Buehler

Atomistic Modeling of Materials Failure
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
ISBN 978-0-387-76425-2 e-ISBN 978-0-387-76426-9

DOI: 10.1007/978-0-387-76426-9
Library of Congress Control Number: 2008927204
c 2008 Springer Science+Business Media, LLC
All rights reserved. This work may not be translated or copied in whole or in part without the written
permission of the publisher (Springer Science+Business Media, LLC, 233 Spring Street, New York, NY
10013, USA), except for brief excerpts in connection with reviews or scholarly analysis. Use in connection
with any form of information storage and retrieval, electronic adaptation, computer software, or by similar
or dissimilar methodology now known or hereafter developed is forbidden.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if they are
not identified as such, is not to be taken as an expression of opinion as to whether or not they are subject
to proprietary rights.
Printed on acid-free paper
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.
Cambridge, MA Markus J. Buehler

July 14, 2008
Contents
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XVII
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LXIII
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
Part II Basics of Atomistic, Continuum and Multiscale Methods
2 Basic Atomistic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
2.2 Modeling and Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.1 Model Building and Physical Representation . . . . . . . . 34
X Contents
2.2.2 The Concept of Computational Experiments . . . . . . . . 35

2.3 Basic Statistical Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4 Formulation of Classical Molecular Dynamics . . . . . . . . . . . . . 37
2.4.1 Integrating the Equations of Motion . . . . . . . . . . . . . . . 39
2.4.2 Thermodynamic Ensembles and Their Numerical
Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
2.4.3 Energy Minimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.4.4 Monte Carlo Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.5 Classes of Chemical Bonding . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.6 The Interatomic Potential or Force Field: Introduction . . . . . 48
2.6.1 Pair Potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2.6.2 Multibody Potentials: Embedded Atom Method
for Metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.6.3 Force Fields for Biological Materials and Polymers . . . 56
2.6.4 Bond Order and Reactive Potentials . . . . . . . . . . . . . . . 59
2.6.5 Limitations of Classical Molecular Dynamics . . . . . . . . 68
2.7 Numerical Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.7.1 Periodic Boundary Conditions . . . . . . . . . . . . . . . . . . . . 70
2.7.2 Force Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
2.7.3 Neighbor Lists and Bins . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.8 Property Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.8.1 Temperature Calculation . . . . . . . . . . . . . . . . . . . . . . . . . 73
2.8.2 Pressure Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.8.3 Radial Distribution Function . . . . . . . . . . . . . . . . . . . . . . 74
2.8.4 Mean Square Displacement Function . . . . . . . . . . . . . . . 75
2.8.5 Velocity Autocorrelation Function . . . . . . . . . . . . . . . . . 76
2.8.6 Virial Stress Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76
2.9 Large-Scale Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
2.9.1 Historical Development of Computing Power . . . . . . . . 79
2.9.2 Parallel Computing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
2.9.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.10 Visualization and Analysis Methods . . . . . . . . . . . . . . . . . . . . . 83
2.10.1 Energy Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.10.2 Centrosymmetry Parameter . . . . . . . . . . . . . . . . . . . . . . . 86
2.10.3 Slip Vector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
2.10.4 Measurement of Defect Speed . . . . . . . . . . . . . . . . . . . . . 89
2.10.5 Visualization Methods for Biological Structures . . . . . 89
2.10.6 Other Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
2.11 Distinguishing Modeling and Simulation . . . . . . . . . . . . . . . . . . 90
2.12 Application of Mechanical Boundary Conditions . . . . . . . . . . . 90
2.13 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Contents XI
3 Basic Continuum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

3.1 Newton’s Laws of Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
3.2 Definition of Displacement, Stress, and Strain . . . . . . . . . . . . . 97
3.2.1 Stress Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
3.2.2 Equilibrium Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.2.3 Strain Tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.3 Energy Approach to Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.4 Isotropic Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.5 Nonlinear Elasticity or Hyperelasticity . . . . . . . . . . . . . . . . . . . 108
3.6 Elasticity of a Beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.6.1 Reduction Formulas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
3.6.2 Equilibrium Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
3.6.3 Example: Solution of a Simple Beam Problem . . . . . . . 112
3.6.4 Calculation of Internal Stress Field . . . . . . . . . . . . . . . . 113
3.6.5 Differential Beam Equations . . . . . . . . . . . . . . . . . . . . . . 116
3.7 The Need for Atomistic Elasticity: What’s Next . . . . . . . . . . . 119
4 Atomistic Elasticity: Linking Atoms and Continuum . . . . . 121

4.1 Thermodynamics as Bridge Between Atomistic
and Continuum Viewpoints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.2 The Atomic and Molecular Origin of Elasticity:
Entropic vs. Energetic Sources . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.3 The Virial Stress and Strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.4 Elasticity Due to Energetic Contributions . . . . . . . . . . . . . . . . 124
4.4.1 Cauchy–Born Rule . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.4.2 Elasticity of a One-Dimensional String of Atoms . . . . 126
4.4.3 Elasticity and Surface Energy
of a Two-Dimensional Triangular Lattice . . . . . . . . . . . 128
4.4.4 Elasticity and Surface Energy
of a Three-Dimensional FCC Lattice . . . . . . . . . . . . . . . 142
4.4.5 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.5 Elasticity Due to Entropic Contributions . . . . . . . . . . . . . . . . . 149
4.5.1 Elasticity of Single Molecules: Worm-Like-Chain
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.5.2 Elasticity of Polymers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
4.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5 Multiscale Modeling and Simulation Methods . . . . . . . . . . . 157

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.2 Direct Numerical Simulation vs. Multiscale
and Multiparadigm Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 158
5.3 Differential Multiscale Modeling . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.4 Detailed Description of Selected Multiscale Methods
to Span Vast Lengthscales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
5.4.1 Examples of Hierarchical Multiscale Coupling . . . . . . . 160
XII Contents
5.4.2 Concurrent Integration of Tight-Binding,

Empirical Force Fields and Continuum Theory . . . . . . 162
5.4.3 The Quasicontinuum Method
and Related Approaches . . . . . . . . . . . . . . . . . . . . . . . . . . 165
5.4.4 Continuum Approaches Incorporating Atomistic
Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.4.5 Hybrid ReaxFF Model: Integration of Chemistry
and Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
5.5 Advanced Molecular Dynamics Techniques to Span Vast
Timescales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
Part III Material Deformation and Failure
6 Deformation and Dynamical Failure of Brittle Materials . 185

6.1 The Nature of Brittle Fracture . . . . . . . . . . . . . . . . . . . . . . . . . . 186
6.2 Basics of Linear Elastic Fracture Mechanics . . . . . . . . . . . . . . . 189
6.2.1 Energy Balance Considerations: Griffith’s Model
of Fracture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
6.2.2 Asymptotic Stress Field and Stress Intensity Factor . . 194
6.2.3 Crack Limiting Speed in Dynamic Fracture . . . . . . . . . 196
6.3 Atomistic Modeling of Brittle Materials . . . . . . . . . . . . . . . . . . 197
6.4 A One-Dimensional Example of Brittle Fracture:
Joint Continuum-Atomistic Approach . . . . . . . . . . . . . . . . . . . . 201
6.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202
6.4.2 Linear-Elastic Continuum Model . . . . . . . . . . . . . . . . . . 204
6.4.3 Hyperelastic Continuum Mechanics Model
for Bilinear Stress–Strain Law . . . . . . . . . . . . . . . . . . . . . 207
6.4.4 Molecular Dynamics Simulations
of the One-Dimensional Crack Model:
The Harmonic Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
6.4.5 Molecular Dynamics Simulations
of the One-Dimensional Crack Model:
The Supersonic Case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
6.4.6 Discussion and Conclusions . . . . . . . . . . . . . . . . . . . . . . . 219
6.5 Stress and Deformation Field near Rapidly Propagating
Mode I Cracks in a Harmonic Lattice . . . . . . . . . . . . . . . . . . . . 223
6.5.1 Stress and Deformation Fields . . . . . . . . . . . . . . . . . . . . 225
6.5.2 Energy Flow near the Crack Tip . . . . . . . . . . . . . . . . . . 227
6.5.3 Limiting Velocities of Mode I Cracks
in Harmonic Lattices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229
6.5.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 230
Contents XIII
6.6 Crack Limiting Speeds of Cracks: The Significance

of Hyperelasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
6.6.1 Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
6.6.2 Crack Speed and Energy Flow . . . . . . . . . . . . . . . . . . . . 238
6.6.3 Hyperelastic Area . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 239
6.6.4 How Fast can Cracks Propagate? . . . . . . . . . . . . . . . . . . 242
6.6.5 Characteristic Energy Length Scale in Dynamic
Fracture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
6.6.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
6.7 Crack Instabilities and Hyperelastic Material Behavior . . . . . 249
6.7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 251
6.7.2 Design of Computational Model . . . . . . . . . . . . . . . . . . . 252
6.7.3 Computational Experiments . . . . . . . . . . . . . . . . . . . . . . 255
6.7.4 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . 258
6.8 Suddenly Stopping Cracks: Linking Atomistic Modeling,
Theory, and Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
6.8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
6.8.2 Theoretical Background of Suddenly
Stopping Cracks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 262
6.8.3 Atomistic Simulation Setup . . . . . . . . . . . . . . . . . . . . . . . 264
6.8.4 Atomistic Simulation Results of a Suddenly
Stopping Mode I Crack . . . . . . . . . . . . . . . . . . . . . . . . . . 268
6.8.5 Atomistic Simulation Results of a Suddenly
Stopping Mode II Crack . . . . . . . . . . . . . . . . . . . . . . . . . . 278
6.8.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 286
6.9 Crack Propagation Along Interfaces of Dissimilar
Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 287
6.9.1 Mode I Dominated Cracks at Bimaterial Interfaces . . 289
6.9.2 Mode II Cracks at Bimaterial Interfaces . . . . . . . . . . . . 294
6.9.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297
6.10 Dynamic Fracture Under Mode III Loading . . . . . . . . . . . . . . . 299
6.10.1 Atomistic Modeling of Mode III Cracks . . . . . . . . . . . . 300
6.10.2 Mode III Cracks in a Harmonic Lattice –
The Reference Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
6.10.3 Mode III Crack Propagation in a Thin Stiff
Layer Embedded in a Soft Matrix . . . . . . . . . . . . . . . . . 301
6.10.4 Suddenly Stopping Mode III Crack . . . . . . . . . . . . . . . . 303
6.10.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
6.11 Brittle Fracture of Chemically Complex Materials . . . . . . . . . 304
6.11.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305
6.11.2 Hybrid Atomistic Modeling of Cracking
in Silicon: Mixed Hamiltonian Gormulation . . . . . . . . . 307
6.11.3 Atomistic Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
6.11.4 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
XIV Contents
6.11.5 Dynamical Fracture Mechanisms . . . . . . . . . . . . . . . . . . 311

6.11.6 Reactive Chemical Processes and Fracture
Initiation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
6.11.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
6.12 Summary: Brittle Fracture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
6.12.1 Hyperelasticity can Govern Dynamic Fracture . . . . . . . 323
6.12.2 Interfaces and Geometric Confinement . . . . . . . . . . . . . 326
7 Deformation and Fracture of Ductile Materials . . . . . . . . . . 327

7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
7.2 Continuum Theoretical Concepts of Dislocations
and Their Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
7.2.1 Properties of Dislocations . . . . . . . . . . . . . . . . . . . . . . . . 329
7.2.2 Forces on Dislocations . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
7.2.3 Rice–Thomson Model for Dislocation Nucleation . . . . 332
7.2.4 Rice–Peierls Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
7.2.5 Link with Atomistic Concepts . . . . . . . . . . . . . . . . . . . . . 338
7.2.6 Generalized Stacking Fault Curves . . . . . . . . . . . . . . . . . 338
7.2.7 Linking Atomistic Simulation Results
to Continuum Mechanics Theories of Plasticity . . . . . . 339
7.3 Modeling Plasticity Using Large-Scale
Atomistic Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
7.4 Case Study: Deformation Mechanics of Model FCC
Copper – LJ Potential . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
7.4.1 Model Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
7.4.2 Visualization Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . 345
7.4.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355
7.5 Case Study: Deformation Mechanics
of a Nickel Nanocrystal – EAM Potential . . . . . . . . . . . . . . . . . 357
7.6 Case Study: Multi-Paradigm Modeling of Chemical
Complexity in Mechanical Deformation of Metals . . . . . . . . . . 359
7.6.1 Atomistic Model and Validation . . . . . . . . . . . . . . . . . . . 360
7.6.2 Example Application: Modeling Hybrid
Metal–Organic Systems . . . . . . . . . . . . . . . . . . . . . . . . . . 364
8 Deformation and Fracture Mechanics

of Geometrically Confined Materials . . . . . . . . . . . . . . . . . . . . . 373
8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
8.2 Thin Metal Films and Nanocrystalline Metals . . . . . . . . . . . . . 381
8.2.1 Constrained Diffusional Creep in Ultra-Thin
Metal Films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
8.2.2 Single Edge Dislocations in Nanoscale Thin Films . . . 390
Contents XV
8.2.3 Rice–Thompson Model for Nucleation of Parallel

Glide Dislocations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393
8.2.4 Discussion and Summary . . . . . . . . . . . . . . . . . . . . . . . . . 396
8.3 Atomistic Modeling of Constrained Grain Boundary
Diffusion in a Bicrystal Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
8.3.1 Introduction and Modeling Procedure . . . . . . . . . . . . . . 397
8.3.2 Formation of the Diffusion Wedge . . . . . . . . . . . . . . . . . 400
8.3.3 Development of the Crack-Like Stress Field
and Nucleation of Parallel Glide Dislocations . . . . . . . 402
8.3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
8.3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
8.4 Dislocation Nucleation from Grain Triple Junction . . . . . . . . . 408
8.4.1 Atomistic Modeling of the Grain Triple Junction . . . . 409
8.4.2 Atomistic Simulation Results . . . . . . . . . . . . . . . . . . . . . 410
8.4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
8.5 Atomistic Modeling of Plasticity of Polycrystalline Thin
Films . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
8.5.1 Atomistic Modeling of Polycrystalline Thin Films . . . 415
8.5.2 Atomistic Simulation Results . . . . . . . . . . . . . . . . . . . . . 416
8.5.3 Plasticity of Nanocrystalline Bulk Materials
with Twin Lamella . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
8.5.4 Modeling of Constrained Diffusional Creep
in Polycrystalline Films . . . . . . . . . . . . . . . . . . . . . . . . . . 426
8.5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 428
8.5.6 Summary: Results of Modeling of Thin Films . . . . . . . 430
8.6 Use of Atomistic Simulation Results in Hierarchical
Multiscale Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 432
8.6.1 Mechanisms of Plastic Deformation of Ultra-thin
Uncapped Copper Films . . . . . . . . . . . . . . . . . . . . . . . . . . 434
8.6.2 Deformation Map of Thin Films . . . . . . . . . . . . . . . . . . . 434
8.6.3 Yield Stress in Ultra-Thin Copper Films . . . . . . . . . . . 435
8.6.4 The Role of Interfaces and Geometric Confinement . . 436
8.7 Deformation and Fracture Mechanics of Carbon
Nanotubes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
8.7.1 Mesoscale Modeling of CNT Bundles . . . . . . . . . . . . . . . 441
8.7.2 Mesoscale Simulation Results . . . . . . . . . . . . . . . . . . . . . 444
8.7.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 445
8.8 Flaw-Tolerant Nanomaterials: Bulk Fracture
and Deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 446
8.8.1 Strength of Brittle Nanoparticles . . . . . . . . . . . . . . . . . . 446
8.9 Nanoscale Adhesion Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
8.9.1 Strength of Fibrillar Adhesion Systems . . . . . . . . . . . . . 453
8.9.2 Theoretical Considerations of Shape
Optimization of Adhesion Elements . . . . . . . . . . . . . . . . 455
XVI Contents
8.9.3 Atomistic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456

8.9.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 460
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 463
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
List of Figures
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 . . . . . . . . . . . 4
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] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.3 Homogeneous material (subplot (a)) and material with
elliptical hole (subplot (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) . . . . . . . . . . . . . . . . . . . . 7
1.4 Schematic illustration of a failure process by crack extension
in a brittle material. The inlay in the center shows how
chemical bonds rupture continuously, leading to formation
of new fracture surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Overview of different crystal structures, showing the SC,
FCC, and BCC crystal structure . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.6 Dislocations are the discrete entities that carry plastic
(permanent) deformation; measured by a “Burgers vector.”
The snapshots illustrate the nucleation and propagation of
an edge dislocation through a crystal, leading to permanent
deformation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
XVIII List of Figures
1.7 Schematic of brittle (a) vs. ductile (b) materials behavior.

