potential is the first-order approximation of the anharmonic
potential. It will be shown, for the first time at the atomistic scale, that the
sub-Rayleigh crack indeed behaves like a massless particle and that this fea-
ture does not hold for the intersonic case. It will also be demonstrated that
the massless feature of cracks does not hold for nonlinear materials: The crack
does not behave like a massless particle in the nonlinear case. Cracks being
strictly massless is therefore confined to sub-Rayleigh cracks.
1 − ȧ/cR
k(ȧ)I ≈ (6.110)
1 − ȧ/cl
and
1 − ȧ/cR
k(ȧ)II ≈ (6.111)
1 − ȧ/cs
for most practical purposes. The stress intensity factor responds instanta-
neously to a change in propagation velocity. In fracture mechanics, one often
writes the so-called equation of motion of a crack as [22]
where G denotes the dynamic energy release rate and Γ (ȧ) represents the
dynamic fracture toughness, a material property measuring the fracture
resistance. For mode I, one can write more precisely,
EΓ (ȧ)
≈ g(ȧ). (6.113)
(1 − ν 2 )KI (a, 0)2
In linear elastic fracture mechanics, the bulk elastic properties consist of elas-
tic constants, while the effects of loading and geometry are included in the
expression for KI (a, 0). Equations of this type can be integrated to obtain
a solution for a(t), if Γ (ȧ) and KI (a, 0) are both known. We would like to
remark that the crack propagation history a(t) could in principle be solved
using molecular dynamics simulations. In contrast, the dynamic fracture resis-
tance Γ (ȧ) cannot be determined from continuum mechanics theory and is
also difficult to be measured by experiments. The massless behavior of the
crack is also reflected by the fact that only the first derivative of the crack
tip position appears in the equation of motion (6.112). This is different from
a moving dislocation: It takes an infinite time for the static field to establish
itself around a suddenly stopping dislocation [38]. Equations (6.109)–(6.114)
are only valid for sub-Rayleigh crack growth [22,23,208,217]. In the intersonic
case, it will take some time after the crack has completely stopped before the
shear and Rayleigh waves reach the tip, as will be discussed shortly.
The analytical solution for the suddenly stopping crack can be derived
using the superposition principle: First, the solution for a crack propagat-
ing at a constant velocity v is determined. Subsequently, the solution for a
moving dislocation is superposed to negate the crack opening displacement
ahead of the crack tip where the crack has stopped. The solution for mode
I loading was derived in [234], and that for mode II sub-Rayleigh cracks was
264 Atomistic Modeling of Materials Failure
first considered in [257]. The fundamental solution for the intersonic crack
was recently determined in [208, 217]. In these papers, the Wiener–Hopf tech-
nique [22] was used to address this problem, by transforming the problem
into the complex space and then applying the theory of complex functions. A
scalar Wiener–Hopf problem is derived which can be solved by transforming
the complex functions back to real space employing the deHoop method of
integral inversion.
The crack is stopped after it reaches its limiting speed (the Rayleigh wave
speed, or higher for nonlinear simulations) and has traveled at this velocity
for some time. The value of ystop denotes the position at which the crack
stops, corresponding to the coordinate of the end of the weak layer (Fig. 6.57).
The system is loaded until time tl , after which the boundaries are no longer
displaced but held fixed. Somewhat different loading histories are chosen for
different simulations and will be indicated in the corresponding sections.
Similar to the mode I case, we also perform mode II simulations. For the mode
II simulations, the loading has to be significantly larger to achieve nucleation
of the daughter crack and the limiting speed. The large deformation around
the crack tip leads to large local dilatations soon after the crack has been
stopped. In the simulation, this could cause bonds to break as they are driven
out of the cutoff radius. Nearest neighbors are searched only within a cutoff
radius (rcut = 2 in reduced atomic units). This leads to finite values of rbreak
instead of the theoretical, continuum mechanics assumption rbreak → ∞. The
variable rbreak stands for the atomic separation when atomic bonds snap.
Breaking of atomic bonds in the bulk is avoided by increasing the potential
energy barrier for higher strains by introducing a fourth-order term in the
potential. Modifying the potential gives additional barrier for bond breaking
without affecting the rest of the slab, where the strains are much lower. Only
very localized to the crack tip, and only for a very short time after the crack
is stopped, this modification of the potential marginally affects the dynamics.
This procedure is not applied with the tethered LJ potential (see Sect. 4.4.3),
because the barrier for bond breaking has proven to be high enough due to
its natural stiffening. In addition to a slight opening displacement loading, we
impose a strong shear loading on the outermost rows, displacing the upper
border atoms to the left and the lower border atoms to the right during
loading.
We quickly note here that without the additional potential barrier, disloca-
tions would be observed to emit when the crack is stopped. This phenomenon
shows the competing mechanisms of atom separation and atom sliding in
nature [66]; the former yielding brittle fracture and the latter giving ductile
response. We deliberately avoid such effects because we wish to focus on the
crack dynamics.
Interatomic Potentials
We briefly present an analysis of the interatomic pair potentials used for the
simulations. The choice of simple interatomic force laws is consistent with
our objective to study generic properties of a many-body problem common
266 Atomistic Modeling of Materials Failure
Griffith Analysis
The question why, how, and under which conditions cracks initiate can be
investigated by comparing atomistic and continuum predictions. We assume
that the onset of crack motion is governed by the Griffith criterion. The Grif-
fith criterion predicts that the crack tip begins to propagate when the crack
tip energy release rate G reaches the fracture surface energy 2γ:
G = 2γ. (6.115)
KI2 + KII
2
G= , (6.116)
E
where KI,II are the mode I and mode II stress intensity factors. In both cases,
bonds across the weak layer breaks at rbreak = 1.1625. For the triangular
lattice and the given crack orientation, the fracture surface energy is γ harm =
0.0914 for the harmonic system. For the tethered LJ potential, the fracture
surface energy is determined to be γ LJ = 0.1186 following the same approach
described in Sect. 4.4.3.
6 Deformation and Dynamical Failure of Brittle Materials 267
This analysis follows the considerations in [165] for a mode II crack using
atomistic and continuum methods. To estimate the crack initiation time due
to the applied loading, an infinite plane stress solid containing a semi-infinite
crack is subject to far field tensile (mode I) and shear (mode II) loading.
Initially assuming a perfect solid without a crack, the background stress rate
σ̇11 is given by
2λG
σ̇11 = + 2G ε̇11 (6.117)
λ + 2G
with
νE
λ= (6.118)
(1 + ν)(1 − 2ν)
and
E
G= . (6.119)
2(1 + ν)
The background shear rate is given by
In the case of pure mode I cracks, the shear rate σ̇12 is set to zero in (6.123).
The loading strain rate for the mode I simulations is ε̇xx = 0.000 05. The
predicted crack initiation time for the mode I linear crack is tpred
init = 41.51.
Assuming small perturbations, the crack initiation time for the nonlinear mode
I crack is predicted to be tpred
init = 45.19. We may assume that the nonlinearity
268 Atomistic Modeling of Materials Failure
is localized to the crack tip, and the slab region can be described by small
perturbation elastic properties.
The values are summarized in Table 6.4. The initiation time decreases
with stiffer systems (larger linear spring constants), faster loading, and smaller
values of the fracture surface energy.
For mode II, the loading rates are ε̇xx = 0.000 015 and ε̇xy = 0.000 2. We
predict an initiation time for the crack in an harmonic solid tpred
init = 28.26. For
the nonlinear solid, we predict a slightly higher value tpred
init = 30.76 because
of the higher fracture energy. As before, we assume that the nonlinearity
is localized to the crack tip, and the slab region can be described by small
perturbation elastic properties. The values are summarized in Table 6.4.
In the following we present the results for a suddenly stopping mode I crack.
The plan is to start with the linear system, and subsequently move on to the
nonlinear system.
Harmonic Systems
The crack propagating close to the Rayleigh velocity displays a distinct signa-
ture from cracks at lower speeds. We use the maximum principal stress field to
analyze the simulation results. We find this field to be a simple and powerful
measure to be compared with continuum solutions, because it displays a sig-
nificant dependence on the propagation speed (and can therefore distinguish a
static field from a dynamic field). The stress field close to the crack tip is best
described by the asymptotic solution of continuum mechanics [22]. The field
shows only one maximum for low speeds, and exhibits another maximum for
sufficiently high velocities. The stress state ahead of the crack at high veloci-
ties is more complicated than at low velocities. The asymptotic field obtained
by atomistic simulation is shown in Fig. 6.58a for the quasistatic case (v = 0,
and low velocities), and for the case v = cR in Fig. 6.58b.
Crack initiation time is determined as tinit = 42.12, in good agreements
with the continuum theory prediction, 41.51. The loading is stopped at
6 Deformation and Dynamical Failure of Brittle Materials 269
Fig. 6.58 The asymptotic field of maximum principal stress near a moving crack tip
(a), when v = 0, (b) dynamic field for v ≈ cR , (c) dynamic field for super-Rayleigh
propagation velocities (v > cR )
tl = 72.8 by setting the strain rate to zero. The crack has a velocity v ≈ 6.55
before stopping, close to its limiting speed. The crack speed does not increase
significantly even if the loading is kept for longer time. The maximum strain
is εxx = 0.0073. The stress field, as well as numerical estimation of the crack
velocity, clearly identifies a crack propagating close to Rayleigh velocity. In
Fig. 6.59, the history of the crack length a(t) is shown.
Fig. 6.59 Crack extension history vs. time for the suddenly stopping linear mode I
crack
Once the mode I crack is stopped, two circular waves are emitted from
the crack tip. The first wave front corresponds to the longitudinal wave front,
while the second one is the shear wave front. The Rayleigh surface wave can
270 Atomistic Modeling of Materials Failure
be seen on the plane behind the crack. The static field was measured to
spread out with a velocity v ≈ 7.05, and the longitudinal wave emitted by the
stopped crack is propagating at v ≈ 12.2. Both values are, taking into account
measurement errors, reasonably close to the continuum mechanics prediction.
In the prospective crack plane, the stress field takes on its static counter-
part immediately after the shear wave has passed. Behind the crack tip, the
static field is established after the Rayleigh wave has passed. In other areas,
the static field is only reached in the long-time limit. Continuous wave emission
and rapid attenuation in regions surrounding the crack tip is observed. The
frequency of these waves increases with time. The wave period attains atomic
distance rapidly and elastic energy is dissipated as heat (thermalization). This
is visualized in Fig. 6.60.
Fig. 6.60 Maximum principal stress field for various instants in time, mode I linear
crack
It can be verified that at late stages (after the waves have attenuated), the
static field inside the shear wave front remains constant, and no additional
wave emission is identified (see lower right snapshot of Fig. 6.60). The results
confirm the experimental observations in [23]: In Fig. 6.61a, the evolution of
the maximum principal stress along the prospective crack line is shown. The
evolution of potential energy is shown in Fig. 6.61b. The first kink in the plots
refers to the longitudinal wave front and a second kink corresponds to the
shear wave front at which the static field is radiated. Additional evidence is
provided by different snapshots of the stress field after the crack has been
stopped. These results are depicted in Fig. 6.60.
6 Deformation and Dynamical Failure of Brittle Materials 271
Fig. 6.61 Evolution of principal maximum stress and potential energy along the
prospective crack line, for a linear mode I crack
The stresses ahead of the crack tip are closely related to the stress intensity
factor KI . We choose a fixed location to measure the stress over time. A similar
approach was used in experiment [23]. The result is depicted in Fig. 6.62. The
plot also shows the results of experimental studies of a suddenly stopping
mode I crack [23].
In both experiment and simulation, the stress decays slightly after the
longitudinal wave has reached the measurement location, and increase again
soon afterward. This decrease in stress is related to the arrival of longitu-
dinal wave and persists even when we change the measurement location to
272 Atomistic Modeling of Materials Failure
different y positions. The stresses continuously change until they reach the
corresponding static solution. The result of Fig. 6.62 agrees qualitatively with
experimental data, as can be verified in the plot (see also Figure 2 in [23]).
In particular, we note that the minimum at time step t∗ ≈ 3 and the smaller,
local minimum at time step t∗ ≈ 7 are qualitatively reproduced.
Fig. 6.62 Variation of stress at fixed distance ahead of the stopped linear mode
I crack. At t ≈ 0, the longitudinal wave arrives at the location where the stress is
measured. At t ≈ 8, the shear wave arrives and the stress field behind the crack tip
is static. The plot also includes the results of experimental studies [23] of a suddenly
stopping mode I crack for qualitative comparison (the time is fitted to the MD result
such that the arrival of the shear wave and the minimum at t∗ ≈ 3 match)
Anharmonic Systems
It was shown that the harmonic solid reproduces continuum mechanics solu-
tions, and may serve as a reference system when we further probe into
nonlinear material behaviors. Atomistic simulations provide an extremely
helpful tool to investigate the nonlinear case – a situation which usually cannot
6 Deformation and Dynamical Failure of Brittle Materials 273
Fig. 6.63 Experimental results of static stress field radiated in front of the crack
tip. The measurement at gage 1 is used for comparison with MD results. Reprinted
from [23] Engineering Fracture Mechanics, Vol. 15, pp. 107–114, B.Q. Vu and V.K.
Kinra, Brittle fracture of plates in tension static field radiated by a suddenly stopping
crack, copyright c 1981, with permission from Elsevier
the velocity remains at the value it has at the moment where the strain rate
is set to zero. We make a few remarks at this point:
1. The crack speed is significantly higher than in the linear case (7.5 vs. 6.55;
this is about 16% higher limiting speed than predicted by the linear theory).
2. The crack speed is also larger than the corresponding wave velocity in
the far-field. This second finding is in good agreement with other results
we have obtained with bilinear hyperelastic potentials (see Chap. 6.6), as
well as previous studies on the topic of hyperelastic brittle fracture [27,
248]. These results indicate that the local stress state at bond breaking
is important [27, 219, 248]. The higher local wave speed leads to a higher
limiting velocity. The crack can funnel energy faster than the far-field wave
speeds would allow.
3. The dynamic maximum principal stress field provides signatures of non-
linear material response. For a super-Rayleigh crack, this field is shown in
Fig. 6.58c. The stresses are higher compared to the harmonic case, and the
angular variation of the asymptotic field is different.
The histories of crack tip position and the velocity for the super-Rayleigh crack
can be found in Fig. 6.64 which plots the limiting speed calculated from the
atomistic simulations and visualizes how the crack accelerates and approaches
its limiting speed.
The history of maximum principal stress in the line ahead of the crack tip
is shown in Fig. 6.65a, and the potential energy field is shown in Fig. 6.65b.
Even in the nonlinear case, we can identify “bulk wave fronts” associated with
a localized group of nonlinear waves. The distributions of stress and energy
along y-direction are different in the linear and nonlinear cases.
In very early stages, the shear wave front propagates with vy ≈ 8.6. Later,
when the stress in y-direction is reduced at the crack tip, we measure vy ≈ 7.3.
The bulk of longitudinal waves emitted from the stopped crack is moving at
vy ≈ 15.3 in early stages, and at vy ≈ 12.3 later, approaching the linear
sound velocity. This can be attributed to the fact that the material ahead of
the stopped crack is not strained as severely in the y-direction. The propaga-
tion speed in the x-direction (orthogonal to the crack) remains higher than
that in the linear case. The longitudinal wave front orthogonal to propagat-
ing direction is moving faster than ahead of the crack at late stages. Similar
finding applies qualitatively to the shear wave front. The results show that
the wave velocity depends on the stress state, and is significantly affected by
the loading condition. The fact that the wave fronts propagate faster orthog-
onal to the propagation direction leads to elliptical wave fronts. In particular,
the local wave speeds differ significantly from the linear elastic wave speeds.
This observation is found in all of the nonlinear simulations. The discontinu-
ities of the longitudinal and shear wave front are smeared out compared to
the corresponding harmonic simulation. This observation is again consistent
6 Deformation and Dynamical Failure of Brittle Materials 275
Fig. 6.64 Crack tip history a(t) and crack tip velocity v as a function of time,
suddenly stopping mode I crack. The limiting speed according to the linear theory
is denoted by the black line (Rayleigh velocity), and the super-Rayleigh terminal
speed of the crack in the nonlinear material is given by the blueish line. When the
crack stops, the crack speed drops to zero
with the idea that there is no unique wave speeds near the crack tip. There
exists a train of “longitudinal” and “shear” waves associated with the rapidly
changing stress state near the crack tip. Consequently, the static field is not
established as soon as the crack has stopped behind the shear wave. In one of
the simulations, it takes δt ≈ 61 since stopping of the crack for the stresses
to reach a static, constant value at δy = 15 ahead of the crack tip. This time
is found to be shorter if the stresses are measured closer to the crack tip. For
example, at δy = 5 ahead of the crack tip, the time to establish the static
field is determined to be δt ≈ 30. The closer to the crack tip, the less the
time required to establish the static field. The reason could be nonlinear wave
dispersion. We emphasize that such large changes in the stress after the shear
wave has passed are not observed in the linear case. The time it takes until
the stresses do not change any more depends on the strength of the nonlin-
earity and on the amount of lateral loading. For the tethered LJ potential,
it is observed that for more compliant systems and longer loading time this
effect becomes more severe, presumably due to larger displacements and more
nonlinear dispersion. As in the harmonic case, emission of elastic waves occurs
276 Atomistic Modeling of Materials Failure
Fig. 6.65 Evolution of principal maximum stress and potential energy along the
prospective crack line; for a mode I nonlinear crack
Fig. 6.66 Maximum principal stress field for various instants in time, for mode I
nonlinear crack
simulation. One can observe the difference in shape compared to the linear
case shown in Fig. 6.62.
Fig. 6.67 Variation of stress at fixed distance ahead of the stopped nonlinear mode
I crack
278 Atomistic Modeling of Materials Failure
Discussion – Mode I
Harmonic Systems
The objective is to validate the theoretical results derived in [208, 217] using
computer experiments. In particular, we will show that the static field does
not spread out until the Rayleigh wave carrying the mother crack reaches
the stopped daughter crack. We will determine the stresses slightly ahead of
the crack tip from the molecular dynamics data and show similarity to the
continuum solution. The linear solution will serve as the reference system
when we probe into nonlinear material behaviors.
The time to crack initiation is found to be tinit = 32.04, in good agreement
to the continuum theory prediction tpred
init = 28.26. The daughter crack nucle-
ates at t = 82, this is, δt ≈ 50 later than the initiation of the mother crack.
6 Deformation and Dynamical Failure of Brittle Materials 279
Fig. 6.68 Schematic of waves emitted at a suddenly stopping mode II crack; (a)
stopping of daughter crack, (b) stopping of mother crack
Fig. 6.69 Evolution of (a) principal maximum stress and (b) potential energy along
the prospective crack line; for linear supersonic crack
Anharmonic Systems
The linear solution has reproduced results similar to the continuum mechanics
solution of the problem. We further consider the dynamics of a suddenly
stopping crack in a nonlinear material.
6 Deformation and Dynamical Failure of Brittle Materials 281
Fig. 6.70 Potential energy field for various instants in time, mode II linear crack
Fig. 6.71 Variation of stress at fixed distance ahead of the stopped intersonic mode
II crack
Fig. 6.72 Crack extension history vs. time for the supersonic mode II crack. The
dashed line is used to estimate the time when the mother crack comes to rest
Fig. 6.73 Evolution of (a) principal maximum stress and (b) potential energy along
the prospective crack line; for nonlinear supersonic crack
observed in the nonlinear mode I crack. In this simulation, the static stresses
are reached soon after the bulk of shear waves has passed. The delay required
to establish the static field after the shear waves have passed, which we have
observed in the nonlinear mode I case, is not observed in this simulation, but
284 Atomistic Modeling of Materials Failure
Fig. 6.74 Potential energy field around the crack tip for various times, suddenly
stopping mode II crack
The potential energy field around the crack tip is shown in Fig. 6.74 for
various times. The velocity of the wave fronts depends on the angle. This leads
to elliptical shapes of the wave front, as also observed in the mode I case. Like
in previous simulations, a thermalization effect is observed. In particular, after
stopping of the daughter crack, energy is found to be dissipated as heat, as
can be verified from Fig. 6.74 (right top). The normalized stresses σ ∗ vs. time
are shown in Fig. 6.75 for an anharmonic simulation.
Comparing this result to the linear case, we find similarity until the shear
wave arrives. Unlike the linear case in which stresses decrease strongly after
the mother crack arrives, this stress remains constant until the mother crack
arrives and then quickly increase to the steady state value.
To investigate the nonlinear dynamics further, we will briefly discuss a
second simulation. To obtain softer elastic properties, we alter the tethered
LJ potential and choose 0 = 1. The corresponding harmonic wave velocities
are cR = 4.8, cs = 5.2 and cl = 9. As we mention in Sect. 6.8.4, this softening in
combination with higher lateral strain allows us to study the nonlinear effect
better. The loading is kept for a longer time than in previous simulations.
All other parameters are kept constant. The daughter crack achieves a speed
of v ≈ 12 and is truly supersonic. The mother crack propagates at v ≈ 5.8,
which is super-Rayleigh. The wave fronts become very difficult to identify,
because they are smeared out more than in the previous simulation. The
shape of the discontinuities is clearly elliptical. The longitudinal wave front
6 Deformation and Dynamical Failure of Brittle Materials 285
Fig. 6.75 Normalized stresses σ ∗ vs. time, suddenly stopping supersonic mode II
crack
of the stopped daughter crack is propagating with v ≈ 11.95, and the shear
wave front is associated with a strong discontinuity propagating at v ≈ 7.80
through the solid. We remark that the velocity of the longitudinal wave front
is almost identical to the propagation velocity of the daughter crack. Once the
mother crack reaches the stopped daughter crack, there is no more significant
discontinuity. This is very different from the harmonic case where each wave
front is clearly identified by a distinct discontinuity. We find that the mother
crack, represented by a surface wave after the secondary crack is nucleated, is
smeared out in the nonlinear case. Therefore the arrival of the mother crack is
not the arrival of a singularity, but of a distributed stress concentration. The
static field is not instantaneously established. The stress intensity continuously
increases from the point where the daughter crack has been stopped until it
becomes fully steady state. A slight increase of stresses until all waves have
passed is observed, just as in the nonlinear mode I case.
This also helps to explain the difference in shape of energy and stress
distribution in the prospective crack line, when we compare the harmonic
with the anharmonic simulations. Moreover, the static field is not established
until all waves have passed.
Discussion – Mode II
2. After the mother crack has reached the stopped crack tip, the stress field
(as well as the energy field) ahead of the crack tip is static behind the shear
wave front in, and is static behind the Rayleigh surface wave behind the
crack. Both observations match the prediction by continuum mechanics.
In the nonlinear case, the definition and observation of the longitudinal wave
front is difficult. The wave fronts are not as sharp and discontinuities not as
strong, which may cause the field behind the shear wave front of the mother
crack to change continuously during some transition time until all waves have
passed.
6.8.6 Discussion
Fig. 6.76 Geometry of the simulations of cracks at bimaterial interfaces (note that
in many of the subsequent plots the crystal slab is rotated by 90 degrees)
Fig. 6.77 Crack tip history and crack velocity history for a mode I crack propagating
at an interface with Ξ = 10. Subplot (a) shows the crack tip history, and subplot
(b) shows the crack tip velocity over time. A secondary daughter crack is born
propagating at a supersonic speed with respect to the soft material layer
In the simulations, the left (upper) part of the slab is the stiff solid, while
the right part has lower Young’s modulus and is soft. We consider the case
290 Atomistic Modeling of Materials Failure
Fig. 6.78 The plot shows the stress fields σxx , σyy , and σxy for a crack at an interface
with elastic mismatch Ξ = 10, before a secondary crack is nucleated. In contrast to
the homogeneous case, the deformation field is asymmetric. The dark grey shades
corresponds to large stresses, and the lighter grey shades to small stresses
when the elastic mismatch Ξ = 10. For comparison, the elastic mismatch as
between PMMA and aluminum is about 15. Figure 6.77a shows the crack tip
history, and b shows the crack tip velocity over time. The crack nucleates at
time t ≈ 35, and quickly approaches the Rayleigh speed of the soft material
v → cr,0 ≈ 3.4. As loading is increased, the crack speed increases slightly and
becomes super-Rayleigh. We observe a large jump in the crack velocity at
t ≈ 110, when a secondary crack is nucleated
√ which quickly approaches the
Rayleigh speed of the stiff material v → Ξcr,0 ≈ 10.7517 > cl,0 ≈ 6.36. The
secondary crack is nucleated approximately at a distance ∆a = 11 ahead of the
mother crack and propagates with Mach 1.7 through the material! Nucleation
of secondary cracks under mode I loading is only found under high-strain rate
loading (ε̇xx = 0.000 05). If the strain rate is too low, the crack moves at a
super-Rayleigh speed until the solid has separated.
The mechanism of nucleation of a secondary crack is reminiscent of the
mother–daughter mechanism, a phenomenon so far only observed in cracks
under mode II loading.
The result suggests that at a bimaterial interface, mode I cracks under
very large loading can propagate with the Rayleigh speed of the stiffer mate-
rials, and cracks can reach speeds beyond the fastest wave speeds in the soft
material. This observation is surprising and has not been reported in experi-
ment so far [216]. In experimental studies of mode I cracks along interfaces,
the crack slightly exceeds the Rayleigh speed of the soft material but is never
observed to become intersonic or supersonic.
Figure 6.78 shows the stress field, and Fig. 6.79 the particle velocity field
before the secondary crack is nucleated. At the time the snapshots are taken,
6 Deformation and Dynamical Failure of Brittle Materials 291
Fig. 6.79 The plot shows the particle velocity field (a) u̇x and (b) u̇y for a crack
at an interface with elastic mismatch Ξ = 10, before a secondary crack is nucleated.
The asymmetry of the particle velocity field is apparent
Fig. 6.80 The plot shows the potential energy field for a crack at an interface with
elastic mismatch Ξ = 10. Two Mach cones in the soft solid can clearly be observed.
Also, the mother and daughter crack can be seen. In the blow-up on the right, the
mother (A) and daughter crack (B) are marked
Fig. 6.81 The plot shows the stress fields σxx , σyy , and σxy for a crack at an
interface with elastic mismatch Ξ = 10. In all stress fields, the two Mach cones in
the soft material are seen. The mother crack appears as surface wave behind the
daughter crack
secondary crack propagates supersonically through the material and the Mach
cones in the right half space (soft material) is clearly visible.
Fig. 6.82 The plot shows the particle velocity field (a) u̇x and (b) u̇y for a crack
at an interface with elastic mismatch Ξ = 10. The shock fronts in the soft solid are
obvious
wave speed of the stiff material is smaller than the longitudinal wave speed of
the soft material).