In brittle fracture, 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
dislocations. 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 . . . . . . . . . . . . . . . . . . . . . 13
1.8 Brittle (a) vs. ductile (b) materials behavior observed in
atomistic computer simulations. In brittle materials failure,
thousands of cracks break the material. In ductile failure,
material is plastically deformed by motion of dislocations . . . . 14
1.9 Overview over timescales and lengthscales associated with
various problems and applications of mechanical properties
(adapted from [2]) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.11 Nanomechanical experiments of bending deformation of
200-nm gold nanowires. Subplot (a) depicts a schematic of
a fixed wire in a lateral bending test with an AFM tip.
Subplots (b–e) depict AFM snapshots of the mechanical
deformation the nanowire. Subplot (b) depicts results after
elastic deformation, subplots (c) and (d) shows results after
successive plastic manipulation, and subplot (e) shows an
SEM image following the bending test. The SEM picture
agrees in detail with the AFM image shown in subplot (d).
All scale bars are 1 µm. Reprinted with permission from
Macmillan Publishers Ltd, Nature Materials [3] c 2005 . . . . . 21
List of Figures XIX
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
XX List of Figures
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 fracture. 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 . . . . . . . . . . . . . . . 26
1.16 AFM experiments of protein unfolding. Subplot (a): Force
peaks corresponding 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 sawtooth pattern. Reprinted with
permission from Macmillan Publishers Ltd,
Nature [6] c 1999 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.17 Large deformation of a protein, here an example of unfolding
of the enzyme lysozyme, result of a reactive force field
simulation. The distance between the ends of the protein
(Cα -atom of the terminal residues) is continuously increased
by applying a continuously increasing force [7]. As the force
is increased, the protein molecule undergoes significant
structural changes relative to its initial folded configuration . . 28
2.1 Molecular dynamics can be used to study material

properties at the intersection 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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
List of Figures XXI
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 . . . . . . . 34
2.3 Molecular dynamics generates the dynamical trajectories of
a system of N particles by integrating Newton’s equations of
motion, with suitable initial and boundary conditions, and
proper interatomic potentials, while satisfying macroscopic
thermodynamical (ensemble-averaged) constraints, leading
to atomic positions ri (t), the atomic velocities vi (t), and
their accelerations ai (t), all as a function of time, for all
particles i = 1 . . . N , each of which has a specific mass mi . . . 38
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”) . . . 41
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. . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.6 Example result of an energy minimization, here an example
of minimizing the structure of a solvated protein (lysozyme).
As the number of iterations progresses, the total potential
energy decreases, until it converges and reaches a constant
value (see [8] for further details) . . . . . . . . . . . . . . . . . . . . . . . . . . 44
2.7 Illustration of generation of random perturbation from an
initial state A toward a state B, as typically performed in
Monte Carlo schemes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
XXII List of Figures
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 perturbation
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 . . . . . . . . . . . . . 46
2.9 Schematic of the typical characteristic of a chemical bond,
showing repulsion at small distances below the equilibrium
separation r0 , and attraction at larger distances . . . . . . . . . . . . 47
2.10 Atoms are composed of electrons, protons, and neutrons.
Electrons and protons are negative and positive charges of
the same magnitude. In classical molecular dynamics, the
three-dimensional atom structure is replaced by a single
mass point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
2.11 Overview over different simulation tools and associated
lengthscale and timescale . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
2.13 Replacing a full-electron representation of atom–atom
interaction by a potential function that only depends on the
distance r between the particles . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.14 Plot of the LJ potential and its derivative (for interatomic
forces) in a parametrization for copper as reported in [9] . . . . 53
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 . . . . . . . . . . . . . . . . . . . . . . . . . 54
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 . . . . . . . . . . . . . . . . . . . . . . . . 56
List of Figures XXIII
2.17 Chemical complexity in proteins involves a variety of

chemical elements and different chemical bonds between
them. The snapshot shows a small alpha-helical coiled coil
protein domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
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 . . . 57
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 experimental 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
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 . . . . . . . . . . . . . . . 62
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
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]) . . . . . . . . . . . . . . . . . . 66
XXIV List of Figures
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] . . . . . . . . . . . . . . . . . . . . . 67
2.24 The ReaxFF force field fills a gap between quantum
mechanical methods (e.g., DFT) and empirical molecular
dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
2.25 Schematic of the numerical scheme in carrying out molecular
dynamics simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
2.26 Schematic of the numerical scheme in carrying out molecular
dynamics simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
2.29 Method to calculate the radial distribution function g(r) . . . . 74
2.30 Radial distribution function g(r) for various atomistic
configurations, including a solid (crystal), a liquid and a gas . 75
2.31 Velocity autocorrelation function (VAF) for a gas, liquid,
and solid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
2.33 Example of how to calculate the stress tensor in a
1D system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
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 . . . . . . . 80
List of Figures XXV
2.35 Summary of top 10 of the TOP500 supercomputer list, as of

Spring 2008 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
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 Computational
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
2.37 Rendering of a large molecular dynamics simulation on a
tiled display at USC, showing hypervelocity impact damage
of a ceramic plate with impact velocity 15 kms−1 , where one
quarter of the system is cut to show the internal pressure
distribution (the projectile is shown in white). This figure
illustrates how novel visualization schemes provide analysis
methods for ultra large-scale simulations. Reprinted from
Computational 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
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 . . . . . . . . . . . . . . . . . 85
XXVI List of Figures
2.39 Application of the energy method to visualize fracture

surfaces in a computational fracture experiment. Only high
energy atoms are shown by filtering them according to their
potential energy. This enables an accurate determination
of the geometry of cracks, in particular of the crack tip.
Typically, the analysis is confined to a search region (shown
as a dashed line) to avoid inclusion of effects of free surfaces . 86
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
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 . . . . . . . . . . 88
2.42 Analysis of a simple alpha-helix protein structure, with
different visualization options, plotted using VMD [148] . . . . . 90
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 . . . . . . . . . . . 91
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
2.45 Steered molecular dynamics simulations of I27 extensibility
under constant 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 . . . . . . . . . . . . . 94
List of Figures XXVII
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96
3.2 Example for deformation of a beam due to mechanical
loading of a distributed 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. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
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 . . . . . . . . . . . . . . . 99
3.5 The most general state of stress acting on a infinitesimal
material element. All stresses shown in the figure have
positive sense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
3.6 Infinitesimal element with stresses and body forces fi acting
as volume forces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.7 This schematic explains the condition that σij = σji so that
there is no moment on the infinitesimal element, since it
cannot rotate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
3.8 Schematic to illustrate the definition of the deformation
tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
3.9 Schematic to illustrate the difference between rotational and
deformation part of the deformation tensor . . . . . . . . . . . . . . . . 104
3.10 Illustration of the concept of nonlinear elasticity or
hyperelasticity. Subplot (a) shows the stress–strain
relationship, and subplot (b) depicts the tangent modulus
as a function of strain. Linear elasticity is based on the
assumption that the modulus is independent of strain.
However, most real materials do not show this behavior.
Instead, they show a stiffening effect (e.g., rubber, polymers,
biopolymers) or a softening effect (e.g., metals, ceramics) . . . . 109
3.11 Geometry of the beam . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
XXVIII List of Figures
3.12 Solution field for a simply supported beam under dead

load ρg, showing the shear force Qy , bending moment My ,
rotation ωy , and the beam axis displacement uz . . . . . . . . . . . . 112
3.13 Demonstration of the concept of the Navier–Bernouilli
assumption . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.1 Example to illustrate Cauchy–Born rule in a one-dimensional