Fig. 6.83 Atomic details of nucleation of the secondary crack under tensile domi-
nated loading. The plot shows the shear stress field σxy near the crack tip. Atoms
with the energy of a free surface are drawn as larger atoms. The plot suggests that a
maximum peak of the shear stress ahead of the crack tip leads to breaking of atomic
bonds and creation of new crack surfaces. After the secondary crack is nucleated (see
snapshots (2) and (3)), it coalesces with the mother crack and moves supersonically
through the material (snapshot (4))
location of the maximum tensile stress σyy does not coincide with the location
of nucleation of the secondary crack; rather, the latter was found to correlate
with a peak in shear stress ahead of the mother crack. We therefore conclude
that there exists a peak in shear stress ahead of a mode I dominated interfacial
crack moving at the Rayleigh wave speed of the soft material and this peak
shear stress leads to subsequent nucleation of a secondary crack which breaks
the sound barrier at the soft Rayleigh-wave speed.
Fig. 6.84 Crack tip history for a mode II crack propagating at an interface with
Ξ = 3. The plot illustrates the mother–daughter–granddaughter mechanism. After a
secondary daughter crack is born travelling at the longitudinal wave speed of the soft
material, a granddaughter crack is born at the longitudinal wave speed of the stiff
material. The granddaughter crack propagates at a supersonic speed with respect
to the soft material layer
Fig. 6.86 Supersonic mode II crack motion at a bimaterial interface, stiffness ratio
Ξ = 3. Subplot (a) depicts the potential energy field of a mode II crack at a
bimaterial interface with Ξ = 3, supersonic crack motion. (A) mother crack, (B)
daughter crack, and (C) granddaughter crack. Subplot (b) shows the allowed limiting
speeds and the observed jumps in the crack speed
born at the longitudinal wave speed of the stiff material. The granddaughter
crack propagates at a supersonic speed with respect to the soft material layer.
If the loading is stopped after the granddaughter crack has nucleated, this
velocity is maintained until the whole slab is cracked. For a choice of Ξ = 2,
296 Atomistic Modeling of Materials Failure
Fig. 6.87 The plot shows the potential energy field near a shear loaded interface
crack with stiffness ratio Ξ = 3 (different shades of grey are used to indicate different
levels of stress). The plot shows a small section around the crack tip. The crack
surfaces are highlighted. In the upper left plot, the initial configuration with the
starting crack is shown. As the loading is increased, the mother crack starts to
propagate, eventually leading to secondary and tertiary cracks. Two Mach cones
in the soft solid and one Mach cone in the stiff solid can be observed in the lower
right figure, suggesting supersonic crack motion with respect to the soft material
and intersonic motion with respect to the stiff material
the qualitative behavior is the same. In Fig. 6.86a we depict the potential
energy field near a supersonic mode II crack along a bimaterial interface. We
mark the different cracks: (A) is the mother crack, (B) is the daughter crack,
and (C) refers to the granddaughter crack. Figure 6.86b shows a schematic of
the allowed limiting speeds and the observed jumps in crack speed. Figure 6.87
shows a few snapshots of the potential energy field from the initial configura-
tion until the birth of the granddaughter crack, illustrating the dynamics of
the crack propagation mechanism.
The stress fields for two different instants in time are shown in Fig. 6.88.
Figure 6.88a shows the stress field before nucleation of the daughter crack,
and Fig. 6.88b shows the stress field after nucleation of the granddaughter
crack. Figure 6.88c shows a magnified view into the crack tip region.
Intersonic cracks along bimaterials interfaces have also been observed in
the experiment [267]. Figure 6.89 depicts snapshots from such experimental
6 Deformation and Dynamical Failure of Brittle Materials 297
Fig. 6.88 The plot shows the σxx field of a mode II crack at a bimaterial interface
with Ξ = 3. Subplots (a) and (b) are consecutive time steps, and subplot (c) is a
blowup
6.9.3 Summary
The studies reported in this section show that cracks at interfaces show a
very different dynamics than cracks in homogeneous materials. At the inter-
face, the limiting crack speed is not well defined any more since the wave
velocities change discontinuously across the interface. Both mode I and mode
II cracks can propagate supersonically with respect to the wave speeds in the
soft material. We summarize the main findings.
• In mode I, it is observed that the limiting speed of cracks at bimaterial
interfaces is the Rayleigh wave speed of the stiff material. The nucleation
of a secondary daughter crack from the primary mother crack is observed.
Supersonic crack motion with respect to the soft layer is possible, and the
mother–daughter crack mechanism is reminiscent of the observations in
mode II cracks in the homogeneous case. This is a new phenomenon in
dynamic fracture not observed in experimental studies so far. It is also in
contrast to published experimental results [216]. Preliminary continuum
mechanics analysis stimulated by the atomistic simulation results provides
theoretical evidence that this dynamical phenomena is possible. The anal-
ysis revealed that the energy release rate is positive for crack motion close
to the Rayleigh speed of the stiff material [269].
• In mode II, it is observed that the limiting speed is the longitudinal wave
speed of the stiff material. Supersonic crack motion with respect to the soft
layer is possible, and then we observe a mother–daughter–granddaughter
298 Atomistic Modeling of Materials Failure
Fig. 6.89 An interfacial crack rupturing the bond between Homalite and aluminium,
experimental results. Subplot (a) shows the loading geometry, illustrating how shear
loading is induced by impact loading of the lower, stiffer material. Subplot (b)
shows the subsonic growth phase and subplots (c) and (d) display the intersonic
crack growth phase. Reprinted from Advances in Physics, Vol. 51(4), A. Rosakis,
Intersonic shear cracks and fault ruptures, pp. 1189–1257, copyright c 2002, with
permission from Taylor and Francis
Fig. 6.90 Allowed velocities for mode III crack propagation, linear and nonlinear
case
Fig. 6.91 Crack tip velocity history for a mode III crack propagating in a harmonic
lattice for two different choices of the spring constant ki . The dotted line shows
the limiting speed of the stiff reference system, and the dashed line shows the lim-
iting speed of the soft reference system. Both soft and stiff systems approach the
corresponding theoretical limiting speeds
Here we use the same geometry as shown in Fig. 6.45, with the difference that
the slab is predominantly under mode III loading. The main objective is to
compare the molecular dynamics simulation results of the curve v(h/χ) with
the theoretical prediction.
According to theory [24], the energy release rate for a crack propagating
in a stiff layer with width h is given by
2
hσxz
G= f (v, c0 , c1 ) (6.126)
µ
where f is a function only of the elastic properties of the layer and matrix
material as well as the crack propagation velocity. Using the Griffith condition
G = 2γ, (6.126) can be numerically solved for v. Therefore, the crack velocity
can be expressed as
where
γµ
χ=β 2
(6.128)
σxz
denotes the characteristic energy length scale. The characteristic energy length
scale is defined such that h/χ equals 1 when the increase in crack speed is
50% of the difference between the shear wave speed of soft and stiff material.
302 Atomistic Modeling of Materials Failure
Fig. 6.92 Mode III crack propagating in a thin elastic strip that is elastically
stiff. The potential energy field is shown while the crack propagates supersonically
through the solid. The stiff layer width is h = 50
Most importantly, the crack speed should only depend on the ratio of the
layer width h to the characteristic energy length scale χ.
According to the values of γ, µ, and the applied shear stress σxz for loading
time of tl = 135, γµ/σxz 2
≈ 1.
Figure 6.92 shows a mode III crack propagating in a thin elastic strip which
is elastically stiff. The crack propagates supersonically through the solid, and
the stiff layer width is h = 50. Figure 6.93 depicts the results of a set of
calculations to check of the scaling law for mode III dynamic fracture. The
continuous line corresponds to the analytical continuum mechanics solution,
and the data points are obtained for different simulation conditions. In the
molecular dynamics simulations, the loading σxz , the fracture surface energy
γ as well the elastic properties E are changed independently. The results
show that all velocities fall on the same curve. From comparison of molecular
dynamics results to the continuum solution, the parameter β ≈ 11 and there-
fore χ ≈ 11. When the inner layer width h approaches this length scale, the
crack speed has increased 50% of the difference between soft and stiff shear
wave speed.
For realistic experimental conditions under 0.1% shear strain and a crack
propagating within a thin steel layer, χ is on the order of millimeters. Further
details will be included in a forthcoming publication [273].
6 Deformation and Dynamical Failure of Brittle Materials 303
Fig. 6.93 Check of the scaling law of the mode III Broberg problem. The continuous
line refers to the analytical continuum mechanics solution [24] of the problem. The
parameters γ0 = 0.1029 and τ0 = 0.054
6.10.5 Discussion
The limiting speeds of mode III cracks are found to be in agreement with the
predicted velocities from the continuum mechanics analysis. With the results
discussed in this chapter, it is concluded that for all three modes of loading,
304 Atomistic Modeling of Materials Failure
Fig. 6.94 Suddenly stopping mode III crack. The static field spreads out behind
the shear wave front after the crack is brought to rest
the predicted limiting speeds agree well with the observation in atomistic
simulations.
The most important result of this chapter is that the scaling law found for
mode I cracks also holds for the mode III case. A quantitative comparison with
the theory provided good agreement. This result strongly corroborates the
concept of the energy length scale proposed earlier. The results also suggest, in
accordance with the continuum analysis of the problem, that supersonic mode
III crack motion is possible [24]. The results are also in agreement with recent
theoretical analysis of supersonic mode III crack propagation in nonlinear
stiffening materials [230]. Preliminary molecular dynamics simulations of crack
motion in a material defined by the tethered LJ potentials have also shown
supersonic mode III crack propagation.
Further results of a suddenly stopping mode III crack agree qualitatively
with the continuum mechanics prediction of a suddenly stopping mode III
crack. The static field is found to spread out behind the shear wave front
nucleated by the stopping crack. In harmonic lattices, the mode III carries no
inertia, just like the mode I crack and the sub-Rayleigh mode II crack. These
results are in agreement with continuum theory of dynamic fracture [22].
6.11.1 Introduction
Atomistic modeling fracture of silicon has been the subject of several stud-
ies using empirical force fields [114, 279]. In contrast to many metals, where
fracture and deformation can be described reasonably well using embedded
atom (EAM) potentials [31, 33, 35, 106, 109, 280], a proper description of frac-
ture in silicon has proved to be far more difficult, as many models did not
agree with experimental observations. This suggested that silicon requires a
different, more accurate treatment of the atomic interactions. There have been
several earlier attempts to describe fracture of silicon using atomistic meth-
ods (see, for example [114, 174, 184, 224]). Early attempts to model fracture
in silicon used Tersoff’s classical potential [10] (in the following referred to
as “Tersoff potential”) and similar formulations such as the Stillinger–Weber
potential [124]) or the EDIP potential [281]. Simulations carried out with
those potentials have not been able to reproduce experimentally observed
brittle fracture of silicon [222]. It has been suggested that the reason for these
discrepancies between computation and experiment is that the description of
the atomic bonding at large stretch obtained by empirical potentials deviates
significantly from the more accurate, quantum mechanical solution [222].
It is thus believed that to develop models of crack dynamics in silicon
that agree with experimental observations, the accuracy of QM for atoms
near the propagating crack tip is necessary. Baskes and coworkers used their
modified EAM formulation (MEAM) to describe crack motion in silicon [114]
and to investigate the critical load for fracture initiation [279]. Even though
this model leads to improved results compared to Tersoff-type potentials, the
MEAM formulation cannot describe bond formation and breaking of silicon
with other elements such as oxygen.
Here we review the results for an alternative approach utilizing the ReaxFF
reactive force field developed to reproduce the barriers and structures for
reactive processes from QM, but at a computational cost many orders of
magnitude smaller.
Here we use a hybrid simulation technique in which the ReaxFF reactive
potential for silicon [93, 131] is used for a modest region of a few thousand
atoms close to the crack tip while a computationally inexpensive but nonre-
active Tersoff potential [10] is used to describe the other 100,000 more distant
atoms needed to include their elastic constraints on the propagating tip. The
Tersoff potential and ReaxFF lead to similar materials behavior or equation
of state for small strains, but deviate strongly at large strains as demonstrated
previously [93]. The fact that both descriptions overlap for small strains enable
a smooth handshake between the two methods.
As reviewed earlier (in chapter 2), the ReaxFF reactive potentials [93,
131] have been developed to describe combinations of many different elements
across the periodic table, including first row elements (C, O, H, N), metals (Cu,
Al, Mg, Ni, Pt and others), and semiconductors (Si and others) [93, 128–131,
282]. Thus, the methodology described here could potentially be a valuable
tool for describing plasticity and fracture for some materials, where certain
details of the bond breaking process are necessary to model crack propagation.
6 Deformation and Dynamical Failure of Brittle Materials 307
Figure 6.95 depicts the atomistic model. A perfect crystal with an initial crack
of length a serving as the failure initiation point is considered. The thickness
of the system is one unit cell in the z-direction with periodic boundary con-
ditions, corresponding to a plane strain case (size in z-direction Lz = 5.43 Å).
308 Atomistic Modeling of Materials Failure
Fig. 6.95 Geometry used for simulating mode I fracture in silicon. The systems
contain between 13,000 and 113,015 atoms with Lx ≈ 550 Å and Ly ≈ 910 Å. The
numerical model is capable of treating up to 3,000 atoms with ReaxFF in Ωrx
This model is chosen due to computational limitations for keeping the reac-
tive region small. It is noted that this choice of geometry may impose some
constraints on the system, which may effect the possibility of dislocation nucle-
ation. The slab is strained at a strain εxx in mode I prior to simulation. The
boundaries are held fixed during the simulation so that the stress in the mate-
rial can only be relieved by crack propagation. The crack starts to nucleate
shortly after the simulation is started.
crack surfaces, similar behavior is observed: The pure Tersoff model leads to
amorphization at the crack tip, in contrast to the hybrid model that leads to
initiation of brittle fracture.
Fig. 6.96 Crack propagation with a pure Tersoff potential (subplot (a)) and the
hybrid ReaxFF-Tersoff model (subplot (b)) along the [110] direction (energy mini-
mization scheme). The darker regions are Tersoff atoms, whereas the brighter regions
are reactive atoms. The systems contain 28,000 atoms and Lx ≈ 270 Å × Ly ≈ 460 Å
Fig. 6.97 Crack dynamics along the [110] direction at finite temperature (T ≈
300 K), 10% strain applied
hybrid simulations, in particular the onset of the rough surface agree with
experimental studies of cracks propagating along the same crystallographic
planes [25].
Fig. 6.98 Crack dynamics along the [100] direction at finite temperature (T ≈
300 K, 10% strain applied). Shortly after nucleation of the primary crack two major
branches develop along [110] directions
Figure 6.98 shows crack dynamics for an initial crack oriented into the
[100] direction. Experiment shows that the crack branches into multiple (110)
surface directions for such a system [224]. Indeed, it is observed that the
crack branches off into two [110] directions – forming (110) crack surfaces –
shortly after nucleation. This result is consistent with the notion that crack
dynamics is most stable along this direction. This observation suggests that
proper description of the chemistry of bond breaking (with ReaxFF or QM)
agrees with experiment [224]. Branching into [110] directions from an initial
[100] crack is not observed with the pure Tersoff model.
This model also enables to calculate the crack speed as a function of the
applied load. Figure 6.99 plots the crack speed as a function of G/G0 (for two
crystallographic orientations). It is observed that the crack speed remains
zero for G < G0 , as it is expected. However, the crack speed discontinuously
jumps to approximately 3 km s−1 when G = G9 , approaching approximately
4.5 km s−1 for higher loads. This crack dynamics is reminiscent of the lattice-
trapping effect, which has also been observed in other studies.
Fig. 6.99b contains a comparison of the results with experimental and ear-
lier computational results, showing that the model quantitatively reproduces
experimental results of crack speeds. We reiterate that other computational
attempts, for example those using empirical potentials (Stillinger-Weber, Ter-
soff, EDIP) have failed to reproduce the experimental results qualitatively,
as they predict a slower increase of crack speed with increasing crack driving
force or fail to reproduce the brittle character of silicon.
Fig. 6.100 shows the dependence of the average crack velocity as a function
of the steady-state energy release rate, as obtained in experimental studies
[25]. This plot also confirms that the steady-state crack propagation speed
6 Deformation and Dynamical Failure of Brittle Materials 311
Fig. 6.99 Crack speed as a function of load, for the (110) system (subplot (a), and
the (111) system (subplot (b))
Fig. 6.100 Dependence of the average crack velocity in a single crystal of silicon,
as a function of the steady-state energy release rate, as obtained in experimental
studies. The fracture surface is smooth and mirror-like over the entire crack path for
the specimen fractured at the lowest G (open circle). A faceted fracture surface is
observed at higher G (triangles). At the highest G, the fracture surface is very rough
(squares). The continuous line corresponds to the continuum mechanical solution
obtained from an expression similar to the one reviewed in (6.34) [25]. Reprinted
from: T. Cramer, A. Wanner, and P. Gumbsch, Physical Review Letters, Vol. 85(4),
2000, pp. 788–791. Copyright c 2000 by the American Physical Society
Fig. 6.101 This plot shows the dynamic fracture toughness as a function of the
energy release rate (subplot (a)) and the average crack velocity (subplot (b)).
Reprinted from: T. Cramer, A. Wanner, and P. Gumbsch, Physical Review Let-
ters, Vol. 85(4), 2000, pp. 788–791. Copyright
c 2000 by the American Physical
Society
Figure 6.103 depicts the crack tip velocity history and the onset of the
crack tip instability.
Figure 6.104 depicts an analysis of the sequence of atomic events. The
first single bond rupture event is due to a local rearrangement of the atoms.
After this initial event, further crack extension does not occur even though
the load is increased. Instead of crack extension, a change in the local crystal
structure at the tip of the crack is observed. Two of the 6-membered silicon
rings transform into a 5–7 double ring combination, where the 7-membered
6 Deformation and Dynamical Failure of Brittle Materials 313
Fig. 6.102 Series of snapshots of fracture mechanics in silicon, for a crack oriented
in the (111) plane. The figure shows the dynamics of crack extension, leading to
failure of the entire crystal
Fig. 6.103 Velocity–time history of the crack dynamics shown in Fig. 6.102. Soon
after nucleation of the crack, the speeds jumps to values of approximately 2 km s−1 .
The crack instability sets in at approximately 69% of cR , the Rayleigh wave speed
314 Atomistic Modeling of Materials Failure
Fig. 6.104 Analysis of the sequence of atomic events during fracture initiation.
This series of snapshots show the formation of the 5–7 ring defect at the tip of the
crack (a blow-up of this defect structure is shown in subplot (b))
ring is closer to the tip of the crack (see Fig. 6.104b). The creation of this
crystal defect appears to be induced by the increased stresses in the vicinity
of the crack tip.
The eventual crack nucleation after the 5–7 double ring has been formed
occurs not at the primary crack tip. Instead, a small secondary microcrack
forms ahead of the primary crack (Fig. 6.104c), which eventually reunites
with the primary crack (Fig. 6.104d). Subsequently, the crack begins to prop-
agate at a speed close to 2 km s−1 . This minimum crack speed coincides with
the smallest admissible speed observed under constant load (see Fig. 6.104c)
6 Deformation and Dynamical Failure of Brittle Materials 315
Fig. 6.105 Comparison of the prediction of LEFM, the prediction of the modified
LEFM model (6.129), and molecular dynamics simulation results
Secondly, the 5–7 double ring effectively leads to an increased energy bar-
rier for crack nucleation. The apparent fracture surface energy under presence
of the 5–7 double ring is increased to γ5−7 > γ0 . After the critical load is
reached sufficient to break the 5–7 double ring, the crack propagates con-
tinuously, without formation of the 5–7 double ring, while the crack senses
a fracture resistance according to the surface energy of the perfect crystal,
γ0 . The jump of crack speed to a finite speed immediately after nucleation
can therefore be explained based on a simple consideration of the dynamical
energy balance. This behavior can be expressed in this simple model, where
the crack speed as a function of the crack driving force G is given by
v(G) 0 if G < G0,5−7 ,
= (6.129)
cR 1 − G0 /G if G ≥ G0,5−7 ,
316 Atomistic Modeling of Materials Failure
where G0 is the critical energy release rate according to the (111) fracture
surface energy γ5−7 , and G0,5−7 is the critical energy release at which fracture
actually occurs (due to the fracture surface energy γs ).
The crack speed at the critical fracture load G = G0,5−7 is
while
0 ≤ v < cR (6.133)
Fig. 6.106 Comparison of the crack dynamics under mode I (subplot (a)) and
mode II loading (subplot (b)) [26]
6.11.7 Summary
Fig. 6.107 Crack dynamics in silicon without oxygen (O2 molecules) (subplots (a)
and (c)) and with oxygen molecules present (subplots (b) and (d)). Subplots (a) and
(b) show the results for 5% applied strain, whereas subplots (c) and (d) show the
results for 10% applied strain. The darker grey regions are Tersoff atoms, whereas the
brighter regions correspond to ReaxFF atoms. The systems contain about 13,000
atoms, with Lx ≈ 160 Å × Ly ≈ 310 Å. This demonstrates the dramatic effect of
oxygen in making Si brittle
equations of state and various reactions in clusters and was not modified to
improve the agreement with experimental results.
The results show that in addition to instability driving forces such as
energy flow [28] and change of asymptotic stress field [27, 29], chemical rear-
rangements may contribute to fracture instabilities, rendering the situation
more complex for many materials, suggesting an intimate connection between
fracture mechanics and chemistry [274, 275].
Even though this analysis was focused on silicon, similar mechanisms may
occur in other covalently bonded materials. A wider implication of this study is
that properties such as the fracture surface energy are not material parameters
alone, but may depend on the physical state of the system, as exemplified by
generation of the 5–7 double ring that is formed under large stresses near the
tip of a crack.
6 Deformation and Dynamical Failure of Brittle Materials 319
Fig. 6.108 This plot shows the difference in large-strain elasticity between the
Tersoff potential and ReaxFF, while both descriptions coincide at small strain. This
result demonstrates the importance of large-strain properties close to breaking of
atomic bonds [27–29]
In Sect. 6.6, the large-strain elastic properties were changed, while keeping
the small-strain elastic properties constant to systematically investigate the
effect on crack dynamics. The main finding is that the elasticity of large strains
can dominate the dynamics of fracture, in contrast to the predictions by many
existing theories [22]. With a new concept of the characteristic energy length
scale χ in dynamic fracture the experimental and computational results can
be explained. This length scale immediately explains under which conditions
hyperelasticity is important and when it can be neglected. Cracks moving in
solids absorb and dissipate energy from the surrounding material. The new
length scale characterizes the zone near the crack tip from which the crack
draws energy to sustain its motion. When materials are under extreme stress,
this length scale extends only a few dozens nanometers. One of the important
consequences of this is that cracks can move supersonically in contrast to
existing theories. The finding that the crack speed increases continuously as
the size of the hyperelastic region expands (shown in Figs. 6.40 and 6.41)
can be explained by the interplay of the hyperelastic region size and the
characteristic energy length scale, and is in qualitative agreement with the
findings in the one-dimensional model depicted in Fig. 6.20a. Stimulated by
the results discussed in Sect. 6.6 [28], intersonic mode I cracks as shown in
Fig. 6.42 have recently been verified in the laboratory [250].
In the following section, we investigated the effect of hyperelasticity on the
stability of cracks. It is known that cracks propagate straight at low velocities,
but start to wiggle when the crack speed gets larger [155, 241]. One of the
theoretical explanations [22, 210] is that the hoop stress becomes bimodal
at a speed of 73% of Rayleigh wave speed. Indeed, we find in the atomistic
simulation that a crack in a harmonic lattice becomes unstable at a speed of
about 73% of Rayleigh wave speed, in good agreement with the continuum
theory. However, if a softening potential is used, the instability occurs at
lower speeds! In contrast, it was demonstrated that if material stiffens with
strain, the instability occurs at higher speeds than predicted by linear elastic
theory. We therefore conclude that hyperelasticity governs the dynamic crack
tip instability.
Additional studies focused on cracks at interfaces. We reviewed studies of
mode I and mode II cracks along interfaces of elastically dissimilar materi-
als. In mode I, it is observed that cracks are limited by the Rayleigh wave
speed of the stiffer of the two materials provided that sufficient loading is
applied. A mother–daughter mechanism, similar as known to exist at inter-
faces of identical materials under mode II loading [165], is observed that allows
the crack to break through the sound barrier. In mode II, we find that the
crack speed is limited by the longitudinal wave speed of the stiff material
and observe a mother–daughter–granddaughter crack mechanism. Whereas
a mother–daughter mechanism has not been observed in mode I cracks, the
mother–daughter–granddaughter mechanism has been observed in mode II
cracks along interfaces of elastically harmonic and anharmonic materials.
Most importantly, experimental evidence was reported for the existence of
6 Deformation and Dynamical Failure of Brittle Materials 323
In this section, the effect of the elasticity of large strains on the dynamics of
fracture was was one of the main points of interest. Here we discuss the role
of hyperelasticity in more general terms.
The results suggest that hyperelasticity has (1) an effect on the crack speed
as well as (2) on the instability dynamics of cracks. Unlike in some previous
324 Atomistic Modeling of Materials Failure
studies (e.g. [155]), we used the concept of the weak fracture layer to separate
the two problems of limiting speed and instability from one another to obtain
clean simulation and analysis conditions. This allowed us to investigate the
conditions under which hyperelasticity governs the dynamics of fracture.
The approach of defining model materials is a reasonable method to inves-
tigate some of the fundamentals of dynamic fracture, and may be considered
advantageous over methods where the peculiarities of a specific material are
accounted for. In some previous studies, due to the complexities of the poten-
tial it was difficult to draw general conclusions about crack dynamics in brittle
solids.
It was shown that the key to understand the dynamics of cracks in hyperelastic
materials is a new length scale that characterizes the zone near the crack tip
from which the crack draws energy to sustain its motion.
Fig. 6.109 Different length scales associated with dynamic fracture. Subplot (a)
shows the classical picture [22], and subplot (b) shows the picture with the new
concept of the characteristic energy length χ
We find that the large-strain elastic properties have a strong impact on the sta-
bility of dynamic cracks. Therefore, the dynamics of fracture is predominantly
governed by the large-strain elastic properties of the interatomic potential.