geometry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
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 . . . . . 128
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 . . . . . . . . . . . . . . 131
4.4 Elastic properties associated with the tethered LJ potential,
and in comparison, 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 . . . . . . . . . . . . . . . . . . 132
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
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)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
4.7 The figure shows the stretching of the triangular lattice in
two different directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
List of Figures XXIX
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
[100] crystal orientation, without Poisson relaxation. The
plot shows the elastic properties as a function of strain . . . . . . 145
[100] crystal orientation, triaxial loading. The plot shows
the elastic properties as a function of strain . . . . . . . . . . . . . . . . 146
(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
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
4.20 This plot depicts the entropic response (F < 14 pN) of a
single tropocollagen 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 persistence length of
approximately 16 nm [17]. The force-extension curve shows
a strong hyperelastic stiffening effect (see also Fig. 3.10) . . . . . 152
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
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 . . . . . . . . . . . . . . . . . 158
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
5.3 Example for implementation of a hierarchical multiscale
method, where parameters are passed through various
lengthscales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
List of Figures XXXI
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. Subplot (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. Karpov, S.
Zhang, and H.S. Park, An introduction to computational
nanomechanics and materials, copyright c 2004, with
permission from Elsevier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162
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
Mechanics 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 . . . . . . . . . . . . . . . . . . . . 165
5.6 Results of a simulation of a crack in a thin film constrained
by a rigid substrate, exemplifying a study using a concurrent
multiscale simulation method, the quasicontinuum
approach [20] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
XXXII List of Figures
5.7 Application of the quasicontinuum method in the simulation

of a nanoindentation experiment. Subplots (a) and (b)
depicts a cross-sectional view of the test sample used in
the nanoindentation simulations for increasing indenter
penetration (part of the indenter is also shown). Subplot (c)
plots the dislocation structure at the indenter penetration
corresponding to the indentation depth shown in subplot
(b). Subplot (d) shows a load vs. displacement curve
predicted by full atomistic (LS) and quasicontinuum (QC)
simulations, illustrating that the two methods show excellent
agreement. Reprinted from Journal of the Mechanics and
Physics of Solids, Vol. 49(9), J. Knap and M. Ortiz, An
analysis of the quasicontinuum method, copyright c 2001,
with permission from Elsevier . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.8 The interpolation method for defining a mixed Hamiltonian
in the transition 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
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 . . . . 172
5.10 Example CMDF script (upper part) and schematic of the
structure of CMDF (lower part) . . . . . . . . . . . . . . . . . . . . . . . . . . 175
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
5.12 Atomistic model to study surface diffusion of a single
adatom on a flat [100] copper surface . . . . . . . . . . . . . . . . . . . . . 178
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 178
5.14 Snapshots of states A (perfect step) and B (single atom
hopped away from step) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179
List of Figures XXXIII
5.15 Hybrid CADD-Parallel Replica study of mechanical

twinning of a metal. Subplot (a) shows the simulation
domain, illustrating the continuum/discrete dislocation
regime and the full atomistic domain (blow-up in right part).
Subplot (b) shows a comparison of atomistic simulation
results with the predictions of an analytical model. The plot
shows the time to nucleation of a trailing or twinning partial
versus applied load in Al at a temperature of 300 K. The
circles refer to the multiscale simulation results covering
many orders of magnitudes in timescales. The dashed lines
correspond to the predictions of the analytical. Reprinted
with permission from Macmillan Publishers Ltd, Nature
Materials [21] c 2007 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.1 Picture of shattered glass, a model for a brittle material . . . . . 185

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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 186
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
6.4 Thin strip geometry. The gray arrows indicate the mode I
(tensile loading), by a stress σ0 . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
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 . . . . . . . . . . . . . . . . . 193
6.6 Schematic of cracks under mode I, mode II, and mode III
crack loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
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 . . . . . . . . . . . . . . . . . . . . . . . . . 195
XXXIV List of Figures
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
propagates 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 symbols). 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 . . . . . . . . . . . . . . . . . . . . . 199
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
6.10 Geometry of the one-dimensional model of fracture . . . . . . . . . 202
6.11 One-dimensional atomistic model of dynamic fracture . . . . . . . 203
6.12 Bilinear stress–strain law as a simplistic model of
hyperelasticity (mimicking the behavior shown in Fig. 3.10).
The parameter εon determines the critical strain where the
elastic properties change from local (El ) to global (Eg ) . . . . . . 207
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)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
6.14 Magnitude of the local stress wave for different crack
propagation velocities from atomistic simulations, in
comparison with the theory prediction . . . . . . . . . . . . . . . . . . . . 214
6.15 Dynamic fracture toughness for different crack propagation
velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
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 . . . . . . . . . . . . . . . . . . . . . . . . . . 216
6.17 Prescribed fracture toughness and measured crack velocity
as the crack proceeds along x . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217
List of Figures XXXV
6.18 Strain field of a crack travelling in a material with

periodically varying fracture toughness . . . . . . . . . . . . . . . . . . . . 218
6.19 Elastic properties associated with the biharmonic
interatomic potential, for ron = 1.125 and Eg = 8 = 1/4El . . . 219
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 comparison 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 220
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
6.22 Particle velocity field near a supersonic crack, comparison
between theory and simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
6.23 Simulation geometry and coordinate system for studies of
rapidly propagating mode I cracks in harmonic lattices . . . . . . 224
6.24 Comparison between σxx from molecular dynamics
simulation with harmonic potential and the prediction of
the continuum mechanics theory for different reduced crack
speeds v/cR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
6.25 Comparison between σyy from molecular dynamics
speeds v/cR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 226
6.26 Comparison between σxy from molecular dynamics
speeds v/cR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
6.27 Comparison between hoop stress from molecular dynamics
speeds v/cR . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
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 . . . . . . . . . . . . . . . . . . . . . . . 228
XXXVI List of Figures
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 . . . . . . . . . . . . . . . . . . . 229
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 . . . . . . . . . . . . . . . . . . . . . . . 230
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 . . . . . . . . 231
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 . . . . . . . . 231
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 . . . . . . . . . 232
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) . . . . . . . . . . . . . . . . . . . . 232
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 234
List of Figures XXXVII
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”) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236
6.37 Hyperelastic region in a (a) softening and (b) stiffening
system . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
6.38 Hyperelastic region and enhancement of energy flow in the
(a) softening and (b) stiffening system . . . . . . . . . . . . . . . . . . . . 240
6.39 J-integral analysis of a crack in a harmonic, softening and
stiffening material, for different choices of the integration
path Γ . The straight lines are a linear fit to the results
based on the calculation of the molecular dynamics
simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242
6.41 Shape of the hyperelastic regions for different choices of εon
(the hyperelastic 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244
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 . . . . . . . . 245
6.44 The plot shows a temporal sequence of supersonic mode II
crack propagation. The field is colored according to the σxx
stress component . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
6.45 Geometry of the Broberg problem of a crack propagating in
a thin stiff layer embedded in soft matrix . . . . . . . . . . . . . . . . . . 246
XXXVIII List of Figures
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 . . 247
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 . . . . . . . . . . . . . . . . . . . . . 248
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . 250
6.49 The concept of hyperelastic softening close to bond
breaking, in comparison to the linear elastic, bond-snapping
approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 253
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 . . . . . . . . . . . . . . . . . . . . . . . . . . 254
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
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 . . . . . . . . . . . . . . . . . . . 256
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 . . . . . 257
List of Figures XXXIX
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 . . . . . 258
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
6.56 Allowed velocities for mode I and mode II crack propagation,
linear and nonlinear stiffening case . . . . . . . . . . . . . . . . . . . . . . . 261
6.57 Simulation geometry for the stopping crack simulation . . . . . . 264
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 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
6.59 Crack extension history vs. time for the suddenly stopping
linear mode I crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
6.60 Maximum principal stress field for various instants in time,
mode I linear crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
6.61 Evolution of principal maximum stress and potential energy
along the prospective crack line, for a linear mode I crack . . . 271
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
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
XL 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
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
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
6.85 Crack tip velocity history during the mother–daughter–

granddaughter mechanism, for elastic mismatch Ξ = 3. The
plot is obtained by numerical differentiation of the crack
tip history shown in Fig. 6.84. The crack speed changes
abruptly at the nucleation of the daughter crack, and rather
continuously as the granddaughter crack is nucleated.
Characteristic wave speeds for the stiff and soft solid are
indicated in the plot . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
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 . . . . . . 296
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 . . . . . . . . . 297
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 . . . . . . . . . . 298
6.90 Allowed velocities for mode III crack propagation, linear
and nonlinear case . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
List of Figures XLIII
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
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 . . . . . . . . . . . . . . . . . . . . . . . . 312
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
Letters, Vol. 85(4), 2000, pp. 788–791. Copyright c 2000
by the American Physical Society . . . . . . . . . . . . . . . . . . . . . . . . 312
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313
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)) . . . . . . . . . . . . . . . . . . . 314
6.105 Comparison of the prediction of LEFM, the prediction of
the modified LEFM model (6.129), and molecular dynamics
simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315
6.106 Comparison of the crack dynamics under mode I (subplot
(a)) and mode II loading (subplot (b)) [26] . . . . . . . . . . . . . . . . 317
List of Figures XLV
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 318
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] . . . . . . . . . . . . . . . . . . . . . . 319
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328
7.2 Illustration of splitting of complete dislocation into two
partial dislocations in an FCC lattice . . . . . . . . . . . . . . . . . . . . . 330
7.3 Geometry to explain the Schmid law, illustrating the angles
φ and λ as well as the uniaxially applied stress σy . . . . . . . . . . 331
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 . . . . . 332
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 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
7.6 Geometry of a dislocation close to a surface, at distance d . . . 333
7.7 Geometry of a cracked specimen with a penny-shaped crack
(subplot (a)) and a semi-infinite crack (subplot (b)) . . . . . . . . 335
XLVI List of Figures
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
7.9 Balance of energies before (1) and after (2) nucleation of the
dislocation, to illustrate the concept of energy release rate . . . 336
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
7.11 The multiplicative decomposition F = Fe Fp in continuum
theory of plasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344
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 . 346
7.14 Dislocation–particle interaction in ordered matrix materials.
A dissociated complete dislocation interacts individually
with the particle (subplot (a) for a schematic). A
TEM weak-beam micrograph of the dislocation–particle
interaction in Fe–30 at.%Al is shown in subplot (b).
Reprinted from Acta Metallurgica, Vol. 46(16), pp.
5611–5626, E. Arzt, Size effects in materials due to
microstructural and dimensional constrains: A comparative
review, copyright c 1998, with permission from Elsevier . . . . 347
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347
List of Figures XLVII
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
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 . . . . . . . . . . . . . . . . . . . . . 348
7.18 Generation of point defects due to jogs in screw dislocations.
Two representative 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 . . . . . . . . . . . . . . . . . . . . . . . . 349
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 subsequently 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 centrosymmetry 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 . . . . . . . . . . . . 350
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 . . . . . . . . . 350
7.21 Activation of secondary slip systems and generation of
Lomer–Cottrell locks. Subplot (a) shows a schematic
of the cross-slip mechanism. Subplots (b) and (c) show
details of activation of secondary slip systems (subplot
(c) represents a magnified view of subplot (b), with the
region of interest highlighted by a circle). This mechanism
of cross-slip of partial dislocations, here first observed in
molecular dynamics simulation, was originally proposed
theoretically by Fleischer [37], and contrasts the well-known
Friedel–Escaig mechanism [38] . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
XLVIII 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
7.31 Centrosymmetry analysis of the details of the dislocation

structure, illustrating 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 . . . . . . . . . . . . . . . . . . . . . . . . . 361
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) . . . . 362
7.33 Centrosymmetry analysis plot of a complete dislocation
emitted as two partials from a crack under shear (mode II)
loading moving through EAM and ReaxFF regions (atoms
with perfect FCC coordination have been removed). The
dashed black circles indicate the region of 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–e) are taken at time interval increments of 500 fs each.
The distance between the partials appears unchanged
throughout the nucleation and propagation process, and
does not change as it passes through the handshaking and
ReaxFF regions and back into the EAM region . . . . . . . . . . . . . 363
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, indicating a smooth handshaking between the two
simulation methods without force discontinuities . . . . . . . . . . . 365
7.35 Partial dislocation emission from crack tip for the 110112
crack orientation (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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 367
L List of Figures
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
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
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
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] . . . . . . . . . . . . . . 371
8.1 This graph shows a proposed deformation-mechanism map

for nanocrystalline 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 . . . . . . . . . . . . . . 374
8.2 High resolution TEM images of grain boundaries in
electrodeposited nanocrystalline Ni (a) and nanocrystalline
Cu (b). The samples were produced by gas-phase
condensation. The grain boundaries constitutes itself as a
very narrow region between crystals of different orientation.
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
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
LII List of Figures
8.4 Illustration of the maximum in the strength of

nanocrystalline copper, as shown by molecular dynamics
simulation of up to 100 million atoms [43]. Panel (a) shows
the stress–strain curves for ten simulations with varying
grain sizes. Panel (b) depicts the flow stress, defined as
the average stress in the strain interval from 7 to 10%
deformation. The error bars indicate the fluctuations in
this strain interval (1 standard deviation). A maximum in
the flow stress is seen for grain sizes of 10–15 nm, caused
by a shift from grain boundary mediated to dislocation
mediated plasticity. Reprinted from Science, Vol. 301, J.
Schiotz and K.W. Jacobsen, A Maximum in the Strength of
Nanocrystalline Copper, copyright c 2003, with permission
from AAAS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
8.5 Size dependence of a bioinspired metallic nanocomposite
(schematic of structure shown in left part) [44, 45],
illustrating the existence of “a strongest size” [322]. The
increase in strength for building block dimensions larger than
approximately 50 nm scales rather well according to the
Hall–Petch relationship given in (8.1) . . . . . . . . . . . . . . . . . . . . . 377
8.6 Study of size-scale effects in inorganic materials by using a
focused ion beam (FIB) microscope machining technique,
combined with nanoidentation experiments. 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 . . . . . . . . . . . . . . . . . . . . . . . . 380
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
List of Figures LIII
8.8 Experimental results of the thin film strength dependence at

room temperature, 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 . . . . . . 383
8.9 Discrete dislocation dynamics simulation analysis of the
strength dependence 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
average 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 . . 384
8.10 Distribution of dislocations, result from a discrete
dislocation dynamics simulation analysis. The plot depicts
the dislocation distribution and the stress field σxx at
T = 400 K for films with a grain size of 1 µm and various film
thicknesses. The plus and minus symbols denote positive
and negative dislocations according to the sign convention
defined in Fig. 8.9a (all dimensions are in µm). Reprinted
from Thin Solid Films, Vol. 479(1–2), L. Nicola, E. Van der
Giessen, A. Needleman, Size effects in polycrystalline thin
films analyzed by discrete dislocation plasticity, copyright
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 . . . . . . . . . . . . . . . . . . . . 386
LIV List of Figures
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 . . . . . . . . . . . 387
8.13 Mechanism of constrained diffusional creep in thin films as
proposed by Gao et al. [46] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 387
8.14 Development of grain boundary opening ux normalized by a
Burgers vector over time, for the case of a copper film on
a rigid substrate. The loading σ0 is chosen such that the
opening displacement at the film surface (ζ = 0) at t → ∞
is one Burgers vector [46] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
8.15 Traction along the grain boundary for various instants in
time [46] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 389
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] . . 390
8.17 Geometry and coordinate system of the continuum
mechanics model of constrained diffusional creep . . . . . . . . . . . 391
8.18 Image stress on a single edge dislocation in nanoscale thin
film constrained by a rigid substrate . . . . . . . . . . . . . . . . . . . . . . 391
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 392
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 394
8.21 Dislocation model for critical stress intensity factor for
nucleation of parallel glide dislocations . . . . . . . . . . . . . . . . . . . . 395
8.22 Disordered intergranular layer at high-energy grain
boundary in copper at elevated temperature (85% of melting
temperature) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
8.23 Sample geometry of the atomistic simulations of constrained
diffusional creep in a bicrystal model . . . . . . . . . . . . . . . . . . . . . . 398
List of Figures LV
8.24 Change of displacements in the vicinity of the diffusion