This was exemplified in a study of a harmonic vs. softening and stiffening
potentials. Whereas the crack becomes unstable at 73% of the Rayleigh wave
speed in materials with harmonic interactions, the crack becomes unstable at
speeds much smaller than the Rayleigh wave speed in softening materials. In
contrast, stiffening material behavior allowed cracks close to Rayleigh wave
326 Atomistic Modeling of Materials Failure
This chapter is dedicated to the study of the deformation and fracture behav-
ior of ductile materials. Ductile materials are characterized by their capacity
to withstand large deformation and to be able to deform permanently (see
also the comparison shown in Fig. 1.2). Plastic deformation of metals is often
described using continuum mechanics techniques, such as crystal plastic-
ity theories [285–287] or strain gradient formulations [288, 289]. Significant
research effort has also been put into the development of mesoscopic dis-
crete dislocation dynamics techniques [50, 95–101]. Yet another approach is
to study plasticity using large-scale atomistic simulations. The basic carri-
ers of plastic deformation in crystals are dislocations. Therefore, most of the
discussion in this chapter will be focused on the behavior of these basic ele-
ments of plasticity. In this chapter, we will review a continuum theoretical
and atomistic approach in treating the nucleation, propagation, and interac-
tion of dislocations. The ductile character of the material is captured by the
specific interatomic potential. As discussed earlier, for many metals, appropri-
ate EAM-type potentials have been developed. This discussion will be limited
to FCC crystals (since several well tested interatomic potentials exist for this
class of metals).
7.1 Introduction
The understanding of how materials deform and break is often limited to
phenomenological engineering theories that describe the macroscale materials
behavior, neglecting the underlying atomistic microstructure. However, defor-
mation and fracture of materials is controlled by atom-by-atom processes that
are essentially governed by quantum mechanics. These quantum mechanical
effects that control chemical bonds are neglected in most existing theories.
To include these effects, atomistic models have proven to be a promising
approach that are capable to simulate the motion of all atoms in the mate-
rial, with systems comprising up to several billion particles, thus reaching
328 Atomistic Modeling of Materials Failure
Fig. 7.1 This plot illustrates the difference between a screw dislocation (marked as
“S”) and an edge dislocation (marked as “E”). The graph depicts a crystal with a
curved dislocation line (the thick curved line). When bl, the dislocation has screw
character, and when b⊥l the dislocation has edge character
of plasticity, as discussed in Sect. 1.2.3. Figure 1.6 displays the process of nucle-
ating and propagating a dislocation through a crystal, leading to permanent
plastic deformation. The geometry of a dislocation is characterized by its
Burgers vector b, defining the amount of plastic deformation due to a single
dislocation. The direction of the dislocation is referred to as the dislocation
line, typically referred to as l.
There are two types of dislocations, edge and screw dislocations. Edge
dislocations feature a Burgers vector that is perpendicular to the line direction.
Screw dislocations feature a Burgers vector that is parallel to the line direction.
Figure 7.1 illustrates the difference between a screw dislocation (marked as
“S”) and an edge dislocation (marked as “E”).
A dislocation can move in a crystal, changing its screw or edge character
as the line direction can change. However, the Burgers vector of a dislocation
always remains the same. Two dislocations of opposite Burgers vector can
annihilate each other when they come sufficiently close.
A dislocation in a solid induces a stress field that scales as 1/r, thus showing
a different singularity order as a crack. For an edge dislocation, the stress field
under plane strain condition is given by
µb y(3x2 + y 2 )
σxx = − (7.1)
2π(1 − ν) (x2 + y 2 )2
µb y(x2 − y 2 )
σyy = (7.2)
2π(1 − ν) (x2 + y 2 )2
µb x(x2 − y 2 )
σxy = (7.3)
2π(1 − ν) (x2 + y 2 )2
σzz = ν(σxx + σyy ) (7.4)
σxz = σyz = 0, (7.5)
where µ is the shear modulus and ν is Poisson’s ratio, and x and y are
coordinates. The pressure field around an edge dislocation is
2 µb(1 + ν) y
p= (7.6)
3 2π(1 − ν) x + y 2
2
illustrating that the pressure is compressive above the slip plane (due to the
additional inserted volume) and tensile below the slip plane.
The stress field around a screw dislocation has a simpler form, involv-
ing only shear stress components. The reader is referred to the literature for
further details [38].
Since dislocations induce a distortion of the atomic lattice, a crystal that
contains a dislocation has a higher energy state than a crystal without a
dislocation. The elastic strain energy stored per unit length of dislocation is
330 Atomistic Modeling of Materials Failure
EL = Cµb2 , (7.7)
where C is a constant that depends on the structure of the core of the disloca-
tion (C typically ranges from 0.5 to 1.0), µ is the shear modulus, and b is the
magnitude of the Burgers vector. Dislocations with smaller Burgers vectors
are energetically favored over those with large Burgers vectors. This equa-
tion can also be used to determine if dislocation reactions are energetically
favorable, a method referred to as “Frank’s rule.”
Dislocations move through a crystal lattice through either glide or climb.
In dislocation glide, the Burgers vector and the glide direction lie in the same
plane, and during its motion no volume is generated or taken away. In dis-
location climb, motion of the dislocation involves generation or dissipation
of volume, which may lead to generation of vacancy defects or interstitials.
The repeated slip of dislocations in the same direction leads to shearing of
the material, resulting in macroscopic permanent plastic deformation. In dis-
location climb, the Burgers vector and the glide planes lie in different crystal
planes. Climb of dislocations is typically diffusion controlled (to ensure mass
transport to sustain vacancies or interstitials).
Fig. 7.2 Illustration of splitting of complete dislocation into two partial dislocations
in an FCC lattice
The two partial dislocations repel each other. However, the separation of
the partial dislocations is associated with an energy penalty due to creation
of a stacking fault. The competition between these two effects leads to an
equilibrium separation d of the partial dislocations which scales as
b2
d∼ . (7.9)
γsf
Fig. 7.3 Geometry to explain the Schmid law, illustrating the angles φ and λ as
well as the uniaxially applied stress σy
Fig. 7.4 A cracked body, with remotely applied tensile and shear stresses σ∞ and
τ∞ . Large resolved shear stresses on specific slip planes are the key drivers for
dislocation nucleation
Fig. 7.5 Balance of forces on a dislocation close to a crack, here illustrating the
competition between the image force pulling the dislocation back to the surface
(Fim ) and the stress induced force pushing the dislocation away from the crack tip
(Fstress )
treated as a particle, which moves in the direction of the resulting force vector
Ftot . It remains in equilibrium if the resulting force vector is zero Ftot = 0.
This idea was initially proposed by Rice and Thomson [66]. Two major
forces act on a dislocation that is nucleated at a tip of a crack: First, the
image force that drives the dislocation back into the surface. Second, a force
that results from the asymptotic stress field at the tip of the crack.
µb2 L0
Fim = − (7.11)
4π(1 − ν) d
where µ is the shear modulus, b Burgers vector, d the distance of the dislo-
cation to the surface (Fig. 7.6). The parameter L0 refers to the line length of
the dislocation (which will cancel out later). The physical origin of the image
force is the fact that adding an additional half plane of atoms close to a sur-
face is energetically expensive, and therefore, the dislocation is being pulled
back into the surface. This model can also be envisioned by considering a dis-
location with opposite sign at the “other” side of the surface, which leads to
an attraction toward the surface to annihilate itself (see Fig. 7.6). Note that
this result can be derived from considering the change in elastic energy of a
dislocation with respect to its position d, F = −∂E/∂d, or by considering
(7.7) (during annihilation b changes from a finite value to zero).
To quantify the force due to the external stress field, we consider the
Peach–Koehler equation
F = P × l, (7.12)
where
Pi = σij bj (7.13)
and li is the direction of the dislocation line.
Now we consider the second force acting on the dislocation, due to the
crack tip stress field. Considering the case shown in Fig. 7.4 or Fig. 7.5,
334 Atomistic Modeling of Materials Failure
where τ is the local stress at the crack tip, which can be written as
KI
τ (d) = √ fτ (θ), (7.15)
2πd
with KI being the stress intensity factor that relates to the overall geometry
of the specimen, noting that KI ∼ σ∞ (for mode II, KII ∼ τ∞ ). The function
fτ (θ) is the characteristic angular variation of the particular stress component
considered.
Therefore, the force acting on the dislocation due to the crack tip stress
field is
KI
Fstress (d) = √ fτ (θ)bL0 . (7.16)
2πd
The forces Fstress and Fim act in opposite directions. We assume that at the
critical moment of dislocation nucleation, these two forces must be equal so
that the dislocation starts to move to the right. Since
1
Fim ∼ (7.17)
d
and
1
Fstress ∼ √ , (7.18)
d
the dislocation will continue to move away from the crack tip once the critical
load is reached.
For a geometry as shown in Fig. 7.7a,
√
KI = σ∞ πa. (7.19)
Fig. 7.7 Geometry of a cracked specimen with a penny-shaped crack (subplot (a))
and a semi-infinite crack (subplot (b))
Alternatively, for the geometry shown in Fig. 7.7b and mode II shear loading
(τ∞ ),
H 2(1 + ν)(1 − 2ν)
KII = τ∞ . (7.23)
2 ν(1 − ν)
Then,
πb µ
τ∞ ≈ 2 , (7.24)
H 4π(1 − ν)
and
b
τ∞ ∼ . (7.25)
H
A drawback of the Rice–Thomson model is that the nucleation condition
depends on the critical dislocation distance d at the moment of nucleation,
which is a priori an unknown.
336 Atomistic Modeling of Materials Failure
Fig. 7.8 Concept of stacking fault energy, considering Peierl’s concept of periodic
shear stress variation along a slip plane, as originally proposed in [30]. Subplot (a)
depicts the geometry, subplot (b) the variation of the shear stress with the distance δ
from the crack tip (coordinate system shown in subplot (a)), and subplot (c) depicts
the variation of the elastic energy
Fig. 7.9 Balance of energies before (1) and after (2) nucleation of the dislocation,
to illustrate the concept of energy release rate
7 Deformation and Fracture of Ductile Materials 337
A few years after his original model, Rice introduced another model, referred
to as the Rice–Peierls model. This model is not based on a force balance but
rather on an energy balance [30]. Rice argued that at the critical point of
nucleation of a dislocation, the energy released due to slip equals the unstable
stacking fault energy, γus . The concept of the unstable stacking fault energy
is depicted in Fig. 7.8.
Considering Fig. 7.9, the change in potential energy due to slip of a distance
∆x (this parameter could be one Burgers vector) is given by
2
τ∞
∆W = W(2) − W(1) = − ∆xHW, (7.26)
2µ
where H is the height of the system (see Fig. 7.7b), and W is the out of plane
thickness. Further,
µ
µ = (7.27)
(1 + ν)(1 − 2ν)
for plane strain.
Equation (7.26) can be rewritten as
∆W ∆W τ2
= = − ∞ H, (7.28)
∆xW ∆(xW ) 2µ
corresponding to the energy released per unit crack area advance (note that
∆(xW ) = ∆xW ). The critical condition for dislocation nucleation is then
given when the energy released per unit crack advance equals the energy to
overcome the energy barrier for dislocation nucleation, γus (energy per unit
area):
2
τ∞
H = γus . (7.29)
2µ
Solving (7.29) for the critical remote shear stress τ∞ yields
2γus µ
τ∞,disl = . (7.30)
H
An alternative mechanism for dislocation nucleation is generation of two new
surfaces (that is, brittle fracture). The energy necessary to create two new
surfaces is 2γs and thus the critical stress is
4γs µ
τ∞,surf = . (7.31)
H
It is evident that the ratio
γus
α= (7.32)
2γs
338 Atomistic Modeling of Materials Failure
determines which of the two mechanisms sets in first, that is, if the material
is ductile (α < 1) or if it is brittle (α > 1). We note that for the general
case of mode I loading and more general slip systems, the expressions take
a somewhat more complex form. However, the basic scaling relationship of
critical energies remains identical.
Fig. 7.10 Calculation of generalized stacking fault (GSF) curves for different EAM
potentials fitted to nickel. We consider potentials by Oh and Johnson [31] (O&J),
Angelo et al. [32] (AFB), Mishin et al. [33] (M&F) as well as Voter and Chen [34,35].
Subplot (a) illustrates the calculation method of the GSF curve by sliding two parts
of the crystal along the [112] direction. Subplot (b) shows the GSF curves for the
four different potentials
Atomic measures for quantities like stress or elastic strain are well described
in the literature and it has been shown in several cases that good agree-
ment of continuum mechanics theories and atomistic simulation results can
be obtained, even in the dynamic cases as shown in this book (see, for instance,
the studies discussed in Chap. 6.5). However, no direct link between continuum
mechanics concepts of plasticity such as strain gradient theories of plasticity
has been established so far. In this section, we discuss how such coupling could
in principle be achieved (based on the discussion reported in [39]).
We assume that the deformation gradient is multiplicatively decomposed,
thus F = Fe Fp , where the lattice distortion is assumed to be contained in the
elastic part Fe , and the plastic slip is contained in Fp [290]. Such deformation
340 Atomistic Modeling of Materials Failure
where
dlk
ηk = b. (7.36)
dv
An integral formulation of (7.35) is given by
1
A= dl ⊗ b. (7.37)
∆v ⊥ in ∆v
Note that statistically stored dislocations do not contribute to A in crystal
plasticity since dislocation dipoles cancel. Curved dislocation lines can be
approximated by straight dislocation segments.
7 Deformation and Fracture of Ductile Materials 341
We will continue with a review of some of the activities and the historical
development of atomistic simulations of dislocations and dislocation interac-
tions in metals, and illustrate that progress in this field was highly coupled to
advances in computer resources.
Early studies by Hoagland et al. [293] and deCelis et al. [294] treated
only a few hundred atoms. The researchers studied the competition of duc-
tile vs. brittle behaviors of solids using quasistatic methods and investigated
how and under which conditions dislocations are generated at a crack tip.
Such microcracks can be found in virtually any real materials (referred to
as material flaws), and serve as seeds for defect generation (see also Fig. 1.7
and the associated discussion). The studies were small in size, and only a
few dislocations could be simulated. Due to the lack of dynamic response
and the system size limitations, the treatments were valid only until the first
dislocation moved a small fraction of the sample size away from the crack tip.
Computational resources rapidly developed during the 1990s (see Fig. 2.34).
Cleri et al. [9] studied the atomic-scale mechanism of crack-tip plasticity using
around 80, 000 atoms. They investigated dislocation emission from a crack tip
by extracting the atomic-level displacement and stress fields, so as to link the
molecular dynamics results to continuum mechanics descriptions of brittle vs.
ductile behavior in crack propagation [30, 66, 67]. Zhou and coworkers [295]
performed large-scale molecular dynamics simulations and carried out simu-
lations of up to 35 million atoms to study ductile failure. In these simulations,
the atoms interact with Morse pair potentials as well as more realistic EAM
potentials. They observed emission of dislocation loops from the crack front,
and found that the sequence of dislocation emission events strongly depends
on the crystallographic orientation of the crack front. They assumed that sys-
tems comprising of 3.5 million atoms are sufficient to study the early stages
of dislocation nucleation (since they observed the same feature independent
of the system size).
Later, simulations with more than 100 million atoms showed generation of
“flower-of-loop” dislocations at a moving crack tip [146]. It was observed that
the generation of dislocation loops in a rapidly propagating crack occurs above
a critical crack speed, suggesting a dynamic brittle-to-ductile transition.
Other studies focused on the creation, motion, and reaction of very few
dislocations in an FCC lattice, with the objective to understand the fun-
damental principles. Research activity was centered on atomistic details of
the dislocation core making use of the EAM method [296, 297]. Zhou and
Hoolian [298] performed molecular dynamics simulations of up to 3.5 million
atoms interacting with EAM potentials (they used up to 35 million atoms
with pair potentials). They studied the intersection of extended dislocations
in copper and observed that the intersection process begins with junction for-
mation, followed by an unzipping event and partial dislocation bowing and
cutting. These are unique studies, whose results can be immediately applied
in mesoscopic simulations. Additional research was carried out to investigate
the screw dislocation structure and interaction in a nickel FCC lattice by Qi
7 Deformation and Fracture of Ductile Materials 343
Both systems feature a small single nanocrystal with small surface cracks,
under mode I tensile loading. In the first case study, two opposing surface
cracks on opposite faces of a three-dimensional FCC solid cube are considered.
Figure 7.12a and b) shows the geometry and the crystallographic orientation
[39, 301].
With tensile loading, the emission of thousands of dislocations from two
sharp cracks is observed. The dislocations interact in a complex way, reveal-
ing three fundamental mechanisms of work-hardening in this ductile material.
These are (1) dislocation cutting processes, jog formation, and generation of
trails of point defects; (2) activation of secondary slip systems by Frank–Read
344 Atomistic Modeling of Materials Failure
Fig. 7.12 Simulation geometry, lattice orientation, and time-sequence of the work-
hardening simulation. (a) Simulation geometry and (b) lattice orientation, also
defining the directions for the x, y, z coordinate system
The slab (Fig. 7.12a) is initialized to zero temperature, and a mode I tensile
strain of 4% is imposed on the outer most columns of atoms defining the
opposing vertical y − z-faces of the slab, and kept constant throughout the
simulation. The initial purely elastic strain is relaxed into plastic strain during
the course of the simulation. Since an N V E ensemble is used, the relief of the
potential energy causes an increase of temperature in the crystal.
To “see” into the interior of the solid, only those atoms are shown that have
potential energy greater than or equal to −6.1, where the ideal bulk value is
−6.3. This approach was used very effectively in several atomistic studies using
a single crack for displaying dislocations, microcracks, and other imperfections
in crystal packing.
This filtering scheme reduces the number of atoms seen by approximately
two orders of magnitude in 3D; the visible atoms are associated with faces
of the slab and initial notch, surfaces created by crack motion, local inter-
planar separation associated with the material’s dynamic failure at the tip,
and topological defects created in the otherwise perfect crystal. Because of
periodic boundary conditions, the vertical faces are not exterior surfaces and
therefore transparent. For the analysis, the centrosymmetry technique [36] is
also used, which is particularly helpful in distinguishing stacking fault regions
and partial dislocations as well as point defects.
Upon application of loading to the system, the cracks serve as fertile sources
for dislocations. Within a few picoseconds, thousands of dislocations appear
and flow into the interior of the solid. The dislocations from the cracks glide on
two primary glide planes (111) and (111). This is in agreement with the fact
that dislocations are nucleated in the direction of largest shear stress and thus
largest Schmid factor in the K-field of the crack [38, 60]. In this early stage,
the dislocations glide through the initially perfect crystal without sensing any
obstacles. Dislocations with the same Burgers vector and line orientation repel
each other and therefore push those previously created rapidly through the
crystal [38, 60].
The nucleation of partial dislocations is observed because of the very small
stacking fault energy in the LJ model material. The Burgers vectors of the
emitted partials are predominantly [211] and [121] on the (111) glide plane,
as well as [211] and [121] on the (111) glide plane. Similar cutting mecha-
nisms are also observed for dislocations with the other two possible Burgers
vectors. Assuming a positive line direction of the dislocation in the negative z-
direction, the Burgers vectors all have a component in the negative x-direction
(which is [110)) in the upper cloud. In the lower cloud, the Burgers vectors
have the opposite sign.
346 Atomistic Modeling of Materials Failure
Dislocations on the (111) and (111) glide planes in regions very close to
the crack tip are also observed. However, these dislocations sense no strong
elastic driving force that would allow them to flow further away from the
crack tip. The dislocations on the primary glide systems (111) and (111) out
run dislocations on the (111) and (111) glide planes, and in the center of the
simulation sample, only dislocations on primary glide systems occur.
Fig. 7.13 Interaction of a dislocation line with an obstacle in the material. Subplot
(a) shows a sequence of events that illustrate how the dislocation line becomes bent
as it cannot pass through the obstacle. Subplot (b) depicts a schematic that shows
an equivalent force acting on the dislocation line
Fig. 7.15 Schematic that illustrates the formation of vacancies while jogs (visible
as kinks in the dislocation line) are forced to move through the crystal
It is known from the literature [38, 60] that, when two screw dislocations
intersect, each acquires a jog with a direction and length equal to the Burgers
vector of the other dislocation. Upon intersection, the dislocations cannot
glide conservatively since each jog has a sessile edge segment. However, if the
applied stress is sufficient large, the dislocations will glide, and the moving
jogs will leave a trail of vacancies (see schematic in Fig. 7.15), or a trail of
interstitials depending on the line orientation and the Burgers vector of the
reacting dislocations.
348 Atomistic Modeling of Materials Failure
Fig. 7.16 Schematic of different dislocation cutting processes. Subplot (a) shows
two partial dislocations cutting each other. Both dislocations leave a trail of point
defects after intersection (circles). The arrows indicate the velocity vectors of the
dislocations. Subplot (b) shows a partial dislocation (black line) cutting the stacking
fault of another partial dislocation. Dislocation number 1 leaves a trail of point
defects (circles) once it hits the stacking fault generated by dislocation number 2
Fig. 7.17 Atomistic simulation results of different types of point defects: (a) Trail
of partial point defects, (b) vacancy tube, and (c) trail of interstitials. The inlays
provide a detailed atomistic view of the defect structure
b b b b
1 2
2
1
(a) (b)
Fig. 7.18 Generation of point defects due to jogs in screw dislocations. Two rep-
resentative dislocation-cutting processes are shown, (a) leading to formation of an
interstitial, (b) leading to formation of vacancy tubes. In case the edge component
of the jog is smaller than that of a partial Burgers vector, trails of partial point
defects, characterized by generation of local lattice distortion rather than complete
rows of missing or additional atoms, are generated
Fig. 7.19 Generation of trails of point defects in early stages of the simulation.
Dislocation number 1 and number 2 leave a stacking fault plane, which is subse-
quently cut by dislocation number 3. Therefore, two trails of partial point defects
are generated resulting in bowing of dislocation number 3. Subplot (a) shows a cen-
trosymmetry analysis [36] where the stacking fault planes are drawn yellow; subplot
(b) shows an energy analysis of the same region where the stacking fault planes are
not shown. Subplot (c) shows a close-up view on the dislocation cutting process
Fig. 7.20 This plot shows the reaction of the two dislocation clouds originating
from opposing crack tips, causing the generation of numerous point defects. The
circles highlight the region of interest in which the dislocation reactions occur
Due to the crystal orientation and the Burgers vector of the dislocations,
only trails of partial point defects as well as interstitials can be generated from
the cutting processes on the primary glide systems.
7 Deformation and Fracture of Ductile Materials 351
Complete vacancy tubes [149, 305] are not generated until later stages of
the simulation when the dislocation density becomes very large and secondary
slip systems are activated. A large number of such defects are created and
appear as straight, thick lines in the plot of the potential energy of atoms.
The geometry of complete vacancy tubes is shown in Fig. 7.17b. The vacancy
tubes observed in the simulation are only several nanometers long.
The formation of some trails of interstitials is also observed. The number
of such processes is rather small since the energy to create such defects is
extremely large. A trail of interstitials is depicted in Fig. 7.17c. The excess
of vacancy concentration is also found in experiments of materials under
heavy plastic deformation [307–309]. More recently, there are also several
observations of vacancy generation in computer simulations [149, 309, 310].
In particular, the study reported in [309] shows formation of single vacancy
tube due to dislocation cutting processes.
352 Atomistic Modeling of Materials Failure
Fig. 7.22 Subplots (a) and (b) show a view from the [110]-direction (a) before
nucleation of dislocations on secondary glide systems (therefore only straight lines),
and (b) after nucleation of dislocations on secondary glide systems (which appear
as curved lines)
It is found that secondary slip systems are activated once the dislocation
density is above a critical value. Figures 7.21a–c and Fig. 7.22 illustrate the
activation of secondary slip systems.
Fig. 7.23 Subplots (a) and (b) show detailed views on the formation of sessile
Lomer–Cottrell locks, with its typical shape of a straight sessile arm connected to
two partial dislocations
dislocation to conserve the Burgers vector. These sessile segments can clearly
be observed in Fig. 7.22.
The main result is that by this cross-slip mechanism it is possible to
observe cross-slip in a situation when only partial dislocations, together with
very high stresses are present. Even in systems with only partial dislocations,
nature finds a way to relieve elastic energy into secondary slip systems! It is
noted that this mechanism of cross-slip of partial dislocations was proposed in
1959 by Fleischer [37]. This simulation confirms the existence of such mecha-
nism. Formation of sessile locks has recently also been found in deformation
of nanocrystalline materials [311].
The activation of secondary slip systems is important for the hardening
process because dislocation cutting or the formation of sessile locks generates
a large number of additional defects. Dislocations on secondary slip systems
cannot move easily in the beginning. However, soon afterward, numerous new
defects are generated. In particular, it is found that the formation of com-
plete vacancy tubes at this stage dramatically increases the concentration of
vacancies. It is noted that cross-slip mechanisms have been studied in other
molecular simulations previously [299, 312–314], and we refer the reader to
these articles for further reference.
The simulation reveals the final sessile structure of a large number of dis-
locations in the late stages of the plastic deformation. When a situation is
reached where the plastic deformation of a solid has generated such a high
dislocation density that dislocation motion is hindered by their mutual inter-
actions, one generally speaks of the work-hardening regime of deformation. In
the molecular dynamics simulation, the deformation is large enough that this
work-hardening regime is reached quickly. In this final stage, a structure com-
posed of point defects, sessile dislocations, and partial dislocations is observed.
This geometric arrangement explains the particular structure dominated by
the straight defect segments in the final stages of simulation.
Fig. 7.24 The final network from a distant view, including a blow-up to show the
details of the network [39]. The characteristic structure of the network is due to the
fact that all sessile defects (both trails of partial and complete point defects) as well
as sessile dislocations as part of the Lomer–Cottrell locks assume tetrahedral angles
and lie on the edges of Thompson’s tetrahedron. The wiggly lines in the blow-up
(see the right half of the figure) show partial dislocations, and the straight lines
correspond to sessile defects
Fig. 7.24 reveals the final network from a distant view. The blow-up shows
a more detailed energy analysis of the network. The characteristic struc-
ture of the network is due to the fact that all sessile defects (trails of point
defects as well as 16 [110]-sessile dislocations as part of the Lomer–Cottrell
locks) assume tetrahedral angles and lie on the edges of Thompson’s tetra-
hedron. This immediately explains the particular structure of the observed
sessile network.
In contrast to the initial stages of plastic deformation, where dislocation
glide occurred easily, dislocation motion is essentially stopped due to the work
hardening and plastic relaxation.
7 Deformation and Fracture of Ductile Materials 355
7.4.4 Summary
Fig. 7.25 Development of the density of different defects during the simulation
using a method of separating defects of different energies [39]
The model simulation reviewed here has revealed some of the fundamen-
tal, classical mechanisms of work hardening, [1, 38, 60] in a single computer
simulation, including: (1) dislocation cutting processes, (2) cross-slip, and (3)
formation of sessile dislocation locks. The study reviewed here exemplifies
methods to analyze ultra-large-scale simulations. Similar techniques, based on
energy filtering, geometrical analysis, and centrosymmetry parameter studies
may be helpful for other future investigations. The collective operation of
the basic hardening mechanisms apparently constricts the mobility of disloca-
tions. A large ensemble of defects self-organizes into a complex defect network
with a regular structure leading to a final defect network composed of trails
of partial point defects and complete vacancy tubes as well as some trails of
interstitials, sessile dislocations, and partial dislocations. The characteristic
structure of the final network is given by the geometrical condition that the
sessile defects appear as straight lines lying at the intersection of stacking
faults, thus along the sides of Thompson’s tetrahedron.