wedge over time. The continuous dark line corresponds to
the continuum mechanical solution . . . . . . . . . . . . . . . . . . . . . . . 400
8.25 Diffusional flow of material into the grain boundary. Atoms
that diffused into the grain boundary are highlighted . . . . . . . 401
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 403
8.27 Nucleation of parallel glide dislocations from a crack,
showing the dynamical 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 . . . . . . . . . . . . . . . 405
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 . . . . . . . . . . . . . . . 408
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 . . . . . . . . . 411
8.30 Schematic of dislocation nucleation from different types
of grain boundaries. Individual misfit dislocations at
low-energy grain boundaries serve as sources for dislocations.
At high-energy grain boundaries, there is no inherent,
specific nucleation site so that the point of largest resolved
shear stress, the grain triple junction, serves as nucleation
point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
8.31 Deformation twinning by repeated nucleation of partial
dislocations. Repeated slip of partial dislocations leads to
generation of a twin grain boundary . . . . . . . . . . . . . . . . . . . . . . 412
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) . . . . . . . . . . . . . . . . . . . . . . . 412
LVI List of Figures
8.33 Generation of trails of point defects. Subplot (a): Dislocation

cutting processes with jog formation and generation of trails
of point defects. Both dislocations leave a trail of point
defects after intersection. The arrows indicate the velocity
vector of the dislocations. Subplot (b): Nucleation of
dislocations on different glide planes from grain boundaries
generate a jog in the dislocation line that causes generation
of trails of point defects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
8.34 Geometry for the studies of plasticity in polycrystalline
simulation sample . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
8.35 Atomistic model of the polycrystalline thin film. Only
surfaces (brighter coloring) and grain boundaries (darker
color) are shown . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416
8.36 Nucleation of threading dislocations in a polycrystalline
thin film. Subplot (a) shows a view into the interior,
illustrating how threading dislocations glide by leaving an
interfacial segment. Subplot (b) shows a top view into the
grain where the surface is not shown. The plot reveals that
the dislocation density is much higher in grains 3 and 4 . . . . . 418
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 . . . . . . . . . . . . . . . . . . 418
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
8.39 Sequence of a nucleation of a threading dislocation, view
at an inclined angle from the film surface. Threading
dislocations are preferably nucleated at the grain
boundary–surface interface and half-loops grow into the film
until they reach the substrate. Due to the constraint
by the substrate, threading dislocations leave an interfacial
segment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 419
List of Figures LVII
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 . . . . . . . . . . . . . . . . . . 420
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 . . . . . . . . . . . . . . 420
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
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 . . . 422
8.44 Nanostructured material with twin grain boundary
nanosubstructure. The light gray lines inside the grains refer
to the intragrain twin grain boundaries. The thickness of
the twin lamella is denoted by dT . . . . . . . . . . . . . . . . . . . . . . . . 423
8.45 Simulation results of nanostructured material with twin
lamella substructure 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
thinner twin lamella (dT ≈ 2.5 nm). Motion of dislocations
is effectively hindered at twin grain boundaries . . . . . . . . . . . . . 424
LVIII List of Figures
8.46 Simulation results of nanostructured material with twin

lamella substructure 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, dislocations 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 425
8.47 Modeling of constrained diffusional creep in polycrystalline
samples; nucleation of threading vs. parallel glide
dislocations [49]. The blowup in panel (c) shows an energy
analysis of the dislocation structure and visualizes a parallel
glide dislocation nucleated from a grain boundary diffusion
wedge. The surface steps indicate that threading dislocations
have moved through the grain and no threading dislocations
exist in grain 1. The dark lines show the network of parallel
glide dislocations in grain 1 (in other grains we also find
parallel glide dislocations in snapshot (d) but they are not
shown) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 427
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
List of Figures LIX
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 diffusion initiation, the yield
stress is film-thickness independent as also observed in
experiment [51] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 433
8.50 Deformation map of thin films, different regimes. Thin films
with thicknesses in the submicron regime feature several
novel mechanisms next to the deformation by threading
dislocations (a). Plasticity can be dominated by diffusional
creep and parallel glide dislocations (b), purely diffusional
creep (c), and no stress relaxation mechanism (d) . . . . . . . . . . 435
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) . 436
8.52 Geometry of a (15,15) single wall carbon nanotube (SWNT)
shown in different views . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
8.53 Compressive deformation mechanics of a CNT with length
L = 6 nm and a diameter of d = 1 nm ((7,7) armchair
CNT). The plot illustrates the increase of strain energy as
a function of compressive strain (subplot (a)), and shows
associated deformation mechanisms (subplot (b)). The
analysis revealed that CNTs begin to buckle according to
a shell-like behavior. Subplot (c) depicts a similar analysis,
showing bending of a (13,0) CNT (length L = 8 nm and a
diameter of d = 1 nm). The strain energy density increases
according to a harmonic behavior until the buckling point is
reached. Reprinted from: B.I. Yakobson, C.J. Brabec, and
J. Bernholc, Physical Review Letters, Vol. 76(14), 1996, pp.
2511–2514. Copyright c 1996 by the American Physical
Society . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 439
8.54 Shell-rod wire transition of CNTs under compressive loading
(schematically 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 . . 440
8.55 Illustration of the process of coarse graining a CNT, leading
to a bead-type representation [52] . . . . . . . . . . . . . . . . . . . . . . . . 441
LX List of Figures
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 . . . . . . . . 443
8.57 Bundle of several individual SWNTs as obtained from a
molecular simulation model [52] . . . . . . . . . . . . . . . . . . . . . . . . . . 444
8.58 Response of a CNT bundle to mechanical compressive
loading. Even for relatively small strains, the structure
starts to buckle, eventually leading to significantly deformed
and buckled shapes [52] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
8.59 Full atomistic calculations of the properties of ultra-long
CNTs (subplot (a)) [371, 372] and corresponding results
obtained using the mesoscale model (subplot (b)) [52] . . . . . . . 445
8.60 Bone-like materials consist of a hierarchical microstructure
made of nanoscale constituents [53, 54]. Left : The plot
depicts the microstructure of such bone-like biological
materials at the smallest scale. Such materials typically
consist of fragile, brittle mineral platelets embedded in
protein matrix materials as soft as human skin. The
combination of these two phases in a nanocomposite results
in superior materials properties. In the studies, we focus on
the fracture properties of mineral platelets since they play a
critical role in determining the strength of these materials.
Right : The tension–shear chain model showing the path
of load transfer in the mineral–protein composites. The
mineral platelets carry tensile load and the protein transfers
loads between the platelets via shear. In this section we
focus on the strength of the mineral platelets . . . . . . . . . . . . . . 447
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) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 448
List of Figures LXI
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
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 distribution. 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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 452
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, Biological and artificial attachment devices:
Lessons for materials scientists from flies and geckos, pp.
1245–1250, copyright c 2006, with permission
from Elsevier . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 454
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 455
LXII List of Figures
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456
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 ) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
8.69 Stress distribution in the rigid punch slightly before
complete detachment (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 concentration 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 . . . . . . . . . . . . . . . . . . . . . . . . . . 458
8.70 Stress distribution in the elastic punch slightly before
complete detachment (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 corresponds to the Young’s modulus obtained
for k0 = 57.23). We find that thestress distribution becomes
homogeneous for large ratios of Rcr /R, in agreement with
the other results (see Figs. 8.63 and 8.69) . . . . . . . . . . . . . . . . . . 459
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 . . . . . 460
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 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
List of Tables
2.1 Overview of various thermodynamical ensembles (the

parameter µ is the chemical potential) . . . . . . . . . . . . . . . . . . . . . 40
2.2 Overview over force fields suitable for organic substances . . . . . 60
2.3 Explanation of the different measures of computing power, as
well as a description when it became available. In comparison
a state-of-the art personal computer (PC, laptop) provides a
performance of approximately 30 GFLOPS . . . . . . . . . . . . . . . . . 79
2.4 Centrosymmetry parameter ci for various types of defects,
normalized by the square of the lattice constant a20 . In the
visualization scheme, we choose intervals of ci to separate
different defects from each other . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
2.5 Distinguishing modeling and simulation, for tasks associated
with classical molecular dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 91
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 . . . . . . . . . . . . . . . . . . . 133
4.2 Failure strain of the two-dimensional solid associated with
the harmonic potential with snapping bonds under different
modes of uniaxial loading for rbreak = 1.17 . . . . . . . . . . . . . . . . . . 136
4.3 Elastic properties and wave speeds associated with the
harmonic potential (see (2.34)) in a 3D solid for different
choices of the spring constant k, cubical crystal orientation . . . 143
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 . . . . . 144
[100] [110] [111]
4.5 Cohesive strains εcoh , εcoh , and εcoh for the LJ potential
and the EAM potential. In all potentials, the weakest pulling
direction is the [110] directions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.6 Summary of fracture surface energies for a selection of
different potentials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
LXIV List of Tables
4.7 Summary of frequently used units to SI units and/or