The results illustrate that even though the LJ potential is a simplistic
model for interatomic bonding, it is nevertheless capable of capturing most of
the predicted hardening mechanisms [1, 38, 60]. A computer simulation using
the LJ potential under extreme conditions of very large strains seems to be
able to reproduce the essential deformation mechanisms of natural crystalline
materials such as metals.
However, in molecular dynamics simulations, systems are generally under
relatively “harsh” conditions, such as high stress and strain rates. This may
7 Deformation and Fracture of Ductile Materials 357
Fig. 7.26 Simulation geometry, lattice orientation, and time-sequence of the EAM
simulation. (a) Simulation geometry and (b) lattice orientation, also defining the
directions for the x, y, z coordinate system
Fig. 7.27 Sequence of snapshots that illustrate the dislocation nucleation process
from the crack tip, for the case of an EAM potential. This snapshot depicts the
early stages of dislocation nucleation, depicting how dislocations grow from the tips
of the crack
Fig. 7.28 Closer view on the dislocation structure at a later stage in the simulation
with the EAM potential (for a system with a single crack at one surface)
Fig. 7.29 Centrosymmetry analysis that shows the activation of secondary slip
systems, at later stages in the simulation with the EAM potential
The successful integration of ReaxFF and Tersoff force fields within the frame-
work of CMDF has already been established for the description of fracture in
silicon, as discussed in Sect. 6.11. Here the validity of coupling of reactive force
fields and embedded atom EAM potentials within CMDF particularly suitable
7 Deformation and Fracture of Ductile Materials 361
Fig. 7.30 Centrosymmetry analysis of the details of the dislocation structure, for
the EAM simulation. In contrast to the LJ simulations, here complete dislocations
(that is, leading and trailing partial dislocations) are emitted at the crack tip. As
a consequence, the dislocation cutting products (partial point defects) have a finite
length. Subplot (a) depicts a sequence of two snapshots, illustrating the growth of
the dislocation network at the crack tip. Subplot (b) shows a detailed view of the
dislocation structure
Fig. 7.31 Centrosymmetry analysis of the details of the dislocation structure, illus-
trating that the point defect trails are of finite length in the case of the EAM
simulation. Subplot (a) shows the network with stacking faults, and subplot (b)
shows the network without stacking faults. The thick lines refer to the trails of point
defects
362 Atomistic Modeling of Materials Failure
for modeling metals is discussed. Specifically, the focus of the studies reviewed
here is on combining the Ni/Ni–O ReaxFF potential and Baskes-Daw’s Ni
EAM potential.
Fig. 7.32 Summary of the two loading conditions considered here, shear loading
(left) and shock loading (right). The figure also depicts which domains are handled
by EAM (dark grey) and which domains are handled with ReaxFF (light grey)
The partial dislocations are allowed to pass through a ReaxFF region with
a suitable transition region from EAM to ReaxFF. A distance of 7 Å is chosen
for the transition region width and 7 Å for the buffer region width.
The reactive region spans approximately 1,000 atoms and lies in the path
of the partials slip direction. The attempt is to see any change in behavior as
the partials pass into the reactive region through the EAM potential region.
As results in Fig. 7.33 show, the partials maintain the same separation as they
pass through the reactive region. The speed and direction of slip are also not
affected. Further simulations with varying values of the transition region width
down to 1 Å, while keeping the buffer region width fixed at 7 Å and the total
number of nickel atoms in the reactive region about the same, show no change
in the behavior of the system of two partials as it moves across the boundary
of the two force fields. This can be attributed to the close match between the
ReaxFF and EAM force fields equations of state over a large range of strain
values. This ensures that the (HReaxFF − HEAM ) term in (5.6) is small for this
system, and hence, (5.7) is still valid for this system with a small transition
7 Deformation and Fracture of Ductile Materials 363
A single nickel crystal is subjected to uniaxial shock load in the [110] direc-
tion with a piston speed of 4 km/s (schematic of the loading conditions see
Fig. 7.32, right part). The elastic shock front that moves out in the EAM
modeled region is allowed to pass through a ReaxFF region with the same
handshaking conditions as discussed earlier. The shape of the shock front
and velocity are observed as it passes through different atomistic potential
regions [40].
As seen in Fig. 7.34, the shock front maintains its shape as it passes through
the reactive region. This shows that the coupling region between the two
potentials does not spuriously reflect elastic waves or change the straight
geometry of the shock front.
The behavior of dislocations and elastic waves as they pass through the
coupling region from one atomistic potential region to another shows no
anomalies. In particular, the shock front line maintains its straight geometry,
which proves that the wave speeds for both methods are similar.
Fig. 7.34 Velocity profile in x-direction for a uniaxial shock wave in a system with
ReaxFF and EAM regions coupled [40, 41]. The shock wave front can be identified
as a vertical line of high velocity (darker color) atoms. The dotted circles contain
all atoms modeled by ReaxFF potential (the exterior of the circle contains a skin of
ghost atoms with zero weight on the forces, while the entire interior is completely
reactive). Subplots (a), (b), and (c) depict the velocity profile in the x-direction
across the sample as the shock wave passes through. There appears to be no change
in shock front profile as it encounters and passes through the reactive regions, indi-
cating a smooth handshaking between the two simulation methods without force
discontinuities
of different slips systems being activated first. We focus on two cases to show
the variety in deformation mechanisms depending upon crack orientation:
• Ni crystal with crack in 110112 orientation with tensile loading in 110
direction
• Ni crystal with crack in 111112 orientation with tensile loading in 111
direction
The simulation cell size is chosen 150 Å × 150 Å × 8 Å with free surfaces
in the x and y-directions, and periodic boundary conditions (PBC) in the z-
direction. The cracks have an elliptical shape, and are inserted in the middle of
the samples in the x − y plane, and run through the entire sample thickness in
the z-direction. The crack propagation is studied under tensile mode I loading
with and without the presence of a single oxygen molecule in the crack region
to discern changes in mechanism upon presence of oxygen molecules.
The reactive regions in the multiscale modeling framework are defined by
a radius of reactive region around each oxygen atom in the system. A region of
5 Å radius is chosen in the nickel system around an oxygen atom within which
the influence of the oxygen atom can be felt and these atoms are described
by a reactive Ni–O force field. The handshaking transition region is taken as
7 Å in width and the buffer region with ghost atoms is also 7 Å in thickness,
keeping the same values as in the test cases discussed above. The union of all
reactive regions triggered by the oxygen atoms constitutes the entire reactive
domain in the system. This method ensures that ReaxFF is applied solely to
primarily describe metal-organic chemistry, while the EAM method handles
dislocation propagation and reactions in the bulk [40, 41].
The system is first set up with no oxygen atoms inside the crack region, and
tensile loading at a constant strain rate is applied. Due to the time scales
accessible to molecular dynamics, this strain rate is rather high, approaching
1010 s−1 . In this system, all atoms are modeled by an EAM potential as the
absence of oxygen atoms triggers no reactive regions.
This system is thus a reference case for the nickel–oxygen system. As seen
in Fig. 7.35, the crack tip emits partial dislocation on one of the inclined 111
planes at the crack tip, at a strain of 0.039, and emits a second partial at the
other crack tip end at a slightly larger strain of 0.041.
Now we consider the behavior of this system under presence of an oxygen
molecule (Fig. 7.36). A single oxygen molecule is randomly placed inside the
crack and the system is allowed to equilibrate chemically over 10,000 molecular
dynamics integration steps at a time-step of 0.5 fs. During this relaxation
phase, the system includes approximately 750 atoms that are modeled by a
reactive potential. Different starting positions are used for the oxygen molecule
to investigate if there is a stochastic effect of the initial position of O2 . An NVT
7 Deformation and Fracture of Ductile Materials 367
Fig. 7.35 Partial dislocation emission from crack tip for the 110 112
crack orien-
tation (only part of the crystal close to the crack tip is shown), for the case when
no oxygen atoms present (all-atom EAM) [40,41]. Defects produced at the crack tip
are circled. Subplots (a) and (b) show emission of partials from top and bottom of
the crack tip at critical strains of 0.039 and 0.041
Fig. 7.36 Partial dislocation emission from crack tip for the 110 112
crack ori-
entation (only part of the crystal close to the crack tip is shown), under presence
of oxygen molecules in the void inclusion (same coordinate system as shown in
(Fig. 7.35) [40,41]. Subplots (a),(b) and (c),(d) depict results for two different start-
ing positions of a single O2 molecule in the crack ((a) and (b): oxygen molecule at the
tip of the crack, (c) and (d): oxygen molecule at the side of the crack face). Defects
produced at the crack tip are circled and the dotted line indicates the approximate
region of atoms modeled by ReaxFF. Subplot (a) shows the structure of the crack
tip after 10,000 equilibrium molecular dynamics integration steps, and (b) shows the
first partial initiating at a strain of 0.049. Subplot (c) depicts the resulting crack
structure after 10,000 molecular dynamics integration steps for a different starting
position of the O2 molecule (placed at the side of the crack), and (d) shows the
partial initiation for this case at a strain of 0.041. The figures show lattice defects
produced in the bulk crystal surrounding the crack as a result of oxidation at crack
surface, even before strain is applied
7 Deformation and Fracture of Ductile Materials 369
Fig. 7.37 Partial dislocation emission from crack tip for the 111 112
case, results
for the case without oxygen molecules. Subplot (a) shows brittle crack opening for
the no oxygen case at a strain of 0.05, but the results depicted in subplot (b) shows
that the crack tip emits dislocations as well at a strain of 0.051 [40, 41]
Fig. 7.38 Partial dislocation emission from crack tip for the 111 112
case, results
for the case with oxygen molecules. Subplot (a) shows the structure of the crack
tip with O2 after 10,000 equilibrium molecular dynamics integration steps. The
differently shaded region to the left of the crack represents a grain boundary (GB)
that has formed during the oxidation process. In subplot (b) at a strain of 0.03 a
void has initiated close to crack tip. The crack tip starts to open up in subplot (c)
at a strain of 0.05, indicating a brittle fracture mode [40, 41]
oxygen attaches at the bottom crack tip end in this case, and upon tensile
loading, void formation starts at a strain of 0.018 and the void joins the crack
tip at a strain of 0.023. Further opening of the crack is through opening of
the crack tip in a brittle manner at a strain of 0.049.
On the other hand, placing the O2 molecule away from the crack tip so
that the oxygen reacts away from the tip leads to defect formation which is
not in vicinity of the tip. This leads to similar crack propagation behavior
as for the no oxygen case, with brittle fracture at xx = 0.054. The crack
propagation takes place in regions away from the oxide clusters in all cases.
These studies show that that oxidation can induce generation of numerous
defects including dislocations, even under conditions when no strain is applied.
These processes can significantly change the dominating fracture mechanisms.
The local chemical reactions lead to crystal defects that can spread far away
from the tip of the crack.
Figure 7.39 summarizes the critical failure strains for different cases stud-
ied. The results indicate that the failure strain is drastically reduced under
presence of oxygen molecules, almost by a factor of two. However, the location
of where the oxygen attacks matters: If attachment occurs in the vicinity of
the crack tip (first two cases on the left), the failure strain is reduced. If attack
occurs away from the crack tip, failure strain is not reduced and is very close
to the case without any oxygen present.
7 Deformation and Fracture of Ductile Materials 371
Fig. 7.39 Summary of critical failure strains for different cases studied (a schematic
of each case is shown below the graph). The results indicate that the failure strain is
drastically reduced under presence of oxygen molecules, almost by a factor of two.
However, the location of where the oxygen attacks matters: If the attack occurs in
the vicinity of the crack tip (such as in the first two cases on the left), the failure
strain is reduced (and found to be in the range of 2–3%). If oxygen attack occurs
away from the crack tip, the failure strain is not reduced and is then very close to
the case without any oxygen present (at approximately 5%) [40, 41]
8
Deformation and Fracture Mechanics
of Geometrically Confined Materials
8.1 Introduction
The strength of materials depends on the geometry of their microstructure
[306]. For instance, it has been established that in most metals, by decreasing
the grain size, the strength of the material can be increased. Therefore, fine-
grained materials are typically stronger than coarse-grained materials. The
yield strength increases according to
ky
σY = σ0 + √ , (8.1)
d
where d is the grain size, σ0 is the yield stress of a single crystal, and ky is a
material constant. This leads to the following scaling relation
1
σY ∼ √ . (8.2)
d
This is referred to as the Hall–Petch behavior, and can be derived based
on considerations of dislocation pileups in the grains [1,306]. It is a prominent
example of a geometric confinement effect.
However, materials cannot get infinitely strong as suggested by (8.1). For
instance, at elevated temperatures [1, 306, 319, 320] deformation by creep or
other grain boundary mediated mechanisms plays an important role in mate-
rials with small grain sizes. Fine-grained materials thus tend to fail rapidly
374 Atomistic Modeling of Materials Failure
Fig. 8.1 This graph shows a proposed deformation-mechanism map for nanocrys-
talline materials obtained from molecular dynamics simulation results. The map
shows three distinct regions in which either complete extended dislocations
(Region I) or partial dislocations (Region II), or no dislocations at all (Region
III) exist during the low-temperature deformation of nanocrystalline FCC metals.
Reprinted with permission from Macmillan Publishers Ltd, Nature Materials [42] c
2004
under loading, and cannot get infinitely strong contradicting the prediction of
(8.1). Recent research results suggest that even at low temperatures, materials
with ultra-fine grain sizes behave quite differently from coarse-grained mate-
rials. For instance, in nanostructered materials, the role of grain boundaries
becomes increasingly important leading to previously unknown deformation
mechanisms. Even though it is generally accepted that grain boundaries pro-
vide sources and sinks for dislocations, its role in doing so is still not well
understood. Molecular dynamics simulations have provided valuable insight
into the deformation mechanisms of nanocrystalline materials as shown in
Fig. 8.1. This plot depicts a deformation-mechanism map for nancrystalline
FCC metals as reported in [42].
Experimental techniques are capable of studying the details of the struc-
ture of grain boundaries, reaching a resolution down to individual atoms.
Figure 8.2 depicts high resolution TEM images of grain boundaries in elec-
trodeposited nanocrystalline Ni (a), and nanocrystalline Cu (b). The capa-
bility to analyze the structure of materials at nanoscale, combined with
the analysis of mechanical properties by tensile tests or nanoindentation
enables one to make structure–property links for nanoscale materials
features.
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 375
Due to the increased volume fraction of grain boundary regions, the behav-
ior of grain boundaries is critically important in nanocrystalline materials. One
of the reasons for the increasing importance of grain boundaries is that clas-
sical mechanisms of dislocation generation (e.g., Frank–Read-sources) cannot
operate in nanocrystals, because they would not fit within the grain. In addi-
tion, defects such as grain boundaries interact in complicated ways with other
defects like dislocations. An important consequence of this is that despite the
prediction by (8.1), the strength of nanomaterials does not increase continu-
ously with decreasing grain size. Below a critical grain size, experiments have
shown that the strength decreases again [321]. This is referred to as the inverse
Hall–Petch effect [69, 306, 321]. In this regime, it was proposed that the yield
stress scales as √
σY ∼ d, (8.3)
although physical foundation of such material behavior is yet to be explored
[303]. Such behavior indicates that there may exist a maximum of strength
for a certain grain size, described as “the strongest size” by Sidney Yip in
1998 [322]. One of the major objectives of recent research is to quantify this
376 Atomistic Modeling of Materials Failure
Fig. 8.3 Normalized critical stress as a function of grain size, d. The plot illustrates
the variation of the flow stress (normalized by the value of the critical flow stress
at the maximum strength at the transition grain size). The transition from the
hardening regime to the softening regime is associated with an increased role of
grain boundary mechanisms (the dashed line was added to the plot to guide the
eye). Reprinted from Acta Materialia, Vol. 51, K.S. Kumar, H. Van Swygenhoven and
S. Suresh, Mechanical behavior of nanocrystalline metals and alloys, pp. 5743–5774,
copyright c 2003, with permission from Elsevier
Fig. 8.6 Study of size-scale effects in inorganic materials by using a focused ion
beam (FIB) microscope machining technique, combined with nanoidentation exper-
iments. Ultra-small pillars of different sizes were created using FIB and subsequently
deformed plastically under compression from the top surface. Subplot (a) shows a
SEM image of the microsample after testing. The dislocation slip lines are clearly
visible at the surface. Subplot (b) shows the dependence of the yield strength on
the inverse of the square root of the sample diameter for Ni3 Al-Ta. The linear fit to
the data predicts a transition from bulk to size limited behavior at approximately
42 µm. The parameter σys denotes the stress for breakaway flow. Reprinted from
Science, M.D. Uchic, D.M. Dimiduk, J.N. Florando and W.D. Nix, Vol. 305(9), pp.
987–989, Sample Dimensions Influence Strength and Crystal Plasticity, copyright c
2004, with permission from AAAS
Fig. 8.7 Polycrystalline thin film geometry. A thin polycrystalline copper film is
bond to a substrate (e.g., silicon). The grain boundaries are typically predominantly
orthogonal to the film surface
Fig. 8.8 Experimental results of the thin film strength dependence at room tem-
perature, here for copper thin films plotted as the inverse of the film thickness. With
decreasing film thickness, flow stress initially rises, but then exhibits a plateau at
approximately 630 MPa for films 400 nm and thinner. Each data point is an average
of flow stresses from several thermal cycles, with a scatter of less than 5% in each
case. Reprinted from [51] Acta Materialia, Vol. 51, T.J. Balk, G. Dehm and E. Arzt,
Parallel glide: Unexpected dislocation motion parallel to the substrate in ultrathin
copper films, copyright pp. 4471–4485, c 2003, with permission from Elsevier
films often results from a lack of active dislocation sources rather than from
the energetic effort associated with dislocation motion.
Discrete dislocation dynamics simulations have directly confirmed the
increase of the strength of thin films [349]. Figure 8.9 depicts the result of
this analysis. It was found that for most cases, the thin film strength increase
does not obey a simple scaling of power-law type, depending on the particular
grain size. The strengthening exponent ranges from 0.5 (Hall–Petch behav-
ior) to values greater than 1 (Freund–Nix behavior of thin films). Figure 8.10
shows the distribution of dislocations for different film thicknesses.
In copper films, the regime where plastic relaxation is limited by disloca-
tion nucleation and carried by glide of threading dislocations reaches down to
film thicknesses of about hf ≈ 400 nm [51]. For yet thinner films experiments
reveal a film-thickness-independent flow stress [51, 346, 350], as also shown in
384 Atomistic Modeling of Materials Failure
Fig. 8.9 Discrete dislocation dynamics simulation analysis of the strength depen-
dence of a thin metal film. Subplot (a) shows the discrete dislocation model of the
polycrystalline film on a semi-infinite elastic substrate. Subplot (b) shows the aver-
age film strength at T = 400 K vs. the film thickness hf . Reprinted from Thin Solid
Films, Vol. 479(1–2), L. Nicola, E. Van der Giessen and A. Needleman, Size effects
in polycrystalline thin films analyzed by discrete dislocation plasticity, copyright
c
2005, with permission from Elsevier
Fig. 8.11 Change of maximum shear stress due to formation of the diffusion wedge.
In the case of no traction relaxation along the grain boundary, the largest shear
stress occur on inclined glide planes relative to the free surface. When tractions are
relaxed, the largest shear stresses occur on glide planes parallel to the film surface
Fig. 8.12 TEM micrographs of an unpassivated copper film showing parallel glide
dislocations (subplot (a), Cu film with film thickness hf = 200 nm), and a passivated
copper film showing threading dislocations (subplot (b), self-passivated Cu-1%Al
film, film thickness hf = 200 nm). Reprinted from [51] Acta Materialia, Vol. 51, T.J.
Balk, G. Dehm and E. Arzt, Parallel glide: unexpected dislocation motion parallel
to the substrate in ultrathin copper films, pp. 4471–4485, copyright c 2003, with
permission from Elsevier
traction along the grain boundary diffusion wedge becomes fully relaxed and
crack-like on the scale of a characteristic time τ . The timescale at that diffusion
takes place is usually much larger than that of dislocation glide. However,
in the nanoscaled structures investigated here, the model can explain the
observed deformation rates even at room temperature, because the timescale
of diffusional creep is inversely proportional to the cube of the characteristic
structural length (similarly to Coble creep, see (8.4)).
In the continuum model [46], diffusion is modeled as dislocation climb in
the grain boundary. The solution for a single edge dislocation near a sur-
face is used as the Green’s function to construct a solution with infinitesimal
Volterra edge dislocations [38, 352, 353]. The basis for the continuum model-
ing is the solution for the normal traction σxx along the grain boundary due
to insertion of a single dislocation (material layer of thickness b) along (0, ζ)
(corresponding to a climb edge dislocation). The coordinate system is given
in Fig. 8.17.
faster than the first eigenmode, so that the diffusion process is dominated by
the first eigenmode.
The continuum mechanics model was further advanced to capture the
effect of surface diffusion [48]. No difference in the qualitative behavior was
found, and stress decay in the film is still exponential with a characteristic
time proportional to the cube of the film thickness. Further details could be
found in [48].
In all cases considered in the literature [46, 48], with the proper definition
of the characteristic time τ , stress decay could be described by an exponential
law of the form
t
σgb (t) = σ0 exp −λ0 (8.6)
τ
with a geometry-dependent constant
Note that d characterizes the grain size and σ0 stands for the laterally applied
stress as discussed above. Equation (8.7) is an empirical formula and is valid
for 0.2 ≤ hf /d ≤ 10.
Fig. 8.15 Traction along the grain boundary for various instants in time [46]
Figures 8.14–8.16 show several numerical examples. Figure 8.14 shows the
opening displacement along the grain boundary for several instants in time.
Figure 8.15 shows the traction along the grain boundary for various instants
in time. These examples show that in the long time limit t → ∞, the solution
approaches the displacement of a crack.
390 Atomistic Modeling of Materials Failure
Fig. 8.16 Stress intensity factor normalized by the corresponding value of a crack
over the reduced time t∗ = t/τ for identical elastic properties of substrate and film
material (isotropic case), rigid substrate (copper film and rigid substrate), and soft
substrate (aluminum film and epoxy substrate) [46–48]
Figure 8.16 shows the stress intensity factor normalized by the corre-
sponding value of a crack over the reduced time t∗ = t/τ for identical elastic
properties of substrate and film material (isotropic case), rigid substrate (cop-
per film and rigid substrate), and soft substrate (aluminum film and epoxy
substrate). The results indicate that in a film on a stiff substrate, the stress
intensity factor of a crack is reached faster compared to the homogeneous
case. Similarly, the stress intensity factor of a crack is reached slower in the
case of a film on a soft substrate compared to the homogeneous case. Table 8.1
summarizes the material parameters used for the calculation.
Mass transport from the surface into the grain boundary toward the substrate
is modeled as climb of edge dislocations [46]. At nanoscale, the fact that
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 391
Fig. 8.17 Geometry and coordinate system of the continuum mechanics model of
constrained diffusional creep
Fig. 8.18 Image stress on a single edge dislocation in nanoscale thin film constrained
by a rigid substrate
The image stress on the dislocation for different film thicknesses is shown
in Fig. 8.18. The thinner the films, the stronger gets the effect of the geometric
confinement.
Between the film surface and the film–substrate interface, the image force
is found to attain a minimum value at ζEQ ≈ 0.4hf . Therefore, from the
energetic point of view, a minimum critical stress is required to allow even a
single climb edge dislocation to exist in the grain boundary. The thicker the
film, the smaller the critical stress.
This analysis suggests that consideration of single, discrete dislocations
can become very important for the nanoscale thin films. The requirement that
an edge dislocation in the film is in a stable configuration could be regarded
as a necessary condition for constrained grain boundary diffusion to initiate
and proceed. If more than one dislocations are stored in the grain boundary,
even stronger image forces are expected since different dislocations repel each
other.
Fig. 8.19 Critical stress as a function of film thickness for stability of one, two, and
three dislocations in a thin film. The critical stress for the stability of one dislocation
(continuous line) is taken from the analysis shown in Fig. 8.18. The curves for more
dislocations (dashed lines) in the grain boundary are estimates
value. For two dislocations to be stable inside the film, the critical stress is
even higher. Consequently, the stress relief due to insertion of dislocations
will also be quantized. Most importantly, as the film thickness increases the
critical stresses for stable dislocations in the film get smaller and smaller,
eventually approaching the limit when the role of single dislocations can be
neglected and the quantization is negligible. Similar observations have been
made in discrete dislocation modeling of constrained diffusional creep in thin
films [50].
Fig. 8.20 Rice–Thomson model for nucleation of parallel glide dislocations. Subplot
(a) shows the force balance in case of a crack and subplot (b) depicts the force
balance in case of a diffusion wedge
of the free surface (image dislocation), and Fstep due to creation of a surface
step (in the following, we assume Fstep Fimage ).
Close to a diffusion wedge, Fstep = 0 since no surface step is involved
and a dipole must be created in order to nucleate a parallel glide dislocation
from the wedge. This leads to a dipole interaction force Fdipole . The dipole
consists of a pair of dislocations of opposite signs, one pinned at the source
and the other trying to emerge and escape from the source. The pinned end
of the dipole has the opposite sign to the climb dislocations in the diffusion
wedge and can be annihilated via further climb within the grain boundary.
The annihilation breaks the dipole free and eliminates the dipole interaction
force so that the emergent end of the dipole moves away to complete the
nucleation process. Therefore, it seems that there could be two possible sce-
narios for dislocation nucleation at a diffusion wedge. In the first scenario, the
nucleation condition is controlled by a critical stress required to overcome the
dipole interaction force. In the second scenario, the nucleation criterion is con-
trolled by the kinetics of climb annihilation within the grain boundary which
breaks the dipole interaction by removing its pinned end and setting the other
end free. No matter which scenario controls the nucleation process, the climb
annihilation of edge dislocations in the grain boundary must be completed
and will be the rate limiting process. The force balance on the dislocation is
illustrated in Fig. 8.20b for two different, subsequent instants in time.
Fig. 8.21 Dislocation model for critical stress intensity factor for nucleation of
parallel glide dislocations
PG E(2πbx )s
Kcr = . (8.12)
8π(1 − ν 2 )
In comparison, a balance of critical stress required to break the dipole
interaction in front of a diffusion wedge yields a similar nucleation criterion
PG E(2πbx )s
Kdw = . (8.13)
4π(1 − ν 2 )
For copper with E = 150 GPa, s = 0.31, and ν = 0.33 the predicted values
PG
are Kcr ≈ 12.5 MPa ms and Kdw PG
≈ 25 MPa ms , and we note a factor of 2
PG PG
difference in critical K-values, Kdw /Kcr = 2, for dislocation nucleation at a
diffusion wedge and at a crack tip.
The study of single edge dislocation showed that in film thicknesses of sev-
eral nanometers, image stresses on climb edge dislocations can be as large as
1 GPa. This further supports the hypothesis that single dislocations become
important in small dimensions and that the discrete viewpoint of dislocation
climb needs to be adapted. It also supports the view of a critical stress for
diffusion initiation described above.
A criterion in the spirit of the Rice–Thomson model was proposed to
describe the conditions under which parallel glide dislocations are nucleated
from diffusion wedges and cracks. The most important prediction of this model
is that the critical stress intensity factor for parallel glide dislocation nucle-
ation from a diffusion wedge is twice as large compared to the case of a crack.
stresses in the film are relaxed by mass diffusion from the surface into the
grain boundary. This leads to formation of a novel material defect referred to
as the diffusion wedge. A crack-like stress field is found to develop around
the diffusion wedge as the traction along the grain boundary is relaxed and
the adhesion between the film and the substrate prohibits strain relaxation
close to the interface. The diffusion wedge causes nucleation of dislocations
on slip planes parallel to the plane of the film.
It is found that nucleation of such parallel glide dislocations from a diffu-
sion wedge can be described by a critical stress intensity factor similar to the
case of a crack.
Atomistic modeling of thin film mechanics becomes feasible with the advent
of massively parallel computers on timescales and lengthscales comparable
with those usually attained in experimental investigations. Due to the time
limitation of the classical molecular dynamics method (time intervals are typ-
ically <10−8 s), simulations must be performed at elevated temperatures to
accelerate the dynamics of grain boundary diffusion.
To make the diffusive processes accessible to the molecular dynamics
timescale, the simulations are carried out at temperatures between 80% and
90% of the melting temperature. However, it is found that the phenomenon
of grain boundary diffusion wedge and the associated dislocation mechanisms
persist at very high temperatures. This makes it possible to simulate this
specific phenomenon. Generally it is still very difficult to simulate diffusion-
related phenomena by molecular dynamics, and advanced molecular dynamics
methods (e.g., TAD) are better suited.
At elevated temperatures, grain boundary diffusion in a bulk material was
successfully modeled recently [68, 69, 303], where grain sizes up to 15 nm were
considered in a model system of Pd. Recent work [68,355,356] suggests that at
398 Atomistic Modeling of Materials Failure
In this section, we discuss the change of the displacement field as the diffusion
wedge builds up and show that the displacement field becomes crack-like.
Further, we show the diffusive displacement of atoms and hence prove that
diffusional mass transport from the surface along the grain boundary leads to
formation of a diffusion wedge.
Fig. 8.24 Change of displacements in the vicinity of the diffusion wedge over time.
The continuous dark line corresponds to the continuum mechanical solution
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 401
The snapshots in Fig. 8.24 show how the displacement changes as mate-
rial diffuses into the grain boundary. The horizontal coordinates have been
stretched by a factor of 10 in the x-direction to make the crystal lines clearly
visible
Fig. 8.25 Diffusional flow of material into the grain boundary. Atoms that diffused
into the grain boundary are highlighted
The atoms transported along the grain boundary add to either one of the
two grains. This result illustrates that the continuum mechanics assumptions
[46–48] are valid also on the atomistic level.
It is observed that “classical” threading dislocations which become oper-
ative when stresses in the film are high enough to allow nucleation of
dislocations [68,154]. A frequently observed phenomenon is the emission of dis-
locations from the grain boundary on inclined < 111 > glide planes [154,345].
In the sample with hf ≈ 30 nm, such threading dislocations are nucleated at a
stress level of σxx ≈ 2.4 GPa. We observe that thinner films require a higher
critical stress for dislocation nucleation from the grain boundary, in qualita-
tive agreement with the prediction by the 1/hf scaling law for the yield stress.
In films thinner than 10 nm it requires extremely high stresses to nucleate
inclined dislocations, which renders this mechanism almost impossible. The
studies show that grain boundaries are, as proposed in the literature, fertile
sources for dislocations in small-grained materials [68, 154].
Another issue in terms of dislocation nucleation is the stability and mobil-
ity of the grain boundary. It is observed that the grain boundary forms jogs
at elevated temperatures and relatively low stresses (contrary to intuition, at
high stresses the grain boundary remains straight). The diffusion path can be
severely suppressed and the local stress concentration at the kink serves as a
ready source for dislocations. The grain boundary does not remain straight
and oscillates around the initial, straight position.
Continuum theory assumes that dislocations are nucleated when the stress
field around the diffusion wedge becomes crack-like. Critical stress intensity
factors for dislocation nucleation measured from the atomistic simulations
are shown in Table 8.2 for different simulations. We use (8.10) to determine
the stress intensity factor. The stress intensity factor is found independent of
geometry (film thickness) and also has similar values at Th = 0.8 and Th = 0.9.
Fig. 8.26 Nucleation of parallel glide dislocations from a diffusion wedge, showing
the dynamical sequence of the process (from top to bottom). The arrows indicate the
position of the partial dislocation nucleated from the diffusion wedge, illustrating
how the dislocation moves away from the source
is therefore completely inside the film material). The core width of the partials
extends to about six Burgers vectors around 1.6 nm. The dislocation moves a
small distance away from the grain boundary to its equilibrium position. When
stresses in the film become larger, it responds by moving further away from
the grain boundary. The nucleation process is highly repeatable. Every time
one parallel glide dislocation is nucleated, one climb edge dislocation is anni-
hilated, leading to a decay in stress intensity. The additional time required to
nucleate another parallel glide dislocation is determined by the time required
for diffusion to recover the critical stress intensity. This time is much less than
the initial time required to form the diffusion wedge. After the first disloca-
tion is nucleated, more and more parallel glide dislocations are observed. In
the confined, finite simulation geometry, the emitted parallel glide dislocations
form a “secondary pileup” close to the boundary of the simulation cell. In sim-
ulations at lower temperatures (T ≈ 800 K), we also observe constrained grain
boundary diffusion and the formation of a diffusion wedge with a lattice dis-
placement field similar to that of a crack. However, due to the time constraints
of molecular dynamics, nucleation of parallel glide dislocation is not observed.
The nucleation of parallel glide dislocations from a crack in a bimaterial
layer is shown in Fig. 8.27. For numerical reasons, the loading rate is chosen
higher than in the previous case and the temperature in the simulations is
about 300 K.
After an incipient dislocation is formed, a dislocation nucleates and moves
away from the crack tip. The crack tip is blunted, and each time a paral-
lel glide dislocation is nucleated, one surface step is formed. This process is
also highly repeatable, as lateral strain is increased. The nucleation of parallel
glide dislocations from a crack tip is observed at loading rate a few orders of
magnitude higher than in the case of a diffusion wedge and there seems to be
no rate limitation in the case of a crack. As in the case of a diffusion wedge,
the dislocation glides on a parallel glide plane a few Burgers vectors above the
film–substrate interface. For a crack, nucleation of parallel glide dislocations is
observed in films as thin as 5 nm. This may be because the critical stress inten-
sity factor is smaller than that for a diffusion wedge as indicated in Table 8.2.
The critical stress intensity for parallel glide dislocation nucleation from a
diffusion wedge is about 2.3 times larger than that for a crack. This value is
in good agreement with the estimate based on the Rice–Thomson model.
8.3.4 Discussion
Fig. 8.27 Nucleation of parallel glide dislocations from a crack, showing the dynam-
ical sequence of the process (from top to bottom). The arrows indicate the position
of the partial dislocation nucleated from the crack, illustrating how the dislocation
moves away from the source. Upon nucleation, a surface step is formed due to crack
blunting
and that such transport leads to a crack-like stress field causing nucleation
of a novel dislocation mechanism of parallel glide dislocations near the film–
substrate interface. The atomistic simulations of parallel glide dislocations
being emitted near the root of the grain boundary have further clarified the
mechanism of constrained grain boundary diffusion in thin films and provided
an important link between theory and experiments.
The experimental data suggests that nucleation occurs only at specific grain
boundaries. This can partially be explained by the strong dependence of dif-
fusion coefficients on the structure of the grain boundary [68, 356]. Using
different types of grain boundaries, it was verified that high-energy grain
boundaries exhibit faster diffusivities than low-energy grain boundaries. The
viewpoint proposed in [356] is thus consistent with the simulation results.
The fact that this concept was shown to hold in covalently bonded system,
406 Atomistic Modeling of Materials Failure
palladium as well as copper (the present study) shows that the transforma-
tion of grain boundaries into liquid-like structures may be a more general
concept independent of the details of the atomic bonding. Experimental
results [346, 350] indicate that the nucleation of parallel glide dislocation
occurs repeatedly from grain boundary sources near the film–substrate inter-
face while strain is increased during thermal cycling. The same phenomenon
is observed in the atomistic simulations reported in this chapter, although
the conditions are quite different. Repeatedly emitted parallel glide disloca-
tions form a pileup when they move toward an obstacle, which can be other
grain boundaries (e.g., twins) in the experiments or boundary atoms in the
simulations. Repeated nucleation is possible because by each parallel glide
nucleation, only one climb edge dislocation in the grain boundary is annihi-
lated while many of them remain “stored” in the grain boundary. The total
Burgers vector stored in the grain boundary is found to remain constant.
As discussed in Sect. 8.2.2, in films thinner than 10 nm, image stresses on
climb dislocations can be as large as 1 GPa. This can severely hinder climb
mediated diffusional creep, suggesting that the behavior of discrete disloca-
tions needs to be considered for the nanoscale thin films. This is also supported
by the atomistic results showing that stress cannot be relaxed completely in
extremely thin films. In addition to the theoretical and computational evi-
dence, the results are not contradicting experimental results which often show
large residual stresses in extremely thin films [342].
Employing the molecular dynamics results that nucleation occurs at a
critical stress intensity factor K PG , we estimate the necessary lateral stress σ0
in order to achieve this stress intensity factor at t → ∞. In films thinner than a
critical thickness between 10 and 20 nm, the analysis predicts stresses reaching
the cohesive strength of the material. Hence, before nucleation of parallel
glide dislocations the simulation sample will be destroyed by homogeneous
decohesion. In the present quasi-two-dimensional setup with rigid boundaries,
an additional issue is that dislocations from the boundaries are nucleated
when very large strains are applied.
These considerations suggest a minimum thickness for parallel glide dislo-
cation nucleation. The critical stress intensity factor for dislocation nucleation
from a crack and a diffusion wedge at 0 K is about three times larger than the
values calculated from atomistic simulation results at elevated temperature.
This can be explained by the finite temperature in the simulations. Yet it is
important that both approaches suggest that Kdw PG PG
/Kcr ≈ 2.
The discussion reveals that the diffusion wedge has similar properties as
a crack, but requires a larger stress intensity factor to nucleate a dislocation.
The reason for this is that in the case of a diffusion wedge, a dislocation
dipole needs to be formed and the dipole interaction force is twice as strong
as the image force on an emergent dislocation near a crack tip. This is an
important result of atomistic modeling that corroborates the assertion made
in the development of the Rice–Thomson model in Sect. 8.2.3.
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 407
8.3.5 Summary
Fig. 8.28 Geometry for studies of plasticity in grain triple junctions. A low-energy
grain boundary is located between grains 1 and 2, and two high-energy grain
boundaries are found between grains 2 and 3 and between 3 and 1
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 409
Here we summarize the details of the atomistic modeling procedure for this
case. To focus on the nucleation process of dislocations and the effect of dif-
ferent types of grain boundaries in detail, we consider a tricrystal model with
a triple junction between three grains. The model is constructed such that it
features two high-energy and one low-energy grain boundary. The schematic
geometry is shown in Fig. 8.28. As indicated in Fig. 8.28, cracked grain bound-
aries with traction-free surfaces along zc < z < hf are used to mimic the
existence of diffusion wedges in all of the grain boundaries. We choose zc ≈
1.5 nm so that the crack does not reach the substrate (z = 0 at the substrate).
This is motivated by the results of molecular dynamics simulations showing
that the glide plane of dislocations is not directly at the substrate but a few
atomic layers above.
Loading is applied by prescribing a displacement to the outermost rows at
the boundary of the quadratic slab. Grain 1 has the reference configuration
([110] in the x-direction, [112] in the y-direction). Grain 2 is rotated counter-
clockwise by 7.4◦ , and grain 3 is rotated by 35◦ with respect to grain 1. The
low-energy grain boundary is situated between grains 1 and 2, and the two
high-energy grain boundaries are between grains 2 and 3 and between grains
3 and 1.
The structure of the low-energy grain boundary is significantly different
from that of the high-energy ones. The former is essentially composed of a peri-
odic array of misfit dislocations, with a strongly inhomogeneous distribution of
strain energy along the grain boundary. In contrast, the strain energy along
410 Atomistic Modeling of Materials Failure
The boundary conditions of all models are chosen such that atoms close to the
film–substrate interface are pinned to their initial locations (and also moved
according to the applied strain field), mimicking perfect adhesion of a film
on a stiff substrate. After the initial atomic configuration is created, a global
energy minimization scheme is applied to relax the structure.
The studies are carried out using a microcanonical N V E ensemble with
a quasistatic energy minimization scheme. The biaxial strain in the samples
is gradually increased up to 2.5%. The multibody embedded atom potential
(EAM) potential for copper developed by Mishin and coworkers [107] is used
for these studies.
Analysis Techniques
The simulation results are analyzed with the centrosymmetry technique [36],
which is a convenient way to discriminate between different defects such as
partial dislocations, stacking faults, grain boundaries, surfaces, and surface
steps. In some cases, we will also use the slip vector technique proposed
recently [147]. This method allows us to extract quantitative information
about the Burgers vector and slip plane of dislocations immediately from
the simulation data.
Fig. 8.29 Nucleation of parallel glide dislocations from a grain triple junction. The
plot shows a time sequence based on a centrosymmetry analysis, showing how several
dislocation half loops nucleate and grow into the grain interior
Fig. 8.30 Schematic of dislocation nucleation from different types of grain bound-
aries. Individual misfit dislocations at low-energy grain boundaries serve as sources
for dislocations. At high-energy grain boundaries, there is no inherent, specific nucle-
ation site so that the point of largest resolved shear stress, the grain triple junction,
serves as nucleation point
occurs due to repeated nucleation of partial dislocations with the same Burgers
vector.
Jog Dragging
Fig. 8.32 Dislocation junction and bowing of dislocations by jog dragging. A trail of
point defects is produced at the jog in the leading dislocation, which is then repaired
by the following partial dislocation (this is a similar mechanism as that shown in
Figs. 7.16 and 7.17a)
Once different half-loops grow, they combine with each other while form-
ing jogs since they glide on different glide planes. The jog has a nonglissile
component and cannot move conservatively [38] thus causing generation of
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 413
point defects. This, in turn, exerts a drag force on dislocations causing the
dislocation lines to bow.
A similar mechanism of jog dragging due to point defect generation is
known from dislocation cutting processes as depicted in Fig. 8.33. As discussed
in the literature on dislocation mechanics [38], when two dislocations intersect
each acquires a jog equal in direction and length to the Burgers vector of the
other dislocation. If two screw dislocations intersect, this jog cannot glide
conservatively since it features a sessile edge segment. However, if the applied
stress is large enough, the dislocation with the jog starts to glide and the
jog leaves a trail of vacancies or a trail of interstitials depending on the line
orientation and the Burgers vector of the reacting dislocations.
Fig. 8.33 Generation of trails of point defects. Subplot (a): Dislocation cutting
processes with jog formation and generation of trails of point defects. Both dis-
locations leave a trail of point defects after intersection. The arrows indicate the
velocity vector of the dislocations. Subplot (b): Nucleation of dislocations on dif-
ferent glide planes from grain boundaries generate a jog in the dislocation line that
causes generation of trails of point defects
8.4.3 Discussion
Dislocation nucleation depends on the grain boundary structure: We observe
that low-energy grain boundaries composed of a periodic array of misfit
dislocations provide more fertile sources for threading dislocation nucleation.
At low-energy grain boundaries, dislocations are often observed to nucleate
close to grain boundary misfit dislocations. This can be referred to as an intrin-
sic condition, because the concentration of internal grain boundary stresses
serves as nucleation site for dislocations. Since the incipient dislocations are
often nucleated at different glide planes, complex dislocation reactions take
place when several of them combine to form a single dislocation line. Such
mechanisms can hinder dislocation motion and cause bowing of the dislo-
cation line. The observation of such nucleation-induced jogs with subsequent
generation of trails of point defects has not been described in the literature. In
other computer simulation of ductile materials [14], similar mechanisms have
been observed, suggesting that this mechanism may play a role in hardening
of materials.
In the more homogeneous high-energy grain boundaries, there is inherently
no preferred nucleation site. Therefore, triple junctions of grain boundaries
are preferred as nucleation sites. The overall stress field governs dislocation
nucleation, since such a triple junction provides a location with highest stress
concentration. Different parallel glide dislocations can interact in a complex
way to form networks of dislocations as shown in Fig. 8.32.
Another finding is that partial dislocations dominate plasticity in the simu-
lations, as can be verified in Fig. 8.31, where deformation twinning is depicted.
This indicates that partial dislocations dominate plasticity at nanoscale.
Similar observations have been reported by other groups [154, 303].
Fig. 8.34 Geometry for the studies of plasticity in polycrystalline simulation sample
Fig. 8.35 Atomistic model of the polycrystalline thin film. Only surfaces (brighter
coloring) and grain boundaries (darker color) are shown
In contrast to the modeling with the tricrystal model, the simulation cell is
fully periodic in the x- and y-direction. Loading is applied by homogeneously
straining the sample in the desired direction. Further details on the modeling
technique can be found in Sect. 8.4.1.
The plasticity of thin films will be investigated with and without traction
relaxation at the grain boundaries. This enables one to study the details of
the dislocation nucleation process near the grain boundary–substrate inter-
face. Dislocation mechanisms associated with grain boundary cracks will be
compared and related to experimental results.
The model of constrained diffusional creep [46] predicts that due to mass
transport from the surface into the grain boundary, the tractions along the
grain boundary are relaxed, and thus a crack-like stress field develops. This
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 417
Threading Dislocations
Fig. 8.37 Surface view of the film for different times. The threading dislocations
inside the film leave surface steps that appear as darker lines in the visualization
scheme. This plot further illustrates that the dislocation density in grains 3 and 4
is much higher than in the two other grains
dislocation density in grains 3 and 4 is several times higher than that in grains
1 and 2.
Figure 8.39 shows a sequence of a nucleation of a threading dislocation from
the grain boundary–surface interface. The plot indeed shows that threading
dislocations are nucleated at the grain boundary–surface interface and then
the half-loops grow into the film until they reach the substrate. Due to the con-
straint by the substrate, threading dislocations leave an interfacial segment.
This observation is in agreement with the classical understanding of threading
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 419
Fig. 8.38 Detailed view onto the surface (magnified view of snapshot 4 in Fig. 8.37).
The plot shows creation of steps due to motion of threading dislocations. The surface
steps emanate from the grain boundaries, suggesting that dislocations are nucleated
at the grain boundary–surface interface. From the direction of the surface steps it
is evident that different glide planes are activated
Fig. 8.40 Nucleation of parallel glide dislocations, small grain sizes. The plot shows
that dislocation activity centers on the grain boundary whose traction is relaxed.
Due to the crack-like deformation field, large shear stresses on glide planes parallel
to the film surface develop and cause nucleation of parallel glide (PG) dislocations.
Subplot (a) shows a top view and subplot (b) shows a perspective view. The plot
reveals that there are also threading (T) dislocations nucleated from the grain triple
junctions
Fig. 8.41 Nucleation of parallel glide dislocations, large grains. The plot shows a
top view of two consecutive snapshots. The region “A” is shown as a blow-up in
Fig. 8.43
film–substrate interface. Figure 8.40a shows a top view while Fig. 8.40b shows
a perspective side view of the interior of the film.
The section shown has dimensions of approximately 120 nm × 150 nm, and
the film thickness is hf ≈ 15 nm. The grain diameter in the x-direction is
approximately dx ≈ 40 nm. This plot reveals that not only parallel glide
but also some threading dislocations are generated at the grain boundary–
surface interface. The plot shows that dislocation activity centers on the grain
boundary whose traction is relaxed. Due to the crack-like deformation field,
large shear stresses on glide planes parallel to the film surface develop and
cause nucleation of parallel glide dislocations. A complex dislocation network
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 421
Fig. 8.42 Nucleation of parallel glide dislocations, large grains. The plot shows a
view of the surface. From the surface view it is evident that threading dislocations
are nucleated in addition to the parallel glide dislocations. These emanate preferably
from the interface of grain boundaries, traction-free grain boundaries, and the surface
Fig. 8.43 Nucleation of parallel glide dislocations. The plot shows an analysis of
the complex dislocation network of partial parallel glide dislocations that develops
inside the grains (magnified view of the region “A” marked in Fig. 8.41). All defects
besides stacking fault planes are shown in this plot
Figure 8.43 shows the complex dislocation network of partial parallel glide
dislocations that develops inside the grains. In this plot, the stacking fault
planes are not shown. The bowing of the dislocations indicates that their
motion is hindered by mutual interaction. We observe that formation of jogs
and creation of trails of point defects play a very important role as already
discussed in Sect. 8.4.2.
Modeling
The simulation geometry is depicted in Fig. 8.44. The light grey lines inside
the grains refer to the intragrain twin grain boundaries. The thickness of the
twin lamella is denoted by dT .
Simulation Results
The material is loaded uniaxially in the x-direction. We start with sample (i),
and we consider is a grain size of 12.5 nm × 16.5 nm. The grains have the same
424 Atomistic Modeling of Materials Failure
Fig. 8.45 Simulation results of nanostructured material with twin lamella substruc-
ture under uniaxial loading for two different twin lamella thicknesses. Subplot (a)
shows the results for thick twin lamella (dT ≈ 15 nm > d) and subplot (b) for thin-
ner twin lamella (dT ≈ 2.5 nm). Motion of dislocations is effectively hindered at twin
grain boundaries
We report another study with the same microstructure, but with differ-
ent grain misorientation angles, sample (ii). In this case, we choose the grain
boundary misorientation identical in all grains. Grain 1 is in its reference con-
figuration, grain 2 is rotated by 30◦ , grain 3 by 60◦ , and grain 4 is misoriented
by 90◦ . All grain boundaries are now high-angle grain boundaries.
The results of this calculation are shown in Fig. 8.46a. Unlike in Fig. 8.45,
dislocations are now nucleated at all grain boundaries and the nucleation of
dislocations is governed by the resolved shear stress on different glide planes.
It is observed that dislocations can easily penetrate through the stacking fault
planes generated by motion of other partial dislocations, but build pileups at
the twin grain boundaries. We also observe that dislocations with opposite
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 425
Fig. 8.46 Simulation results of nanostructured material with twin lamella substruc-
ture under uniaxial loading for two different twin lamella thicknesses, all high-energy
grain boundaries. Subplot (a) shows the potential energy field after uniaxial loading
was applied. Interesting regions are highlighted by a circle. Unlike in Fig. 8.45, dis-
locations are now nucleated at all grain boundaries. The nucleation of dislocations
is now governed by the resolved shear stress on different glide planes. Subplot (b)
highlights an interesting region in the right half where i. cross-slip, ii. stacking fault
planes generated by motion of partial dislocations and iii. intersection of stacking
fault planes left by dislocations is observed
426 Atomistic Modeling of Materials Failure
8.5.5 Discussion
The results reviewed in this section can be summarized as follows:
1. Threading dislocations dominate deformation when tractions along the
grain boundaries are not relaxed. However, if the grain boundary tractions
are relaxed, parallel glide dislocations dominate the plasticity of ultrathin
films (hf ≈ 15 nm). Almost all plasticity is carried on glide planes that
are very close to each other. This transition of the maximum shear stress
from inclined planes to planes parallel to the film surface is illustrated in
Fig. 8.11.
2. Dislocation nucleation depends on the grain boundary structure. Low-
energy grain boundaries composed of an array of misfit dislocations provide
more fertile sources of dislocations than high-energy grain boundaries with
a more homogeneous structure. We find that the dislocation density is a few
times higher in grains connected by low-angle grain boundaries. This asser-
tion is further supported by the studies of nanostructured bulk material
described in Sect. 8.5.3.
3. Different parallel glide dislocations can interact in a complex way to form
networks of dislocations as shown in Fig. 8.43 (a blow-up picture, show-
ing only partial dislocations and grain boundaries while filtering out the
stacking faults).
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 429
Fig. 8.48 Series of TEM micrographs of an unpassivated copper film, film thickness
hf = 200 nm, showing the nucleation and propagation of parallel glide dislocations.
Dislocations appear as white lines. A total of ten dislocations (numbered in the plots)
are emitted sequentially from the source at the lower left triple junction. Dislocations
are pushed forward by subsequently emitted dislocations, which in turn are not able
to glide as far into the grain as the earlier dislocations (compare subplots (b), (d),
(f), (h)). Based on their motion and on the grain geometry, dislocations must have
undergone glide on the (111) glide plane parallel to the film–substrate interface.
Reprinted from [51] Acta Materialia, Vol. 51, T.J. Balk, G. Dehm and E. Arzt,
Parallel glide: unexpected dislocation motion parallel to the substrate in ultrathin
copper films, pp. 4471–4485, copyright c 2003, with permission from Elsevier
in Fig. 5.6, for instance. The same behavior was observed in these simulations
as with purely atomistic methods.
Without relaxation of grain boundary tractions, threading dislocations
dominate thin film plasticity, while under grain boundary diffusion, disloca-
tions on parallel glide planes dominate. Threading dislocations are found to
be mostly complete dislocations, while we see a strong tendency to nucleate
partial dislocations in the case of parallel glide dislocations in the nanometer-
sized grains investigated here. This is qualitatively consistent with results of
atomistic modeling of deformation of nanocrystalline materials [154,303,304].
At nanoscale, the role of partial dislocations becomes increasingly impor-
tant. Twinning along parallel planes might become an important deformation
mechanism at high strain rates as shown in Fig. 8.31.
The transition of the deformation mechanism from threading dislocations
to parallel glide dislocations is also observed in recent experimental investiga-
tions [346, 350]. Experiment has clearly confirmed that once grain boundary
diffusion is shut down, for example in very thick films or due to the existence
of a capping layer, threading dislocations dominate plasticity [51, 346, 350].