definition of constants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.1 Activation energy for different state transitions . . . . . . . . . . . . . . 179
6.1 Overview over wave speeds in a variety of materials,

indicating the longitudinal wave speed cl , the shear wave
speed cs , and the Rayleigh wave speed cR . . . . . . . . . . . . . . . . . . 198
6.2 Critical load R0 for fracture initiation, for different values of
the spring constant kp of the pinning potential. The results
are in good agreement with the theory prediction when kp
becomes much smaller than k . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 214
6.3 Change of energy flow to the crack tip, due to a bilinear
softening or stiffening interatomic potential . . . . . . . . . . . . . . . . . 241
6.4 Griffith analysis of the atomistic models, for mode I and
mode II cracks, and different potentials. The predicted
values based on continuum calculations agree well with the
molecular dynamics simulation results . . . . . . . . . . . . . . . . . . . . . . 268
8.1 Material parameters for calculation of stress intensity factor

over the reduced time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
8.2 Critical stress intensity factor K PG for nucleation of parallel
glide dislocations under various conditions, for both a
diffusion wedge and a crack . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
8.3 Summary of mesoscopic parameters derived from atomistic
modeling [52] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 443
Nomenclature
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
Pi Dynamic Poynting vector component

P Magnitude of dynamic Poynting vector
ron Critical atomic separation for onset
of hyperelastic effect (biharmonic potential)
rbreak Critical atomic separation for
bond breaking (biharmonic potential)
kratio Ratio of large-strain to small-strain
spring constant
εon Critical onset strain for hyperelastic
effect (biharmonic potential)
rH (Effective) size of hyperelastic region
χ Characteristic energy lengthscale
t Time (usually in reduced atomic units)
tl Loading time of a cracked crystal
tinit Fracture initiation time
H(s) Heaviside unit step function
δij Kronecker delta function
a Crack length
ȧ = v Time derivative of crack length (=crack speed)
u, ui Displacement vector (i stands for
the spatial direction component)
u̇i Particle velocity vector
A Area (for instance, cross-sectional area of beam)
L Length (for instance, length of beam)
c0 Wave speed in a one-dimensional model
string of atoms
cR Rayleigh wave speed
cs Shear wave speed
cl Longitudinal wave speed
Θ Angle in cylindrical coordinate system at crack tip
r Radial direction in cylindrical coordinate
system at crack tip
σij Stress tensor
εij Strain tensor
σi , εi Principal stress/strain
σΘ , εΘ Hoop stress/hoop strain
σY Yield stress
σth , τth Theoretical maximum stress or shear stress
γs Fracture surface energy
γus Unstable stacking fault energy
γsf Stacking fault energy
∆γ Adhesion energy
G Energy release rate in fracture problems
K, KI,II,III Stress intensity factor (loading modes I, II, III)
Nomenclature LXVII
Σij Characteristic variation of asymptotic stress field

near static or moving crack
m, mi Atomic mass
Ξ Elastic mismatch across bimaterial interface
ci Centrosymmetry parameter of atom i
sαi Slip vector component i of atom α
bi , b Burgers vector (components and vector)
b Magnitude of Burgers vector (=| b |)
a0 Lattice constant of cubic metals
hf Film thickness
d Grain size in a polycrystalline material
ξ Characteristic dimension of nanocrystal
ξp Persistence length of a polymer chain
R Characteristic radius of adhesion punch
τ Characteristic time for stress relaxation
δz Diffusive displacement
Th Homologous temperature
δgb Dgb Grain boundary diffusivity
D Diffusivity
Ds Surface diffusivity
Part I
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
and deformation of different classes of materials, linking the viewpoints of

scientific disciplines such as chemistry and physics with those of engineers.
This merger of scientific and engineering disciplines is an exciting opportunity
that results from research at this fundamental, atomic scale.
This book will provide readers with a basic understanding about the fun-
damentals, application areas, and potential of classical molecular dynamics
for problems in mechanics of materials. Another emphasis is on developing a
sensitivity for the significance of mechanics in different areas, and how atom-
istic and continuum viewpoints can be integrated to build a new platform of
control in the analysis and description of the behavior of complex materials
under extreme conditions. The focus of this book is on atomistic modeling of
small-scale dynamics phenomena with large-scale simulations, combined with
theoretical methods such as continuum mechanics. Throughout the text, the
reader will find a coherent discussion of theory, simulation results, and exper-
imental studies. The discussion of model building and analysis of simulation
results illustrate some of the critical aspects of this computational tool. A
core contribution of this book to the available literature in this field is the
comparative discussion of atomistic simulation results with continuum theo-
ries and other engineering concepts. The main question we are concerned with
1 Introduction 5
is how materials fail under extreme conditions, and how the microscopic or
macroscopic failure processes are related to atomistic details.
1.1 Materials Deformation and Fracture Phenomena:

Why and How Things Break
When materials are deformed, they display a small regime where deformation
is reversible, a behavior referred to as the elastic regime. Once the forces on the
material are increased, deformation becomes irreversible, and the deformation
of a body caused by the applied stress remains after the stress is removed.
This behavior is referred to as the plastic regime, and may cause the material
to fail [1]. In terms of thermodynamics, elastic deformation is characterized by
a reversible process. That is, all mechanical work done on the system is fully
recoverable. In contrast, materials failure represents an irreversible process.
That is, not all mechanical work done on the system can be recovered, as it
is dissipated during processes associated with permanent deformation.
Materials can fail in many ways. Brittle materials like glass shatters and
quickly breaks into many small pieces. Ductile metals can be deformed perma-
nently without breaking, with moderate resistance against the forces. Many
biological tissues such as skin, spider silk, or polymers and rubber are capa-
ble to sustain quite large deformation before they break suddenly. Figure 1.2
depicts a schematic stress–strain curve, comparing a brittle and a ductile
material, introducing one of the most fundamental distinction of material
behavior.
The well-known failure mechanisms are macroscopic phenomena and its
manifestations are often visible to our eyes. However, the origin of virtually all
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]
deformation and fracture phenomena lies at much smaller, atomistic scales.

Whether or not a material behaves brittle or ductile, for instance, depends
strongly on how the atoms and molecules are arranged inside the material,
and how these structures respond to an applied load. Much of this book is
dedicated to elucidate the question of the underlying atomistic and molecular
deformation mechanisms and how they can be linked to macroscale phenom-
ena that are measured in engineering laboratories. In many cases, what mat-
ters for an engineering application is the behavior at larger macroscopic scales.
1.2 Strength of Materials: Flaws, Defects,

and a Perfect Material
How strong are materials, and what are the fundamental mechanisms and fea-
tures in materials that control their deformation behavior and their strength
limit? These aspects have been subject to studies in a variety of scientific and
engineering disciplines. Materials have been studied by physicists, often con-
sidering the smallest lengthscales, featuring a few atoms and below. On the
other hand, chemists considered the bonding between different atoms or
the interactions of different chemical compounds and molecules. Engineers, on
the other hand, have mainly used continuum descriptions of materials, consid-
ering the material as matter that can be divided infinitely many times. These
traditional engineering theories have their origin in neglecting the discreteness
of matter, which is a reasonable assumption when dealing with larger struc-
tures that feature characteristic geometric dimensions much larger than the
inhomogeneities of the material (e.g., molecular structure, grains, particles).
Research carried out over the last few decades revealed that the integration
of these different viewpoints, that is, those of physics, chemistry, and engineer-
ing is critical to make important breakthroughs to understand and improve
the mechanical properties of materials. As we shall discuss in the forthcom-
ing chapters, the mechanical properties of materials are strongly influenced
by the presence of defects or imperfections. Nothing is perfect! There are
many defects in a material, even though through our eyes it may appear as
if no defects are visible. For example a piece of small copper wire will include
many millions of crystal imperfections such as voids, cracks, inclusions of for-
eign materials, and interfaces between crystal grains of different orientations.
The list of possible defects in crystalline materials can be extended much
further. All these defect structures have in common that they represent a
deviation from the ideal, perfect crystal lattice that is considered the refer-
ence structure. Why do defects matter? The reason is that upon application
of an external load to the material, these imperfections lead to a magnifica-
tion of the local stresses, and thereby induce failure since the forces between
atoms become so large that the chemical bonds are sheared or ruptured. If this
occurs collectively over larger lengthscales, the material undergoes permanent
1 Introduction 7
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)
shape change or begins to fracture, if a large number of interatomic bonds are