When grain boundary diffusion is active, either because there is no capping
layer or because the film thickness is sufficiently small, parallel glide dislo-
cations dominate. These observations indicates that mechanisms relaxing the
grain boundary tractions are active during the deformation of ultra-thin films.
Experimental results are in good qualitative agreement with the molecular
dynamics results reviewed here.
Another important feature is that parallel glide dislocations do not glide as
easily along inhomogeneous low-angle grain boundaries as they do along homo-
geneous high-energy grain boundaries as shown in Fig. 8.41. This is explained
by the fact that the low-energy grain boundaries are composed of an array
of misfit dislocations and thus rather inhomogeneous. Similar mechanism has
been observed in experiment. In [51], it was reported that dislocations are
effectively repelled from certain type of grain boundaries causing significant
bowing.
Figure 8.40b reveals that not only parallel glide but also some thread-
ing dislocations are generated at the grain boundary–surface interface. This
observation is in qualitative agreement with experiment [51]. The studies of
constrained grain boundary diffusion in polycrystalline samples led to similar
results and are also in qualitative agreement with experiment.
Fig. 8.49 Flow stress σY vs. the film thickness hf , as obtained from mesoscopic
simulations of constrained diffusional creep and parallel glide dislocation nucleation
(data taken from [50]). The results are shown for two different initiation criteria
for diffusion and a film-dependent source. In the case of a local criterion for dif-
fusion initiation, the yield stress is film-thickness independent as also observed in
experiment [51]
a threshold stress for grain boundary diffusion [50]. The investigation of dif-
ferent conditions for nucleation of climb and glide dislocations, as well as their
interaction with grain boundaries, suggests that the diffusion threshold stress
should only depend on the strain in the top layer of the film, and thus be inde-
pendent of the film thickness. This gives rise to a thickness-independent flow
stress for ultra-thin films, in good agreement with the relevant experimental
results [51].
An important point is that only a local criterion for initiation of diffusion
leads to a film-thickness-independent yield stress [50, 51]. If the source for
diffusion initiation is chosen film-thickness dependent, the yield stress is also a
function of the film thickness. Since the yield stress is found independent of the
film thickness in experiment [51], there is reason to believe that the assertion
of a local criterion for diffusion initiation is correct. This is in contrast to the
nucleation criterion for parallel glide dislocations, which is a global criterion
dependent on the film thickness. The results for the yield stress obtained
from mesoscopic simulation are summarized in Fig. 8.49 [50]. The yield stress
as a function of film thickness is shown for two different initiation criteria
for diffusion (constant source and therefore local criterion, and film-thickness-
dependent source). In the case of a local criterion for diffusion initiation, the
yield stress is film-thickness independent as observed in experiment [51]. The
yield stress increases slightly for very thin films.
The results obtained by this hierarchical multiscale simulation method illus-
trate the usefulness of the atomistic approach and its possible transferability
434 Atomistic Modeling of Materials Failure
to other materials phenomena for which fully atomistic simulations are not yet
feasible.
The results from the numerical modeling reviewed here and in [50] together
with experimental findings reported by different authors [51] allow us to qual-
itatively describe different deformation mechanisms that occur in thin films in
the submicron regime. We propose that there exist four different deformation
regimes. These are:
• Regime A: Deformation with threading dislocations
• Regime B: Constrained diffusional creep with subsequent parallel glide
• Regime C: Constrained diffusional creep without parallel glide
• Regime D: No stress relaxation mechanism with no diffusion and no
dislocation motion
A schematic “deformation map” is plotted in Fig. 8.50. This plot shows
the critical applied stress to initiate different mechanisms of deformation as
a function of the film thickness. We assume that the loading is applied very
slowly and the temperature is sufficiently high such that diffusive processes
are generally admitted.
The critical applied stress to nucleate threading dislocations scales with
1/hf [71–73]. We note that the 1/hf -scaling has been found in two-dimensional
molecular dynamics simulations [361] recently. Two-dimensional mesoscopic
studies [182] revealed qualitatively that the flow stress increases with decreas-
ing film thickness.
For films thicker than a material-dependent value, regime (A) is the dom-
inating deformation mechanism. For thinner films, the stress necessary to
nucleate threading dislocations must be assumed larger than the stress to ini-
tiate grain boundary diffusion. In this regime (B), diffusion dominates stress
relaxation and causes a plateau in the flow stress as shown by the discrete
dislocation modeling.
Parallel glide helps to maintain grain boundary diffusion until the overall
stress level is below the diffusion threshold which is independent of the film
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 435
Fig. 8.50 Deformation map of thin films, different regimes. Thin films with thick-
nesses in the submicron regime feature several novel mechanisms next to the
deformation by threading dislocations (a). Plasticity can be dominated by diffu-
sional creep and parallel glide dislocations (b), purely diffusional creep (c), and no
stress relaxation mechanism (d)
thickness. For yet thinner films, grain boundary diffusion stops before a suf-
ficient stress concentration to trigger parallel slip is obtained as suggested by
our molecular dynamics simulations.
The onset of regime (C) can be described with the scaling of the critical
nucleation stress for parallel glide with 1/hsf (s ≈ 0.5). In this regime, the
flow stress increases again for smaller films, due to the back stress of the
climb dislocations in the grain boundary, effectively stopping further grain
boundary diffusion. If the applied stress is lower than the critical stress for
diffusion, no stress relaxation mechanism is possible. This is referred to as
regime (D). The critical film thickness of 25 nm is estimated based on the
PG
result for the critical Kdw from molecular dynamics simulations.
The investigations of ultra-thin films show the richness of phenomena that
occur, as the dimensions of materials are shrunk to nanometer scale. For
tomorrow’s engineers, such knowledge may be the key to successful design.
The yield stress of thin films resulting from these considerations is summa-
rized in Fig. 8.51 for different film thicknesses. For thicker films, the strength
increases inversely proportional to the film thickness as has been shown in
many theoretical and experimental studies [51,72,96,343]. If the films thickness
is small enough such that grain boundary diffusion and parallel glide are the
prevailing deformation mechanisms, the film strength is essentially indepen-
dent of hf , as shown by the discrete dislocation model (reviewed in Sect. 8.6)
436 Atomistic Modeling of Materials Failure
Fig. 8.51 Deformation mechanism map of thin copper films, here focused on the
yield stress. For films in the submicron regime (thinner than about 400 nm), the
yield stress shows a plateau. This is the regime where diffusional creep and parallel
glide dislocations dominate (regime (B) in Fig. 8.50)
and seen in experiment [51]. However, for films thinner than hf ≈ 25 nm, the
modeling predicts an increase in strength with decreasing film thickness (see
also Fig. 8.49).
In Fig. 8.51, the film thickness of hf ≈ 400 nm below which the yield stress
remains constant, as well as the plateau yield stress of 0.64 GPa are taken
from experimental results of copper thin films [51, 346].
The studies show that interface properties and geometric confinement can
govern the deformation mechanisms in thin films. Important interfaces in thin
films are:
• The film surface
• The grain boundary between two neighboring grains
• The interface of film and substrate (geometrical constraint)
In the following paragraphs, we will discuss the role of these different interfaces
and constraints on the mechanical behavior.
Film Surface
The film surface is important since it allows that atoms diffuse along the
surface into the grain boundary. As shown in [48], the slower of the pro-
cesses surface or grain boundary diffusion controls the dynamics of constrained
diffusional creep.
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 437
The governing character of the grain boundary structure in thin films is found
either when deformation is mediated by diffusional creep or by dislocation
motion: As discussed in [68], the grain boundary structure has a significant
influence on the diffusivities, and therefore determines how fast the tractions
along the grain boundaries are relaxed and a singular stress field develops.
This is also shown in Fig. 8.47 where the depth of grain boundary grooves
is deepest at high-energy grain boundaries corresponding to the fastest dif-
fusion paths. Another indication of this is that high-energy grain boundaries
lead to more pronounced surface grooves than low-energy grain boundaries
when grain boundary diffusion is active. If deformation is carried by nucle-
ation and motion of dislocations, the structure of the grain boundaries also
has a significant influence on the details of deformation: Low-energy grain
boundaries composed of arrays of misfit dislocations are more fertile sources
for dislocation nucleation than homogeneous high-energy grain boundaries.
On the other hand, motion of parallel glide dislocations through grains may
be hindered due to pinning of dislocations when such a inhomogeneous grain
boundary structure is present (see discussion in Sect. 8.5.2).
Geometrical Constraints
The study supports the notion that geometric confinement has strong impact
on the deformation, and could potentially be utilized to create materials with
superior mechanical properties.
Fig. 8.52 Geometry of a (15,15) single wall carbon nanotube (SWNT) shown in
different views
weak dispersive van der Waals interactions (vdW) play a more prominent role,
and often govern the mechanics or self-assembly dynamics of those materials.
The interplay of such adhesive forces with covalent bonding within CNTs is
not well understood for many CNT systems.
The mechanics of carbon nanotubes has been discussed in various arti-
cles published over the last decade, both from a continuum and atomistic
perspective [369]. In a classical article by Yakobsen and coworkers [369], the
behavior of single, free-standing SWNTs under compressive loading was inves-
tigated using classical, molecular dynamics with an empirical Tersoff–Brenner
potential. Yakobson and his co-authors developed a continuum shell model to
describe the buckling modes of the CNTs. Figure 8.53 plots some of the results.
Figure 8.53a shows the strain energy as a function of compressive strain. Asso-
ciated snapshots of the CNT geometry are shown in Fig. 8.53b, illustrating a
link between the kinks in the energy plot and atomistic deformation mech-
anisms. Figure 8.53c shows a similar analysis for bending deformation of a
CNT.
Atomistic simulations of carbon nanotubes have led to significant insight
into their mechanical behavior, as they have confirmed a very high Young’s
440 Atomistic Modeling of Materials Failure
Fig. 8.54 Shell-rod wire transition of CNTs under compressive loading (schemat-
ically shown in subplot (a)), subplot (b) represents shell-like behavior, subplot (c)
shows the behavior of CNTs as a rod, and subplot (d) shows a CNT that behaves
similarly as a wire (or a long polymer monomer) [370, 371]. The series of plots
illustrates the change in compressive behavior as the length of the CNT increases
systematically
This section is kept very brief and we refer the reader to the literature
for additional information. We only highlight a few results of simulation stud-
ies of CNTs and assemblies of CNTs, in particular focused on assembly and
mechanical properties.
Figure 8.54 depicts a shell-rod wire transition of CNTs under compressive
loading. Figure 8.54a represents the shell-like behavior, Fig. 8.54b shows the
behavior of CNTs as a rod, and Fig. 8.54c shows a CNT that behaves similarly
as a wire [370, 371]. The series of plots illustrates the change in compres-
sive behavior as the length of the CNT increases systematically. This result
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 441
Fig. 8.55 Illustration of the process of coarse graining a CNT, leading to a bead-type
representation [52]
U = UT + UB + Uweak , (8.16)
part is given by the sum over all pair-wise and triple (angular) interactions in
the system, thus
UT;weak = φI (r) (8.17)
pairs
for the tensile and weak interactions (both summed pair-wise), and
UB = φbend (ϕ), (8.18)
angles
summed over all triples of particles in the systems. The bending energy of one
triplet is given by a simple harmonic potential
1
φbend (ϕ) = kbend(ϕijk − ϕ0 )2 , (8.19)
2
with kbend as the spring constant relating to the bending stiffness and ϕijk
as the bending angle between three particles i, j, and k (the angle ϕ0 is the
equilibrium angle). This expression is based on the harmonic angle term given
in (2.37), as it is used in the CHARMM potential, for instance. Calculation
of the bending angle requires consideration of the position of three atoms and
the molecular potential is thus a three-body potential.
The nonlinear stress–strain behavior under tensile loading with a bilinear
model (similar to the one that has been used successfully in the studies of
fracture as discussed above): The biharmonic potential is defined as
(0)
dφT (r) k (r − r0 ) if r < r1 ,
= H(rbreak − r) T (1) (8.20)
dr kT (r − r˜1 ) if r ≥ r1 .
(0) (1)
In this equation, kT and kT stand for the small- and large-deformation
(0) (1)
spring constants. The parameter r˜1 = rq − kT /kT (r1 − r0 ) is obtained from
force continuity conditions. This model is chosen to reproduce the nonlinear
elastic and fracture behavior of carbon nanotubes. The availability of two
spring constants enables modeling changes in the elastic properties due to
increasing deformation. The Heaviside function enables us to describe the
drop of forces to zero at the initiation of fracture of the carbon nanotube. We
assume weak, dispersive interactions between either different parts of each
molecule or different molecules, defined by a Lennard-Jones 12-6 (LJ) poten-
tial. The LJ potential has been shown to be a good model for the dispersive
interactions between CNTs. With these three potentials, all terms in (8.16)
are defined.
The mass of each bead is determined by assuming a homogeneous dis-
tribution of mass in the molecular model. The equilibrium distance between
mesoscale particles is chosen to be r0 = 10 Å, based on the requirement that it
is much smaller than the persistence length. All other parameters are chosen
based on the results of full atomistic simulations (see, for instance Fig. 8.56 for
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 443
Fig. 8.56 Atomistic analysis of the tensile properties of a (5,5) CNT [52]. Subplot
(a): Stress vs. strain behavior during stretching of a (5,5) CNT, result obtained using
the Tersoff potential. The straight lines show the mesoscale model that is developed
based on the atomistic simulation results. Subplot (b): Fracture mechanism of a
(5,5) CNT, modeled using the Tersoff potential (plots show the atomistic mechanics
as the lateral tensile strain is increased). Fracture initiates by generation of localized
shear defects in the hexagonal lattice of the CNT, reminiscent of 5–7 Stone–Wales
defects
Table 8.3 Summary of mesoscopic parameters derived from atomistic modeling [52]
If different parts of the tube come sufficiently close, these attractive forces
should also be present and can form self-folded structures of CNTs. Such
self-folded structures of CNTs with extremely large aspect ratio were first
observed in molecular dynamics simulations of (5,5) CNTs using a com-
bined Tersoff–LJ model. The results from such atomistic and corresponding
mesoscale simulations are shown in Fig. 8.59 [370–372].
The stability of the folded structure is governed by the balance of energy
required to bend the CNT and energy gained by formation of weak vdW
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 445
“bonds.” Therefore there exists a minimum critical length of the CNT (depen-
dent on the bending stiffness and the adhesion parameter) at which such a
structure becomes energetically stable. It was proposed that the critical CNT
length (denoted by Lχ ) scales as
EI
Lχ ∼ . (8.21)
γ
Fig. 8.59 Full atomistic calculations of the properties of ultra-long CNTs (sub-
plot (a)) [371, 372] and corresponding results obtained using the mesoscale model
(subplot (b)) [52]
8.7.3 Discussion
here can be straightforwardly implemented for other CNTs than those studied
here, including multiwall CNTs.
of bone. As illustrated in Fig. 8.60 (right), the mineral platelets are critical for
the integrity of the material since they carry the tensile load. We thus focus
on the fracture strength of the brittle platelets.
The strength of a small mineral particle with a crack under mode I tensile
loading as shown schematically in Fig. 8.61 is considered here. We assume that
classical continuum theory of fracture can be applied to describe this material.
From classical fracture mechanics, the critical stress for crack nucleation in
this perfectly brittle material is given by the Griffith condition G = 2γ, where
γ is the fracture surface energy and G is the energy release rate. For the strip
geometry as shown in Fig. 8.61, with strip width ξ, the energy release rate can
be expressed as
σ 2 ξ(1 − ν 2 )
G= , (8.22)
2E
where E is Young’s modulus, ν is Poisson ratio, and σ is the applied stress. At
the critical point of onset of crack motion, the energy released per unit length
of crack growth must equal the energy necessary to create a unit length of
new surface. Using the Griffith condition, (8.22) can be solved for the critical
448 Atomistic Modeling of Materials Failure
Fig. 8.61 The geometry and dimensions of a cracked platelet. This model is used in
the continuum and atomistic studies of fracture at small scales. We consider a thin
strip of width ξ, in which the crack length extends half way through the length of
the slab in the x-direction. The system is under mode I tensile loading as indicated
in the plot (mode I loading in the y-direction)
applied stress
4γE
σ= (8.23)
ξ(1 − ν 2 )
for spontaneous onset of failure.
For decreasing layer width ξ, (8.23) predicts an increasing stress for nucleation
of the crack, approaching infinity as ξ goes to zero. This, however, cannot be
literally accepted, since the stress cannot exceed the theoretical strength of
the material, which is denoted by σth . This immediately yields a critical layer
width ξcr below which fracture cannot be described by the Griffith theories
any more. Instead, the strength of the cracked slab is given by the theoretical
strength of the material, regardless of the presence of a crack. This critical
length can be calculated to be
4γE
ξcr = 2 (1 − ν 2 ) . (8.24)
σth
Atomistic Modeling
layer in the center of the strip. To model bond breaking, we assume that
atoms across this weak layer interact according to the 12-6 Lennard-Jones
(LJ) potential (2.30).
In the simulations, it is assumed that σ = = 1 and r0 = 1.12246 (FCC
lattice constant a ≈ 1.587). Interactions across the weak fracture layer (LJ
potential) are cut off at a critical distance rcut = 2.5. Bonds only break
between atoms that are located at different sides of the weak layer. This
procedure ensures that the crack can only extend along a predefined direction.
For the analysis of the critical length for flaw tolerance, exact knowledge
of elastic properties and fracture surface energy is needed. It can be shown
that Young’s modulus is
4r2
E = 0 k0 (8.25)
3
and Poisson’s ratio is given by ν = 1/3. The surface energy is given by the
expression reviewed in Sect. 4.4.4.
It is emphasized that this setup of bulk material and a weak layer is par-
ticular convenient for the studies because Young’s modulus E can be easily
varied independent of the other variables (γ and σth ), allowing the critical
lengthscale ξcr to be tuned in a range easily accessible to the molecular
dynamics simulations.
All simulation results are expressed in reduced units: Energies are scaled
by the depth of the LJ potential and lengths are scaled by σ. In these
reduced units, the critical length scale is ξcr ≈ 115 for the material parameters
chosen in the simulation (this corresponds to a few tens of nanometers in real
materials).
A critical element of the studies is calculation of stresses from atomistic
simulations. The atomic stress is calculated based on the virial theorem as
reviewed in Sect. 2.8.6. As illustrated in the studies reviewed in Sect. 6.5, the
atomistic definitions of stress near a moving crack tip show reasonable agree-
ment with continuum mechanics predictions. Thus this approach is reasonable
in performing a direct comparison with continuum mechanical theories.
Fig. 8.62 Fracture and adhesion strength as a function of the size of the material.
The plot depicts the results of bulk fracture as well as surface adhesion. The results
are normalized with respect to the theoretical strength and normalized with respect
to the critical lengthscale for flaw tolerance. These results suggest that the principle
of dimension reduction is valid in a variety of systems, including surface adhesion
as well as bulk fracture
Fig. 8.63 Stress distribution ahead of the crack in a thin mineral platelet just before
failure, for different materials sizes (the x-coordinate is scaled by the characteristic
length scale ξcr ). The thinner the slab, the more homogeneous is the stress distri-
bution. When the slab width is smaller than the critical size, the stress distribution
becomes homogeneous and does not depend on the size of the platelet any more (see
values ξcr /ξ > 2.67 and larger). The normal stress σyy is normalized with respect
to the maximum strength at the onset of failure, σth
Fig. 8.64 Microscopic view of the contact elements in flies (left) and geckos (right).
The terminal setae of flies consist of insect cuticle, that is, chitin–fiber reinforced
protein, and have typical dimensions of 2 µm. The terminal elements (“spatulae”) of
geckos are made of keratin and have typical dimensions of 200-nm diameter [55]
(micrographs courtesy of S. Gorb, Max Planck Institute for Metals Research).
Reprinted from Materials Science and Engineering: C, Vol. 26, E. Arzt, Biologi-
cal and artificial attachment devices: Lessons for materials scientists from flies and
geckos, pp. 1245–1250, copyright c 2006, with permission from Elsevier
Fig. 8.65 The schematic of the model used for studies of adhesion. The model
represents a cylindrical Gecko spatula (as shown in Fig. 8.64) with radius attached to
a rigid substrate (left). A circumferential crack represents flaws for example resulting
from surface roughness. The parameter α denotes the dimension of the crack. The
regime 0 < r < αR corresponds to an area of perfect adhesion, whereas αR <
r < R represents regions of no adhesion. This model resembles the effect of surface
roughness as depicted schematically on the right-hand side
the stress field near the edge of the contact area has a square-root singularity
with stress intensity factor
P √
KI = πaF1 (α), (8.26)
πa2
where F1 (α) varies narrowly between 0.4 and 0.5 for 0 < α < 0.8 (α = 1
corresponds to a perfect, defect-free contact). Equation (8.26) is substituted
into the Griffith condition:
KI2
= ∆γ (8.27)
2E ∗
where the factor two is due to the rigid substrate and ∆γ is the adhesion
energy (corresponding to two times the surface energy). The apparent adhesive
strength normalized by the theoretical strength for adhesion,
Pc
σ̂c = (8.28)
πσth R2
is obtained as
σ̂c = βα2 ψ, (8.29)
where
∆γE ∗
ψ= 2 (8.30)
Rσth
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 455
Fig. 8.66 The geometry of the system considered is a periodic array of punches of
radius R. The rigid–elastic interface leads to singular stress concentrations for flat
punches. We vary the shape of the rigid punch surfaces to avoid these singular stress
concentrations
and
2
β= (8.31)
παF12 (α)
as well as
E
E∗ = (8.32)
1 − ν2
where E is Young’s modulus.
The adhesive strength is a linear function of the dimensionless variable ψ
with slope βα2 . The maximum adhesion strength is achieved when the pull-off
force reaches Pc = σth a2 , or σ̂c = α2 , in which case the traction within the
contact area uniformly reaches the theoretical strength σth . This saturation
in strength occurs at a critical size of the contact area
∆γE ∗
Rcr = β 2 2 . (8.33)
σth
This lengthscale corresponds to ξcr described above.
Fig. 8.67 The shape function defining the surface shape change as a function of the
shape parameter. For Ψ = 1, the optimal shape is reached and stress concentrations
are predicted to disappear
Fig. 8.68 Atom rows in the rigid punch are displaced according to the continuum
mechanics solution of the optimal surface shape (theoretical solution see Fig. 8.67).
This method allows achieving a smoothly varying surface and enables a continuous
transition from a flat punch (left) to the optimal shape (right)
Here we report the results of a series of computer experiments with the models
described in the previous section.
The first question we address is how the adhesion strength varies with the
diameter of the spatula. Figure 8.62 shows the results of atomistic simulations
of the adhesion strength as a function of the size of the material R/Rcr (cor-
responding to the geometry shown in Fig. 8.65). Whereas the strength of the
materials interfaces is predicted reasonably well based on Griffith’s theories
for large dimensions, reduction of dimension results in deviation from this
prediction, and eventually failure of the material at its theoretical strength
σth regardless of the presence of the flaw. In practical terms, that means even
if there is surface roughness present, the roughness does not lead to stress
concentration, and therefore the adhesion structure adheres robustly to the
substrate.
458 Atomistic Modeling of Materials Failure
Fig. 8.69 Stress distribution in the rigid punch slightly before complete detach-
ment (the stress is calculated in a thin strip along the diameter, within the area of
contact Rcut = 2αR). The simulations reveal that for large radii, a stress concentra-
tion develops at the exterior sides of the cylinder. For small dimensions, this stress
distribution starts to vanish.
For dimensions smaller than the critical radius for flaw
tolerance (large ratios of Rcut /R), the stress distribution becomes homogeneous
and does not vary with the cylinder diameter any more
Now we focus on the stress distribution across the adhesion element slightly
before detachment occurs. This study includes variation of (1) the dimensions
of the adhesion element, (2) the adhesion energy, and (3) the elastic properties
of the substrate. A rigid punch on an elastic substrate is considered.
Figure 8.69 shows the stress distribution at the punch–substrate interface
close to detachment for various choices of the ratio R/Rcr . The simulations
reveal that for large radii, a stress concentration develops at the exterior sides
of the cylinder. This result is expected from the classical understanding of
fracture mechanics and corresponds to the regime where Griffith’s theory holds
to describe onset of detachment. For small dimensions, the stress distribution
starts to become uniform. For dimensions smaller than the critical radius for
flaw tolerance, the stress distribution becomes homogeneous and does not vary
with the cylinder diameter any more.
Figure 8.70 shows the variations of the stress distribution close to detach-
ment for changes in adhesion energy and elastic properties of the substrate.
These results further support the notion that
Rcr ∼ E (8.35)
and
Rcr ∼ γ. (8.36)
8 Deformation and Fracture Mechanics of Geometrically Confined Materials 459
Fig. 8.70 Stress distribution in the elastic punch slightly before complete detach-
ment (the stress is calculated in a thin strip along the diameter, within the area
of contact Rcut = 2αR). Here we keep the dimension fixed and vary the adhesion
energy (γ0 corresponds to the surface energy) and the elastic properties (E0 corre-
sponds to the Young’s modulus obtained for k0 = 57.23). We find that the stress
distribution becomes homogeneous for large ratios of Rcr /R, in agreement with
the other results (see Figs. 8.63 and 8.69)
Fig. 8.71 Stress distribution along the diameter of the punch for different choices
of the shape parameter describing the punch shape. The results indicate that when
the optimal shape is reached (Ψ = 1), the stress distribution is completely flat as
in the homogeneous case (λ = 1) without stress magnification. We observe that for
Ψ < 1, a stress concentration develops at the boundaries of the punch, whereas for
Ψ > 1 the largest stress occurs in the center
why nature does not primarily optimize shape but instead focuses on reduction
of dimension as a design strategy.
8.9.5 Summary
Fig. 8.72 Adhesion strength for different choices of the shape parameter Ψ . The
results indicate that although optimal adhesion can be achieved at any lengthscale
by changing the shape of the attachment device (by choosing Ψ = 1), robustness
with respect to variations in shape while at the same time keeping a strong adhesion
force can only be achieved at small lengthscales
complex expressions for the atomic interactions allows carrying out funda-
mental parameter studies enabling immediate comparison with continuum
theories. Even though the results do not allow making quantitative predic-
tions about specific materials, the studies may help to develop a deeper
understanding of the mechanics of brittle fracture at nanoscale.
Increase in the available computational power allows, at the same time,
modeling at lengthscales on the order of micrometers. It is believed that
atomistic-based modeling will play a significant role in the future in the area
of modeling nanomechanical phenomena and linking to continuum mechanical
theories.
References
35. A.F. Voter and S.P. Chen. Accurate interatomic potentials for Ni, Al and Ni3 Al
characterization of defects. Mater. Res. Soc. Symp. Proc., 82:175180, 1987.
36. C. Kelchner, S.J. Plimpton, and J.C. Hamilton. Dislocation nucleation and
defect structure during surface-intendation. Phys. Rev. B, 58:11085–11088,
1998.
37. R.L. Fleischer. Cross slip of extended dislocations. Acta Met., 7:134–135, 1959.
38. J.P. Hirth and J. Lothe. Theory of Dislocations. Wiley-Interscience, New York,
1982.
39. M.J. Buehler, A. Hartmaier, H. Gao, F.F. Abraham, and M. Duchaineau.
Atomic plasticity: Description and analysis of a one-billion atom simulation of
ductile materials failure. Comput. Methods Appl. Mech. Eng., 193(48–51):5257–
5282, 2004.
40. D. Sen and M.J. Buehler. Shock loading of bone-inspired metallic nanocom-
posites. Solid State Phenom., 139:11–22, 2008.
41. D. Sen and M.J. Buehler. Simulating chemistry in mechanical deformation of
metals. Int. J. Multiscale Comput. Eng., 5:181202, 2007.
42. V. Yamakov, S.R.Phillpot D. Wolf, A.K. Mukherjee, and H. Gleiter.
Deformation-mechanism map for nanocrystalline metals by moleculardynamics
simulation. Nat. Mater., 3:43–47, 2004.
43. J. Schiotz and K.W. Jacobsen. A maximum in the strength of nanocrystalline
copper. Science, 301:1357–1359, 2003.
44. N. Broedling, A. Hartmaier, M.J. Buehler, and H. Gao. Strength limit of a
bioinspired metallic nanocomposite. J. Mech. Phys. Solids, 56(3):1086–1104,
2008.
45. D. Sen and M.J. Buehler. Crystal size controlled deformation mechanism:
Breakdown of dislocation mediated plasticity in single nanocrystals under
geometric confinement. Phys. Rev. B, 47:195439, 2008.
46. H. Gao, L. Zhang, W.D. Nix, C.V. Thompson, and E. Arzt. Crack-like grain
boundary diffusion wedges in thin metal films. Acta Mater., 47:2865–2878,
1999.
47. L. Zhang. A class of strongly coupled elasticity and diffusion problems in thin
metal films. PhD thesis, Stanford University, Stanford, CA, 2001.
48. L. Zhang and H. Gao. Coupled grain boundary and surface diffusion in a
polycrystalline thin film constrained by substrate. Z. Metallk., 93:417–427,
2002.
49. M.J. Buehler, A. Hartmaier, and H. Gao. Constrained grain boundary diffusion
in thin copper films. Mater. Res. Soc. Proc., 821:P1.2, 2004.
50. A. Hartmaier, M.J. Buehler, and H. Gao. A discrete dislocation model of
diffusional creep. Diffusion and Defect Forum, 224–225:107–128, 2003.
51. T.J. Balk, G. Dehm, and E. Arzt. Parallel glide: Unexpected dislocation motion
parallel to the substrate in ultrathin copper films. Acta Mater., 51:4471–4485,
2003.
52. M.J. Buehler. Mesoscale modeling of mechanics of carbon nanotubes: Self-
assembly, self-folding and fracture. J. Mater. Res., 21(11):2855–2869, 2006.
53. I. Jager and P. Fratzl. Mineralized collagen fibrils: A mechanical model with
a staggered arrangement of mineral particles. Biophys. J., 79(4):1737–1746,
2000.
466 References
54. H. Gao, B. Ji, I.L. Jäger, E. Arzt, and P. Fratzl. Materials become insen-
sitive to flaws at nanoscale: Lessons from nature. P. Natl. Acad. Sci. USA,
100(10):5597–5600, 2003.
55. E. Arzt. Iological and artificial attachment devices: Lessons for materials
scientists from flies and geckos. Mater. Sci. Eng.: C, 26:1245–1250, 2006.
56. C.E. Inglis. Stresses in a plate due to the presence of cracks and sharp corners.
Trans. Inst. Naval Architects, 55:219, 1913.
57. G.I. Taylor. Mechanism of plastic deformation in crystals. Proc. R. Soc. A,
145:362, 1934.
58. J.E. Gordon. The New Science of Strong Materials. Princeton University Press,
Princeton, NJ, 1984.
59. M. Marder and J. Fineberg. How things break. Phys. Today, 49(9):24–29, 1996.
60. D. Hull and D.J. Bacon. Introduction to Dislocations. Butterworth Heinemann,
2002.
61. K.B. Broberg. Cracks and Fracture. Academic Press, San Diego, CA, 1990.
62. A.A. Griffith. The phenomenon of rupture and flows in solids. Phil. Trans.
Roy. Soc. A, 221:163–198, 1920.
63. G.R. Irwin. In: Fracturing of Metals, Fracture dynamics, pages 147–166,
American Society for Metals, Cleveland, OH, 1948.
64. T.L. Anderson. Fracture Mechanics: Fundamentals and Applications. CRC
Press, Boca Raton, 1991.
65. L.B. Freund and S. Suresh. Thin Film Materials. Cambridge University Press,
Cambridge, 2003.
66. J.R. Rice and R.M. Thomson. Ductile versus brittle behavior of crystals. Phil.
Mag., 29:73–97, 1974.
67. J.R. Rice and G.B. Beltz. The activation energy for dislocation nucleation at
a crack. J. Mech. Phys. Solids, 42:333–360, 1994.
68. V. Yamakov, D. Wolf, S.R. Phillpot, and H. Gleiter. Grain-boundary diffusion
creep in nanocrystalline palladium by molecular-dynamics simulation. Acta
Mater., 50:61–73, 2002.
69. V. Yamakov, D. Wolf, S.R. Phillpot, and H. Gleiter. Deformation mechanism
crossover and mechanical behaviour in nanocrystalline materials. Phil. Mag.
Lett., 83:385–393, 2003.
70. H. van Swygenhoven and P.M. Derlet. Grain-boundary sliding in nanocrys-
taline fcc metals. Phys. Rev. B, 64:224105, 2001.
71. L.B. Freund. The stability of a dislocation in a strained layer on a substrate.
J. Appl. Mech., 54:553–557, 1987.
72. W.D. Nix. Mechanical properties of thin films. Metall. Trans. A, 20:2217–2245,
1989.
73. W.D. Nix. Yielding and strain hardening of thin metal films on substrates.
Scripta Mater., 39:545–554, 2001.
74. C.T. Lim, M. Dao, S. Suresh, C.H. Sow, and K.T. Chew. Large deformation
of living cells using laser traps. Acta Mater., 52:1837–1845, 2004.
75. S. Suresh, J. Spatz, J.P. Mills, A. Micoulet, M. Dao, C.T. Limand, M. Beil,
and T. Seufferlein. Connections between single-cell biomechanics and human
disease states: gastrointestinal cancer and malaria. Acta Biomaterialia, 1:15–
30, 2005.
References 467
76. J. Li, M. Dao, C.T. Lim, and S. Suresh. Spectrin-level modeling of the
cytoskeleton and optical tweezers stretching of the erythrocyte. Biophys. J.,
88:3707–3719, 2005.
77. G. Bao and S. Suresh. Cell and molecular mechanics of biological materials.
Nat. Mater., 2(11):715–725, 2003.
78. M.J. Buehler. Nano- and micromechanical properties of hierarchical biological
materials and tissues. J. Mater. Sci., 42(21):8765–8770, 2007.
79. M. Sotomayor and K. Schulten. Single-molecule experiments in vitro and in
silico. Science, 316(5828):1144–1148, 2007.
80. M.J. Buehler and T. Ackbarow. Fracture mechanics of protein materials.
Mater. Today, 10(9):46–58, 2007.
81. B.J. Alder and T.E. Wainwright. Phase transformations for a hard sphere
system. J. Chem. Phys., 27:1208–1209, 1957.
82. B.J. Alder and T.E. Wainwright. Studies in molecular dynamics. 1. General
method. J. Chem. Phys., 31:459–466, 1959.
83. A. Rahman. Correlations in the motions of atoms in liquid Argon. Phys. Rev.,
136:A405A411, 1964.
84. M.P. Allen and D.J. Tildesley. Computer Simulation of Liquids. Oxford
University Press, Oxford, 1989.
85. M.F. Ashby. Modelling of materials problems. J. Comput. Aided Mater. Design,
3:95–99, 1996.
86. S. Yip. Lecture notes of subject 22.00. MIT’s Open Course Ware site,
http://ocw.mit.edu/OcwWeb/Nuclear-Engineering/22-00JSpring-2006/
CourseHome/, 2007).
87. M. Springborg. Density-Functional Methods in Chemistry and Materials Sci-
ence. Wiley research series in Theoretical Chemistry, Wiely, New York,
1997.
88. R.P. Feynman. Feynman Lectures On Physics. Addison-Wesley, Reading, MA,
1970.
89. A.F. Voter, F. Montalenti, and T.C. Germann. Extending the time scale in
atomistic simulation of materials. Ann. Rev. Mat. Res., 32:321–346, 2002.
90. W.K. Hastings. Monte carlo sampling methods using markov chains and their
applications. Biometrika, 57(1):97–109, 1970.
91. M.N. Rosenbluth, A.H. Teller, N. Metropolis, A.W. Rosenbluth, and E. Teller.
Equations of state calculations by fast computing machines. J. Chem. Phys.,
21(6):1087–1092, 1953.
92. R. Car and M. Parrinello. Unified approach for molecular dynamics and density
functional theory. Phys. Rev. Lett., 55:2471, 1985.
93. A.C.T. van Duin, S. Dasgupta, F. Lorant, and W.A. Goddard. Reaxff: A
reactive force field for hydrocarbons. J. Phys. Chem. A, 105:9396–9409, 2001.
94. M.W. Finnis and J.E. Sinclair. A simple empirical n-body potential for
transition metals. Phil. Mag. A, 50:45–55, 1984.
95. R. Madec, B. Devincre, L. Kubin, T. Hoc, and D. Rodney. The role of collinear
interaction in dislocation-induced hardening. Science, 301:1879–1882, 2003.
96. B. von Blanckenhagen, P. Gumbsch, and E. Arzt. Dislocation sources in dis-
crete dislocation simulations of thin film plasticity and the Hall-Petch relation.
Model. Simul. Mater. Sci. Eng., 9:157–169, 2001.
97. H. H. M. Cleveringa, E. van der Giessen, and A. Needleman. Comparison
of discrete dislocation and continuum plasticity predictions for a composite
material. Acta Mater., 45:3163–3179, 1997.
468 References
98. H.H.M. Cleveringa, E. van der Giessen, and A. Needleman. A discrete dislo-
cation analysis of mode I crack growth. J. Mech. Phys. Solids, 48:1133–1157,
2000.
99. H.H.M. Cleveringa, E. Van der Giessen, and A. Needleman. A discrete
dislocation analysis of bending. Int. J. Plast., 15:837–868, 1999.
100. L.P. Kubin, G. Canova, M. Condat, B. Devincre, V. Pontikis, and Y. Bréchet.
Dislocation microstructures and plastic flow: A 3D simulation. Solid State
Phenom., 23–24:455–472, 1992.
101. L.P. Kubin, B. Devincre, G. Canova, and Y. Bréchet. 3-D simulations of
dislocations and plasticity. Key Eng. Mater., 103:217–226, 1995.
102. B. van Beest, G. Kramer, and R. van Santen. Force-fields for silicas and alu-
minophosphates based on ab-initio calculations. Phys. Rev. Lett., 64(16):1955,
1990.
103. P.M. Morse. Diatomic molecules according to the wave mechanics. II-
vibrational levels. Phys. Rev., 34:57–64, 1929.
104. N.J. Wagner, B.L. Holian, and A.F. Voter. Molecular-dynamics simulation of
two-dimensional materials at high strain rates. Phys. Rev. A, 45:8457–8470,
1992.
105. P. Gumbsch, S.J. Zhou, and B.L. Holian. Molecular dynamics investigation of
dynamic crack instability. Phys. Rev. B, 55:3445, 1997.
106. R. Komanduri, N. Chandrasekaran, and L.M. Raff. Molecular dynamics (MD)
simulations of uniaxial tension of some single-crystal cubic metals at nanolevel.
Int. J. Mech. Sci., 43:2237–2260, 2001.
107. Y. Mishin, M.J. Mehl, D.A. Papaconstantopoulos, A.F. Voter, and J.D. Kress.
Structural stability and lattice defects in copper: Ab-initio, tight-binding and
embedded-atom calculations. Phys. Rev. B, 63:224106, 2001.
108. F.J. Cherne, M.I. Baskes, and P.A. Deymier. Properties of liquid nickel: A crit-
ical comparison of EAM and MEAM calculations. Phys. Rev. A, 65(2):024209,
2002.
109. M.I. Baskes. Embedded-atom method: Derivation and application to impu-
rities, surfaces and other defects in metals. Phys. Rev. B, 29(12):6443–6543,
1984.
110. K.W. Jacobsen, J.-K. Norskov, and M.J. Puska. Interatomic interactions in the
effective-medium theory. Phys. Rev. B, 35(14):7423, 1987.
111. K.W. Yacobsen, P. Stoltze, and J.K. Norskov. A semi-empirical effective
medium theory for metals and alloys. Surf. Sci., 366:394, 1996.
112. J.A. Zimmermann. Continuum and atomistic modeling of dislocation nucleation
at crystal surface ledges. PhD thesis, Stanford University, 1999.
113. J.A. Zimmerman, H. Gao, and F.F. Abraham. Generalized stacking fault
energies for embedded atom FCC metals. Model. Simul. Mater. Sci. Eng.,
8:103–115, 2000.
114. J.G. Swadener, M.I. Baskes, and M. Nastasi. Molecular dynamics simulation
of brittle fracture in silicon. Phys. Rev. Lett., 89(8):085503, 2002.
115. A.D. MacKerell and D. Bashford et al. All-atom empirical potential for
molecular modeling and dynamics studies of proteins. J. Phys. Chem. B,
102:3586–3616, 1998.
116. S.L. Mayo, B.D. Olafson, and W.A. Goddard. Dreiding - a generic force-field
for molecular simulations. J. Phys. Chem., 94:8897–8909, 1990.
References 469
117. A.K. Rappe, C.J. Casewit, K.S. Colwell, W.A. Goddard, and W.M. Skiff. Uff, a
full periodic-table force-field for molecular mechanics and molecular-dynamics
simulations. J. Am. Chem. Soc., 114(25):10024–10035, 1992.
118. A.D. Mackerell. Empirical force fields for biological macromolecules: Overview
and issues. J. Comput. Chem., 25(13):1584–1604, 2004.
119. D.W. Brenner. Relationship between the embedded-atom method and Tersoff
potentials. Phys. Rev. Lett., 63:1022, 1989.
120. J. Pauling. Atomic radii and interatomic distances in metals. J. Am. Chem.
Soc., 69(3):542–553, 1947.
121. H.S. Johnston. Adv. Chem. Phys., 3:131, 1960.
122. H.S. Johnston and C.A. Parr. Activation energies from bond energies. I.
Hydrogen transfer reactions. J. Am. Chem. Soc., 85:2544–2551, 1963.
123. J. Tersoff. New empirical model for the structural properties of silicon. Phys.
Rev. Lett., 56(6):632–635, 1986.
124. F. Stillinger and T.A. Weber. Computer-simulation of local order in condensed
phases of silicon. Phys. Rev. B, 31(8):5262–5271, 1985.
125. D.W. Brenner. Empirical potential for hydrocarbons for use in simulating the
chemical vapor-deposition of diamond films. Phys. Rev. B, 42(15):9458–9471,
1990.
126. D.W. Brenner, O.A. Shenderova, J.A. Harrison, S.J. Stuart, B. Ni, and S.B.
Sinnott. A second-generation reactive empirical bond order (rebo) potential
energy expression for hydrocarbons. J. Phys. Condens. Matt., 14(4):783–802,
2002.
127. S.J. Stuart, A.B. Tutein, and J.A. Harrison. A reactive potential for hydro-
carbons with intermolecular interactions. J. Chem. Phys., 112(14):6472–6486,
2000.
128. S. Cheung, W. Deng, A.C.T. van Duin, and W.A. Goddard. ReaxFFmgh reac-
tive force field for magnesium hydride systems. J. Phys. Chem. A., 109:851,
2005.
129. K.D. Nielson, A.C.T. van Duin, J. Oxgaard, W. Deng, and W.A. Goddard.
Development of the ReaxFF reactive force field for describing transition metal
catalyzed reactions, with application to the initial stages of the catalytic
formation of carbon nanotubes. J. Phys. Chem. A., 109:49, 2005.
130. K. Chenoweth, S. Cheung, A.C.T. van Duin, W.A. Kober, and W.A. Goddard.
Simulations on the thermal decompositions of a poly(dimethylsiloxane) poly-
mer using the ReaxFF reactive force field. J. Am. Chem. Soc., 127:7192–7202,
2005.
131. A.C.T. van Duin, A. Strachan, S. Stewman, Q. Zhang, X. Xu, and W.A.
Goddard. Reaxffsio : Reactive force field for silicon and silicon oxide systems.
J. Phys. Chem. A, 107:3803–3811, 2003.
132. A.K. Rappe and W.A. Goddard. Charge equilibration for molecular dynamics
simulations. J. Phys. Chem., 95:3358–3363, 1991.
133. A. Nakano, R. Kalo, K. Nomura, A. Sharma, P. Vashishta, F. Shimojo,
A. van Duin, W.A. Goddard III, R. Biswas, and D. Srivastava. A divide-and-
conquer/cellular-decomposition framework for million-to-billion atom simula-
tions of chemical reactions. Comp. Mat. Sci., 38:642652, 2007.
134. B.P. Uberuaga, S.J. Stuart, and A.F. Voter. Introduction to the time scale
problem. Proc. of Comput. Nanosci. Nanotech., ISBN 0-9708275-6-3:128–131,
2002.
470 References
135. J. Su and W.A. Goddard. Excited electron dynamics modeling of warm dense
matter. Phys. Rev. Lett., 99:185003, 2007.
136. J.A. Zimmerman, E.B. Webb, J.J. Hoytz, R.E. Jonesy, P.A. Klein, and D.J.
Bammann. Calculation of stress in atomistic simulation. Model. Sim. Mat. Sci.
Engr., 12:S319–S332, 2004.
137. C.L. Rountree, R.K. Kalia, E. Lidorikis, A. Nakano, L. van Brutzel, and
P. Vashishta. Atomistic aspects of crack propagation in brittle materials:
Multimillion atom molecular dynamics simulations. Annu. Rev. Mater. Res.,
32:377–400, 2002.
138. F.F. Abraham, R. Walkup, H. Gao, M. Duchaineau, T.D. de La Rubia, and
M. Seager. Simulating materials failure by using up to one billion atoms
and the world’s fastest computer: Work-hardening. P. Natl. Acad. Sci. USA,
99(9):5783–5787, 2002.
139. A. Sharma, R.K. Kalia, and P. Vashishta. Large multidimensional data
vizualization for materials science. Comp. Sci. Eng., 5(2):26–33, 2003.
140. P. Vashishta, R.K. Kalia, and A. Nakano. Multimillion atom molecular dynam-
ics simulations of nanostructures on parallel computers. J. Nanopart. Res.,
5:119–135, 2003.
141. K. Kadau, T.C. Germann, and P.S. Lomdahl. Molecular dynamics comes of age:
320 billion atom simulation on bluegene/L. Int. J. Mod. Phys. C, 17:1755–1761,
2006.
142. W. Gropp, W. Lusk, and A. Skjellum. Using MPI. MIT Press, Cambridge,
MA, 2nd edition, 1999.
143. S. Plimpton. Fast parallel algorithms for short-range molecular-dynamics. J.
Comput. Phys., 117:1–19, 1995.
144. A. Nakano, M.E. Bachlechner, R.K. Kalia, E. Lidorikis, P. Vashishta, and G.Z.
Voyiadijs. Multiscale simulation of nanosystems. Comp. Sci. Eng., 3(4):56–66,
2001.
145. J. Stone, J. Gullingsrud, P. Grayson, and K. Schulten. A system for interactive
molecular dynamics simulation. In: 2001 ACM Symposium on Interactive 3D
Graphics, pages 191–194, 2001.
146. F.F. Abraham, D. Brodbeck, W.E. Rudge, and X. Xu. A molecular dynamics
investigation of rapid fracture mechanics. J. Mech. Phys. Solids, 45(9):1595–
1619, 1997.
147. J.A. Zimmerman, C.L. Kelchner, P.A. Klein, J.C. Hamilton, and S.M. Foiles.
Surface step effects on nanoindentation. Phys. Rev. Lett., 87(16):165507, 2001.
148. W.A. Humphrey, A. Dahlke, and K. Schulten. Vmd: Visual molecular dynam-
ics. J. Mol. Graph., 14:33, 1996.
149. J. Schiotz, T. Leffers, and B.N. Singh. Dislocation nucleation and vacancy
formation during high-speed deformation of fcc metals. Phil. Mag. Lett.,
81:301–309, 2001.
150. D. Faken and H. Jonsson. Systematic analysis of local atomic structure
combined with 3D computer graphics. Comput. Mater. Sci., 2:279, 1994.
151. J.D. Honeycutt and H.C. Andersen. Molecular dynamics study of melting and
freezing of small Lennard-Jones clusters. J. Phys. Chem., 91:4950, 1987.
152. P.M. Derlet, A. Hasnaoui, and H. van Swygenhoven. Atomistic simulations as
guidance to experiments. Scripta Mater., 49:629–635, 2003.
153. D. Farkas, H. van Swygenhoven, and P. Derlet. Intergranular fracture in
nanocrystalline metals. Phys. Rev. B, 66:184112, 2002.
References 471
154. H. van Swygenhoven, P.M. Derlet, and A. Hasnaoui. Atomic mechanism for
dislocation emission from nanosized grain boundaries. Phys. Rev. B, 66:024101,
2002.
155. F.F. Abraham, D. Brodbeck, R.A. Rafey, and W.E. Rudge. Instability
dynamics of fracture: A computer simulation investigation. Phys. Rev. Lett.,
73(2):272–275, 1994.
156. F.F. Abraham, D. Brodbeck, W.E. Rudge, J.Q. Broughton, D. Schneider,
B. Land, D. Lifka, J. Gerner, M. Rosenkranz, J. Skivira, and H. Gao. Ab-
initio dynamics of rapid fracture. Model. Simul. Mater. Sci. Eng., 6:639–670,
1998.
157. P. Klein and H. Gao. Crack nucleation and growth as strain localization in a
virtual-bond continuum. Eng. Fract. Mech., 61:21–48, 1998.
158. P. Zhang, P. Klein, Y. Huang, and H. Gao. Numerical simulation of cohe-
sive fracture by the virtual-internal-bond model. Comput. Modeling Eng. Sci.,
3(2):263–277, 1998.
159. H. Gao and P. Klein. Numerical simulation of crack growth in an isotropic solid
with randomized internal cohesive bonds. J. Mech. Phys. Solids, 46(2):187–218,
2001.
160. M. Born and K. Huang. Dynamical Theories of Crystal Lattices. Clarendon,
Oxford, 1956.
161. K. Huang. On the atomic theory of elasticity. Proc. R. Soc. Lond., 203:178–194,
2002.
162. J.J. Weiner. Hellmann-Feynmann theorem, elastic moduli, and the cauchy
relation. Phys. Rev. B, 24:845–848, 1983.
163. R.W. Ogden. Nonlinear Elastic Deformations. Wiley, New York, 1984.
164. J.E. Marsden and T.J.R. Hughes. Mathematical Foundations of Elasticity.
Prentice Hall, Englewood Cliffs, NJ, 1983.
165. H. Gao, Y. Huang, and F.F. Abraham. Continuum and atomistic studies of
intersonic crack propagation. J. Mech. Phys. Solids, 49:2113–2132, 2001.
166. F.F. Abraham. Dynamics of brittle fracture with variable elasticity. Phys. Rev.
Lett., 77(5):869, 1996.
167. D.H. Boal. Mechanics of the cell. Cambridge University Press, New York, 2002.
168. D.I. Bower. An Introduction to Polymer Physics. Cambridge University Press,
New York, 2002.
169. C. Bustamante, J.F. Marko, E.D. Siggia, and S. Smith. Entropic elasticity of
lambda-phage DNA. Science, 265(5178):1599–1600, 1994.
170. J.F. Marko and E.D. Siggia. Stretching DNA. Macromolecules, 28(26):8759–
8770, 1995.
171. S. Kohlhoff, P. Gumbsch, and H.F. Fischmeister. Crack propagation in b.c.c.
crystals studied with a combined finite-element and atomistic model. Phil.
Mag. A, 64:851–878, 1991.
172. P. Gumbsch. Brittle fracture processes modelled on the atomic scale. Z.
Metallkd., 87:341–348, 1996.
173. V. Bulatov, F.F. Abraham, L. Kubin, B. Devincre, and S. Yip. Connecting
atomistic and mesoscale simulations of crystal plasticity. Nature, 391:669–672,
1998.
174. F.F. Abraham, J.Q. Broughton, N. Bernstein, and E. Kaxiras. Spanning the
length scales in dynamic simulation. Comput. Phys., 12(6):538–546, 1998.
472 References
192. L.E. Shilkrot, R.E. Miller, and W.A. Curtin. Multiscale plasticity modeling:
coupled atomistics and discrete dislocation mechanics. J. Mech. Phys. Solids,
52(4):755–787, 2004.
193. T.D. Nguyen, S. Govindjee, P.A. Klein, and H. Gao. A rate-dependent cohesive
continuum model for the study of crack dynamics. Comput. Meth. Appl. Mech.
Eng., 193(30–32):3239–3265, 2003.
194. F.H. Seitz and J.W. Mintmire. Electrostatic potentials for metal-oxide surfaces
and interfaces. Phys. Rev. B, 50:11996–12003, 1994.
195. X.W. Zhou, H.N.G. Adley, J.S. Filhol, and M.N. Neurock. Modified charge
transfer-embedded atom method potential for metal/metal oxide systems.
Phys. Rev. B, 69:035492, 2004.
196. S.G. Parker, C.R. Johnson, and D. Beazley. Computational steering software
systems and strategies. IEEE Comput. Sci. Eng., 4(4):50–599, 1997.
197. M.J. Buehler, J. Dodson, P. Meulbroek, A. Duin, and W.A. Goddard. The
computational materials design facility (CMDF): A powerful framework for
multiparadigm multi-scale simulations. Mat. Res. Soc. Proc., 894:LL3.8, 2006.
198. T.C. Germann and A.F. Voter. Accelerating molecular dynamics simulations.
Proc. Comput. Nanosci. Nanotech., ISBN 0-9708275-6-3:140–143, 2002.
199. A. Laioa and M. Parrinello. Escaping free-energy minima. P. Natl. Acad. Sci.
USA, 99:12562–12566, 2002.
200. A.F. Voter. Parallel replica method for dynamics of infrequent events. Phys.
Rev. B, 57(22):985–988, 1998.
201. F. Montalenti, A.F. Voter, and R. Ferrando. Spontaneous atomic shuffle in flat
terraces: Ag(100). Phys. Rev. B, 66:205404, 2002.
202. A.F. Voter. Hyperdynamics: Accelerated molecular dynamics of infrequent
events. Phys. Rev. Lett., 78(20):3908–3911, 1997.
203. F. Montalenti, M.R. Sorensen, and A.F. Voter. Closing the gap between exper-
iment and theory: Crystal growth by temperature accelerated dynamics. Phys.
Rev. Lett., 87(12):126101, 2001.
204. I. Kaur and W. Gust. Handbook of Grain and Interface Boundary Diffusion
Data. Ziegler Press, Stuttgart, 1989.
205. M.L. Falk, A. Needleman, and J. Rice. A critical evaluation of cohesive zone
models of dynamic fracture. J. Phys. IV Fr., 11:43–50, 2001.
206. P.A Klein, J.W. Foulk, E.P. Chen, S.A. Wimmer, and H.J. Gao. Physics-
based modeling of brittle fracture: Cohesive formulations and the application
of meshfree methods. Theor. Appl. Fract. Mech., 37:99–166, 2001.
207. J. Fineberg, S.P. Gross, M. Marder, and H.L. Swinney. Instability and dynamic
fracture. Phys. Rev. Lett., 67(4):457–460, 1991.
208. Y. Huang and H. Gao. Intersonic crack propagation–Part II: The suddenly
stopping crack. J. Appl. Mech., 69:76–79, 2002.
209. D. Gross and T. Selig. Fracture Mechanics. Springer Verlag, Heidelberg, 2006.
210. E.H. Yoffe. The moving griffith crack. Phil. Mag., 42:739–750, 1951.
211. B.R. Baker. Dynamic stresses created by a moving crack. J. Appl. Mech.,
29:567–578, 1962.
212. H. Tada and P.C. Paris. The Stress Analysis of Cracks Handbook. ASME Press,
New York, 2001.
474 References
213. W.T. Ashurst and W.G. Hoover. Microscopic fracture studies in 2-dimensional
triangular lattice. Phys. Rev. B, 14(4):1465–1473, 1976.
214. M. Marder and S. Gross. Origin of crack tip instabilities. J. Mech. Phys. Solids,
43(1):1–48, 1995.
215. A.J. Rosakis, O. Samudrala, and D. Coker. Cracks faster than the shear wave
speed. Science, 284(5418):1337–1340, 1999.
216. A.J. Rosakis. Intersonic shear cracks and fault ruptures. Adv. Phys., 51(4):
1189–1257, 2002.
217. Y. Huang and H. Gao. Intersonic crack propagation–Part I: The fundamental
solution. J. Appl. Mech., 68:169–175, 2001.
218. E. Gerde and M. Marder. Friction and fracture. Nature, 413:285, 2001.
219. F.F. Abraham, R. Walkup, H. Gao, M. Duchaineau, T.D. de La Rubia, and
M. Seager. Simulating materials failure by using up to one billion atoms and the
world’s fastest computer: Brittle fracture. P. Natl. Acad. Sci. USA, 99(9):5788–
5792, 2002.
220. F.F. Abraham. How fast can cracks move? A research adventure in materials
failure using millions of atoms and big computers. Adv. Phys., 52(8):727–790,
2003.
221. M. Marder. Molecular dynamics of cracks. Comput. Sci. Eng., 1(5):48–55, 1999.
222. D. Holland and M. Marder. Cracks and atoms. Adv. Mater., 11(10):793, 1999.
223. D. Holland and M. Marder. Ideal brittle fracture of silicon studied with
molecular dynamics. Phys. Rev. Lett., 80(4):746, 1998.
224. J.A. Hauch, D. Holland, M. Marder, and H.L. Swinney. Dynamic fracture in
single crystal silicon. Phys. Rev. Lett., 82:3823–3826, 1999.
225. P. Vashishta, R.K. Kalia, and A. Nakano. Large-scale atomistic simulations of
dynamic fracture. Comp. Sci. Eng., 1(5):56–65, 1999.
226. R. Mikulla, J. Stadler, F. Krul, H.-R. Trebin, and P. Gumbsch. Crack
propagation in quasicrystals. Phys. Rev. Lett., 81(15):3163–3166, 1998.
227. H.-R. Trebin, R. Mikulla, J. Stadler, G. Schaaf, and P. Gumbsch. Molecular
dynamics simulations of crack propagation in quasicrystals. Comput. Phys.
Commn., 121–122:536–539, 1999.
228. C. Kittel. Einführung in die Festkörperphysik. Oldenburg, München, 2nd
edition, 1999.
229. R. Mikulla, J. Roth, and H.-R. Trebin. Simulation of shear-stress in 2-
dimensional decagonal quasi-crystals. Phil. Mag. B, 71(5), 1995.
230. G.F. Guo, W. Yang, and Y. Huang. Supersonic crack growth in a solid of
upturn stress-strain relation under anti-plane shear. J. Mech. Phys. Solids,
51:1971–1985, 2003.
231. K. Hellan. Introduction to Fracture Mechanics. McGraw-Hill, New York, 1984.
232. Hellan. Debond dynamics of an elastic strip, I: Timoshenko-beam properties
and steady motion. Int. J. Fract., 14(1):91–100, 1978.
233. L.B. Freund. A simple model of the double cantilever beam crack proapgation
specimen. J. Mech. Phys. Solids, 25:69–79, 1977.
234. L.B. Freund. Crack propagation in an elastic solid subjected to general loading,
II. nonuniform rate of extension. J. Mech. Phys. Solids, 20:141–152, 1972.
235. S. Fratini, O. Pla, P. Gonzalez, F. Guinea, and E. Louis. Energy radiation of
moving cracks. Phys. Rev. B, 66(10):104104, 2002.
References 475
259. R. Burridge. Admissible speeds for plain-strain self-similar cracks with friction
but lacking cohesion. Geophys. J. R. Astron. Soc., 35:439–455, 1973.
260. J.R. Rice and G.C. Sih. Plane problems of cracks in dissimilar media. Trans.
ASME, 32(2):418–423, 1965.
261. J.R. Rice. Elastic fracture mechanics concepts for interfacial cracks. Trans.
ASME, 55(1):98–103, 1988.
262. A.H. England. A crack between dissimilar media. J. Appl. Mech., 32:400–402,
1965.
263. Williams M.L. The stresses areound a fault or crack in dissimilar media. Bull.
Seismol. Soc. Am., 49:199–204, 1959.
264. W. Yang, Z. Suo, and C.F. Shih. Mechanics of dynamic debonding. Proc. R.
Soc. Lond. A, 433:679–697, 1991.
265. C. Liu, J. Lambros, and A.J. Rosakis. Highly transient elastodynamic crack
growth in a bimaterial interface: higher order asymptotic analysis and experi-
ments. J. Mech. Phys. Solids, 41:1887–1954, 1993.
266. J. Lambros and A.J. Rosakis. Development of a dynamic decohesion criterion
for subsonic fracture of the interface between two dissimilar materials. Proc.
R. Soc. Lond. A, 41:711–736, 1995.
267. A.J. Rosakis. Intersonic crack propagation in bimaterial systems. J. Mech.
Phys. Solids, 6(10):1789–1813, 1998.
268. C. Liu, Y. Huang, and A.J. Rosakis. Shear dominated transonic interfacial
crack growth in a bimaterial–II. Asymptotic fields and favorable velocity
regimes. J. Mech. Phys. Solids, 43(2):189–206, 1993.
269. S. Chen. Personal communication.
270. H.F. Zhang and A.H.W. Ngan. Atomistic simulation of screw dislocation
mobility ahead of a mode III crack tip in the bcc structure. Scripta Mater.,
41(7):737–742, 2002.
271. G. Schoeck. Dislocation emission from crack tips as a variational problem of
the crack energy. J. Mech. Phys. Solids, 44(3):413–437, 1996.
272. F.F. Abraham and H. Gao. Anamalous brittle-ductile fracture behaviors in
FCC crystals. Phil. Mag. Lett., 78:307–312, 1998.
273. S. Chen, M.J. Buehler, and H. Gao. The mode III broberg problem: Supersonic
cracking in a stiff strip. unpublished results.
274. M.J. Buehler, A.C.T. van Duin, and W.A. Goddard. Multi-paradigm modeling
of dynamical crack propagation in silicon using the reaxff reactive force field.
Phys. Rev. Lett., 96(9):095505, 2006.
275. M.J. Buehler, H. Tang, A.C.T. van Duin, and W.A. Goddard. Threshold crack
speed controls dynamical fracture of silicon single crystals. Phys. Rev. Lett.,
99:165502, 2007.
276. T. Cramer, A. Wanner, and P. Gumbsch. Crack velocities during dynamic
fracture of glass and single crystalline silicon. Phys. Status Solidi A, 164:R5,
1997.
277. F. Ebrahimi and L. Kalwani. Fracture anisotropy in silicon single crystal.
Mater. Sci. Eng. A, 268:116–126, 1999.
278. R.D. Deegan, S. Chheda, L. Patel, M. Marder, H.L. Swinney, J. Kim, and
A. de Lozanne. Wavy and rough cracks in silicon. Phys. Rev. E, 67(6):066209,
2003.
279. N.P. Bailey and J.P. Sethna. Macroscopic measure of the cohesive length scale:
Fracture of notched single-crystal silicon. Phys. Rev. B, 68(20):205204, 2003.
References 477
300. G.D. Schaaf, J. Roth, and H.-R. Trebin. Dislocation motion in icosahedral
quasicrystals at elevated temperatures: Numerical simulation. Phil. Mag.,
83(21):2449–2465, 2003.
301. M.J. Buehler, A. Hartmaier, M.A. Duchaineau, F.R. Abraham, and H.J. Gao.
The dynamical complexity of work-hardening: a large-scale molecular dynamics
simulation. Acta Mechanica Sinica, 21(2):103–111, 2005.
302. P. Gumbsch and H. Gao. Dislocations faster than the speed of sound. Science,
283:965–968, 1999.
303. D. Wolf, V. Yamakov, S.R. Phillpot, and A.K. Mukherjee. Deformation mecha-
nism and inverse Hall-Petch behavior in nanocrystalline materials. Z. Metallk.,
94:1052–1061, 2003.
304. V. Yamakov, D. Wolf, S.R. Phillpot, A.K. Mukherjee, and H. Gleiter. Dis-
location processes in the deformation of nanocrystalline aluminium by MD
simulation. Nat. Mater., 1:1–4, 2002.
305. K.W. Jacobsen and J. Schiotz. Computational materials science - nanoscale
plasticity. Nat. Mater., 1:15–16, 2002.
306. E. Arzt. Size effects in materials due to microstructural and dimensional
constraints: A comparative review. Acta Mater., 46:5611–5626, 1998.
307. R. Krause-Rehberg, M. Brohl, H. Leipner, A. Kupsch, T. Drost, A. Polity,
U. Beyer, and H. Alexander. Defects in plastically deformed semiconduc-
tors studied by positron annihilation: Silicon and germanium. Phys. Rev. B,
47:13266, 1993.
308. A.A. Rempel, S.Z. Nazarova, and A.I. Gusev. Intrinsic defects in palladium
after severe plastic deformation. Phys. Stat. Sol., 181:R16, 2000.
309. S.J. Zhou, D.L. Preston, and F. Louchet. Investigation of vacancy formation
by a jogged dissociated dislocation with large-scale molecular dynamics and
dislocation energetics. Acta Mat., 47:2695–2703, 1999.
310. J. Marian, W. Cai, and V.V. Bulatov. Dynamic transitions from smooth to
rough to twinning in dislocation motion. Nat. Mater., 3:158–163, 2004.
311. V. Yamakov, D. Wolf, S.R. Phillpot, A.K. Mukherjee, and H. Gleiter.
Deformation-mechanism map for nanocrystalline metals by molecular-
dynamics simulation. Nat. Mater., 3(1):43–47, 2004.
312. T. Vegge. Atomistic simulations of screw dislocation cross slip in copper and
nickel. Mat. Sci. Eng. A-Struct., 309:113, 2001.
313. T. Vegge and W. Jacobsen. Atomistic simulations of dislocation processes in
copper. J. Phys. Condens. Matt., 14:2929, 2002.
314. M. Wen and D.L. Lin. Atomistic process of dislocation cross-slip in ni3al.
Scripta Mater., 36:265, 1997.
315. V.B. Shenoy, R.V. Kukta, and R. Phillips. Quasicontinuum models of interfa-
cial structure and deformation. Phys. Rev. Lett., 84:1491, 2000.
316. N.F. Mott. The work hardening of metals. Trans. Met. Soc. AIME, 218:962,
1960.
317. M. Rhee, H.M. Zbib, J.P. Hirth, H. Huang, and T.D. de la Rubia. Models for
long-/short-range interactions and cross slip in 3d dislocation simulation of bcc
single crystals. Modeling Simul. Mater. Sci. Eng., 6:467, 1998.
318. P.B. Hirsch and D.H. Warrington. Athe flow stress of aluminium and copper
at high temperatures. Phil. Mag., 6:735, 1961.
319. F.R.N. Nabarro. In: Report of a conference on strength of solids, pages 75–90.
London: Phys. Soc., 1948.
References 479
320. R.L. Coble. A model for boundary controlled creep in polycrystalline materials.
J. Appl. Phys., 41:1679–1682, 1963.
321. T.G. Nieh and J. Wadsworth. Hall-Petch relation in nanocrystalline solids.
Scripta Met., 25(4):955–958, 1991.
322. S. Yip. The strongest size. Nature, 391:532–533, 1998.
323. K.S. Kumar, H. Van Swygenhoven, and S. Suresh. Mechanical behavior of
nanocrystalline metals and alloys. Acta Mater., 51(10):5743–5774, 2003.
324. J. Schiotz, D. DiTolla, and K.W. Yacobsen. Softening of nanocrystalline metals
at very small grain sizes. Nature, 391:561–563, 1998.
325. V. Yamakov, D. Wolf, M. Salazar, S.R. Phillpot, and H. Gleiter. Length scale
effects in the nucleation of extended dislocations in nanocrystalline Al by
molecular-dynamics simulations. Acta Mater., 49:2713–2722, 2001.
326. R. Würschum, S. Herth, and U. Brossmann. Diffusion in nanocrystalline metals
and alloys–a status report. Adv. Eng. Mat., 5(5):365–372, 2003.
327. J. Schiotz, T. Vegge, D. DiTolla, and K.W. Yacobsen. Atomic-scale simula-
tions of the mechanical deformation of nanocrystalline metals. Phys. Rev. B,
60:11971–11983, 1999.
328. H. van Swygenhoven and A. Caro. Plastic behavior of nanophase metals studied
by molecular dynamics. Phys. Rev. B, 58:11246–11251, 1998.
329. H. van Swygenhoven, M. Spaczer, A. Caro, and D. Farkas. Competing plastic
deformation mechanisms in nanophase metals. Phys. Rev. B, 60:22–25, 1999.
330. H. van Swygenhoven, M. Spaczer, and A. Caro. Microscopic description of
plasticity in computer generated metallic nanophase samples: A comparison
between Cu and Ni. Acata Mat., 47:3117, 1999.
331. A. Hasnaoui, H. van Swygenhoven, and P.M. Derlet. Dimples on nanocrys-
talline fracture surfaces as evidence for shear plane formation. Science,
300(5625):1550–1552, 2003.
332. M.W. Chen, E. Ma, K.J. Hemker, Y. Wang H. Sheng, and C. Xuemei.
Deformation twinning in nanocrystalline aluminum. Science, 300:1275–1277,
2003.
333. H. Ikeda, Y. Qi, T. Cagin, K. Samwer, W. L. Johnson, and W.A. Goddard.
Strain rate induced amorphization in metallic nanowires. Phys. Rev. Lett.,
82(14):2900–2903, 1999.
334. G. Rubio-Bollinger, S.R. Bahn, N. Agrait, K.W. Jacobsen, and S. Vieira.
Mechanical properties and formation mechanisms of a wire of single gold atoms.
Phys. Rev. Lett., 87(2):026101, 2001.
335. A.M. Leach, M. McDowell, and K. Gall. Deformation of top-down and bottom-
up silver nanowires. Adv. Funct. Mater., 17(1):43–53, 2007.
336. E.Z. da Silva, A.J.R. da Silva, and A. Fazzio. How do gold nanowires break?
Phys. Rev. Lett., 87(25):256102, 2001.
337. J.K. Diao, K. Gall, M.L. Dunn, and J.A. Zimmerman. Atomistic simulations
of the yielding of gold nanowires. Acta Mater., 54(3):643–653, 2006.
338. K. Gall, J.K. Diao, and M.L. Dunn. The strength of gold nanowires. Nano
Lett., 4(12):2431–2436, 2004.
339. P. Heino, H. Häkkinen, and K. Kaski. Molecular-dynamics study of mechanical
properties of copper. Europhys. Lett., 41:273–278, 1998.
340. M.D. Uchic, D.M. Dimiduk, J.N. Florando, and W.D. Nix. Sample dimensions
influence strength and crystal plasticity. Science, 305:987–989, 2004.
480 References
341. R.P. Vinci, E.M. Zielinski, and J.C. Bravman. Thermal strain and stress in
copper thin films. Thin Solid Films, 262:142–153, 1995.
342. M.J. Kobrinsky and C.V. Thompson. The thickness dependence of the flow
stress of capped and uncapped polycrystalline Ag thin films. Appl. Phys. Lett.,
73:2429–2431, 1998.
343. R.-M. Keller, S.P. Baker, and E. Arzt. Stress-temperature behavior of unpas-
sivated thin copper films. Acta Mater., 47(2):415–426, 1999.
344. M.J. Kobrinsky, G. Dehm, C.V. Thompson, and E. Arzt. Effects of thickness
on the characteristic length scale of dislocation plasticity in Ag thin films. Acta
Mater., 49:3597, 2001.
345. W.D. Nix. Yielding and strain hardening of thin metal films on substrates.
Scripta Mater., 39:545–554, 1998.
346. G. Dehm, T.J. Balk, B. von Blanckenhagen, P. Gumbsch, and E. Arzt. Dislo-
cation dynamics in sub-micron confinement: Recent progress in Cu thin film
plasticity. Z. Metallk., 93(5):383–391, 2002.
347. M. Legros, K.J. Hemker, A. Gouldstone, S. Suresh, R.-M. Keller-Flaig, and
E. Arzt. Microstructural evolution in passivated Al films on Si substrates
during thermal cycling. Acta Mater., 50:3435–3452, 2002.
348. B. von Blanckenhagen, P. Gumbsch, and E. Arzt. Dislocation sources and the
flow stress of polycrystalline thin metal films. Phil. Mag. Lett., 83:1–8, 2003.
349. L. Nicola, E. Van der Giessen, and A. Needleman. Discrete dislocation analysis
of size effects in thin films. Thin Solid Films, 479:329–338, 2005.
350. T.J. Balk, G. Dehm, and E. Arzt. Observations of dislocation motion and stress
inhomogeneities in thin copper films. Mat. Res. Soc. Symp. Proc., 673:2.7.1–
2.7.6, 2001.
351. D. Weiss, H. Gao, and E. Arzt. Constrained diffusional creep in UHV-produced
copper thin films. Acta Mater., 49:2395–2403, 2001.
352. Z. Suo and J.W. Hutchinson. Steady-state cracking in brittle substrates
beneath adherent films. Int. J. Solids Struct., 25(11):1337–1353, 1989.
353. J.D. Eshelby. In: Dislocations in Solids, volume 1, Ed. F.R.N. Nabarro, page
167. North-Holland, Amsterdam, 1979.
354. A.R. Zak and M.L. Williams. Crack point singularities at a bi-material
interface. J. Appl. Mech., 30:142–143, 1963.
355. P. Keblinski, D. Wolf, S.R. Phillpot, and H. Gleiter. Continuous
thermodynamic-equilibrium glass transition in high energy grain boundaries?
Phil. Mag. Lett., 76(3):143–151, 1997.
356. P. Keblinski, D. Wolf, S.R. Phillpot, and H. Gleiter. Self-diffusion in high-angle
fcc metal grain boundaries by molecular dynamics simulation. Phil. Mag. Lett.,
79(11):2735–2761, 1999.
357. P. Lagarde and M. Biscondi. Intercrystalline creep of oriented copper bicrystals.
Mem. Etud. Sci. Rev. Met., 71:121–131, 1974.
358. E. Budke, C. Herzig, S. Prokofjev, and L. Shvindlerman. Study of grain-
boundary diffusion of Au in copper within Σ5 misorientation range in the con-
text of structure of grain boundaries. Defect and Diffusion Forum, 156:21–33,
1998.
359. W.C. Swope, H.C. Andersen, P.H. Berens, and K.R. Wilson. A computer-
simulation method for the calculation of equilibrium-constants for the forma-
tion of physical clusters of molecules - application to small water clusters.
J. Chem. Phys., 76(1):637–649, 1982.
References 481
360. R.E. Peierls. The size of a dislocation. Proc. Pys. Soc., 52:256, 1940.
361. Y.-L. Shen. Strength and interface-constrained plasticity in thin metal films.
J. Mater. Res., 18:2281–2284, 2003.
362. L. Lu, Y. Shen, X. Chen, L. Qian, and K. Lu. Ultrahigh strength and high
electrical conductivity in copper. Science, 304:422–426, 2004.
363. S. Iijima. Helical microtubules of graphitic carbon. Nature, 354(6348):56–58,
1991.
364. A. Modi, N. Koratkar, E. Lass, B.Q. Wei, and P.M. Ajayan. Miniaturized gas
ionization sensors using carbon nanotubes. Nature, 424(6945):171–174, 2003.
365. J.Y. Huang, S. Chen, Z.Q. Wang, K. Kempa, Y.M. Wang, S.H. Jo, G. Chen,
M.S. Dresselhaus, and Z.F. Ren. Superplastic carbon nanotubes - conditions
have been discovered that allow extensive deformation of rigid single-walled
nanotubes. Nature, 439(7074):281–281, 2006.
366. S.E. Moulton, A.I. Minett, and G.G. Wallace. Carbon nanotube based
electronic and electrochemical sensors. Sensor Lett., 3(3):183–193, 2005.
367. V. Sazonova, Y. Yaish, H. Ustunel, D. Roundy, T.A. Arias, and P.L. McEuen. A
tunable carbon nanotube electromechanical oscillator. Nature, 431(7006):284–
287, 2004.
368. W.D. Zhang, F. Yang, and P.Y. Gu. Carbon nanotubes grow to pillars.
Nanotechnology, 16(10):2442–2445, 2005.
369. B.I. Yakobson, C.J. Brabec, and J. Bernholc. Nanomechanics of carbon tubes:
Instabilities beyond linear response. Phys. Rev. Lett., 76(14):2511–2514, 1996.
370. M.J. Buehler, Y. Kong, and H. Gao. Deformation mechanisms of very long
single-wall carbon nanotubes subject to compressive loading. J. Eng. Mater.
Tech., 126(3):245–249, 2004.
371. M.J. Buehler, Y. Kong, Huang Y., and H.J. Gao. Self-folding and unfolding of
carbon nanotubes. J. Eng. Mater. Techno., 128(1):3–10, 2006.
372. W. Zhou, Y. Huang, B. Liu, K.C. Hwang, J.M. Zuo, M.J. Buehler, and H. Gao.
Self-folding of single- and multiwall carbon nanotubes. Applied Physics Letters,
90:3107–3110, 2007.
373. H.J. Gao and S.H. Chen. Flaw tolerance in a thin strip under tension. J. Appl.
Mech.-Trans. ASME, 72(5):732–737, 2005.
374. H.J. Gao. Application of fracture mechanics concepts to hierarchical biome-
chanics of bone and bone-like materials. Int. J. Fract., 138(1–4):101–137,
2006.
375. M.J. Buehler, H. Yao, H. Gao, and B. Ji. Cracking and adhesion at small scales:
Atomistic and continuum studies of flaw tolerant nanostructures. Model. Sim.
Mat. Sci. Eng., 14:799–816, 2006.
376. H. Gao and H. Yao. Shape insensitive optimal adhesion of nanoscale fibrillar
structures. P. Natl. Acad. Sci. USA, 101:7851–7856, 2004.
377. H. Gao, B. Ji, M.J. Buehler, and H. Yao. Flaw tolerant bulk and surface
nanostructures of biological systems. Mech. Chem. Biosyst., 1(1):37–52, 2004.
378. H. Gao, X. Wang, H. Yao, S. Gorb, and E. Arzt. Mechanics of hierarchical
adhesion structures of geckos. Mech. Mat., 37(2–3):275–285, 2005.
379. K.L. Johnson, K. Kendall, and A.D. Roberts. Surface energy and contact of
elastic solids. Proc. R. Soc. A, 324:301–313, 1971.
Index
N 2 scaling, 72 mechanical, 90
µVT, 40 steered molecular dynamics, 92
Brittle failure, 12
Abell-Tersoff approach, 60 Brittle versus ductile, 12
Adhesion, 439, 452, 453 Brittle-to-ductile transition, 342
Adhesion strength, 459 Buffer region, 171
Advanced molecular dynamics methods,
177
CADD, 167
Aluminum
Canonical ensemble, 40
nanocrystalline, 379
Carbon nanotubes, 438
AMBER, 56
Catalysis, 66
Analysis techniques, 410
Cauchy relation, 55
Asymptotic stress field, 194
Centrosymmetry parameter, 86
Atomic hypothesis, 35
Centrosymmetry technique, 410
Atomic interactions, 35
Charge equilibration, 63
Atomistic simulations, 33
Atomistic theory, 16, 121 CHARMM, 56
Averaging, 36 Chemical bonding, 46
Avogadro’s number, 79, 155 Chemical complexity, 56, 62, 63, 304,
359
Barrier Classical molecular dynamics, 37
dislocation motion, 424 CMDF, 173
BCC, 9 Coble creep, 378
bcc packing, 13 Cohesive zone, 446
Beam elasticity, 110 Common neighbor analysis, 90
Berendsen thermostat, 40 Computational efficiency, 54
Billion-atom simulation, 341 Computational Materials Design
Bimaterial interface, 287 Facility
Biological materials, 56 CMDF, 173
Biomechanics, 9, 452 Computer experiments, 35
Bond order potentials, 59, 63 Computer power, 79
Bookkeeping, 73 Computing power, historical
Boundary conditions, 90, 399 development, 79
displacement, 90 Concurrent multiscale modeling, 162
484 Index