compromised.
The effect of a defect in changing the distribution of forces within the
material can be demonstrated mathematically by using a simple example that
compares a material with and without an elliptical hole as shown in Fig. 1.3.
For the homogeneous material (Fig. 1.3a), the stress inside the material is
identical to the applied stress. However, the presence of an elliptical hole
leads to a magnification of the stress in the vicinity of the corners of the hole
(Fig. 1.3b). The stress magnification was first quantified by Inglis in 1913 [56],
and it can be shown as:

a
σyy = σ 1 + 2 , (1.1)
ρ
where 2a is the extension of the elliptical inclusion in the x-direction and ρ is

the radius of curvature at the sharp corners. This equation includes the special
case of a homogeneous material (for a = 0), when σyy = σ. This shows that in
this case the stress magnification vanishes. The sharper the radius (that is, for
smaller ρ), or the longer the elliptical hole (that is, for larger a), the greater
the stress magnification at the tip of the hole. This example illustrates the
reason why defects can be sources of failure, since for sufficiently large loads
or sufficiently sharp holes, the local stresses are large enough to break or
shear the bonds between the atoms. The details of how exactly these bonds
respond to extreme loading conditions control the overall material behavior.
It all starts at the atomic scale!
Fig. 1.4 Schematic illustration of a failure process by crack extension in a brittle

material. The inlay in the center shows how chemical bonds rupture continuously,
leading to formation of new fracture surfaces
One mode of failure is the repeated rupture of interatomic bonds, leading

to spreading of fractures in the material as shown in Fig. 1.4. This example
further illustrates that in order to understand materials failure, it is not simply
possible to average over microscopic features. The presence of a tiny defect in
the material can make a huge difference, as it is the source for steady growth
of a fracture that will eventually destroy the entire material or structure.
1.2.1 Crystal Structures and Molecular Packing
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
As indicated above, imperfections in the material structure may give rise

to stress magnifications, leading to breaking of atomic bonds. This breaking of
atomic bonds typically induces additional defects within the crystal. Some of
these defects play a role as mediators of deformation, with the most prominent
example being dislocations. Dislocations are imperfections of a crystal lattice

that allow the material to permanently change its shape. We will discuss
various defects often found in crystals in the following few sections.
1.2.2 Cracks
Cracks or sometimes referred to as fractures are a prominent example of a

material defect. Physically, cracks are represented by surface areas inside a
solid, in other words, an inclusion of “void” inside a crystal lattice. Cracks also
often appear at surfaces of materials. Mathematically, cracks are represented
by traction-free boundary conditions at the surface of the crack, that is, no
stresses can be transmitted from one surface of the crack to the other one.
Similarly as shown in Fig. 1.3, the tip of cracks typically represents loca-
tions of extremely large stresses and thus interatomic forces, which are much
larger than the background load applied. In fact, an infinitely sharp crack
represents a mathematical singularity for the stresses, which scale as
1
σ∼√ . (1.2)
r
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
Figure 1.6 shows the concept of a dislocation in a simple, schematic illustra-

tion. In order to nucleate a dislocation, the material must be sheared locally.
This local shear requires large forces, sufficient to break atomic bonds that
have formed across the sheared plane. For instance, the large stresses in the
vicinity of a crack can induce nucleation of dislocations (recall that mathe-
matically, cracks represents singularities of stress). In this sense, cracks and
dislocations form a symbiotic pair: Cracks lead to large stresses that lead to
nucleation of dislocations, which in turn make the material “deformable,” or
ductile.
The behavior of dislocations in crystals is very complex and involves mul-
tiple mechanisms for generation and annihilation, as summarized in [38, 60].
Collective events may occur through interaction among many dislocations or

between dislocations and other defects such as grain boundaries.
1.2.4 Other Defects in Crystals and Other Structures
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.
1.3 Brittle vs. Ductile Material Behavior

As briefly introduced earlier, materials failure is often divided into two generic
types, brittle and ductile. In the brittle case, atomic bonds are broken as
material separates along a crack front. The type of failure of such materials is
often characterized by the simultaneous motion of thousands of small cracks,
as it is for example observed when glass shatters. This type of failure usually
happens rapidly, as cracks may propagate at velocities close to the speed of
sound in materials. An enormous amount of research has been carried out
over the last century, which has been summarized in recent books [22, 61].
The origin of fracture research dates back to the early twentieth century
in studies by Griffith [62] and Irwin [63]. These studies resulted in theories
that enabled a quantitative prediction of the fracture strength of brittle mate-
rials. The so-called Griffith criterion provides a quantitative estimate of the
condition under which material fails, and is based on simple energetic and
thermodynamic arguments. The Griffith criterion is based on the assumption
that brittle materials fail when the mechanical elastic energy released by crack
1 Introduction 13
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
extension, referred to as G, equals the energy required to generate two new

surfaces, given by 2γs :
G = 2γs , (1.4)
where the parameter G is referred to as the mechanical energy release rate.
In other words, during brittle fracture, elastic energy is dissipated at a rate
G during the fracture process. The dissipated elastic energy is utilized for the
creation of new fracture surfaces and, as in the case of dynamic fracture, a
continuous rise in the temperature.
This thermodynamic view of fracture was the foundation for the field of
linear elastic fracture mechanics (often referred to as LEFM). This continuum
mechanical theory of fracture is a well-established framework, and has been
summarized in several recent textbooks [22, 61, 64, 65].
In ductile failure, a catastrophic event such as rapid propagation of thou-
sands of cracks is prevented by the activation of an alternative deformation
mechanism. Tough materials like metals (such as copper, nickel, gold) do not
shatter. Instead, they deform and bend easily because plastic deformation
occurs by the motion of dislocations inside the material. The competition
between brittle and ductile failure depends on how the interatomic bonds
close to the tip of a crack respond to the locally large interatomic forces.
Thus the tendency of materials to be ductile or brittle depends quite
strongly on the atomic microstructure, in particular on the relative ease of
either dislocation nucleation vs. the energy required to create new surfaces.
The face centered cubic (FCC) packing is known to have a quite strong ten-
dency toward ductility. In contrast, body centered cubic (BCC) crystals are
less ductile. Glasses, as they not have extended crystallinity because atoms are
randomly packed, have no slip-planes, and therefore dislocation nucleation is
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.
1.4 The Need for Atomistic Simulations

The classical theories of continuum mechanics have been the basis for most
theoretical and computational tools of engineers, forming the foundation for
approaches such as the finite element method (often referred to as FEM),
finite volume methods, or finite difference methods.
It was not until the 1980s that scientists and engineers began to include
atomistic descriptions into models of materials failure and plasticity. For
example, in the early stages of computational plasticity, dislocations and
cracks were often treated using linear continuum mechanics theory, often rely-
ing on phenomenological assumptions. Over the last decades, there has been
a new realization that understanding the nanoscale behavior is required for
understanding how materials fail, which opens great opportunities to pro-
vide new design methods for materials from the bottom up. The significance
of atomistic processes to describe deformation and failure of materials is
apparent from the discussion in the last sections.
The significance of including atomistic or nanoscale mechanisms is also due
to the increasing trend toward miniaturization. As shown in Fig. 1.1, relevant
lengthscales of materials approach several nanometers in modern technol-
ogy (e.g., thin films, semiconductors, new composites, carbon nanotubes, new
nanotechnologies based on biological concepts). Once the dimensions of mate-
rials reach the submicron lengthscale, the continuum description of materials
is questionable and the full atomistic information about the material state is
often necessary to study relevant materials phenomena. In addition the signifi-
cance of nanoscopic phenomena for deformation, engineers can now also design
materials at the nanoscale. The ability to create nanostructures by design has
reached a level of perfection that now enables us to make almost arbitrary
structures and shape at ultra-small scales, either through self-assembly or by
utilizing small-scale material manipulation and cutting techniques (such as

focused ion beam, FIB). This ability opens a new era in the design of materials,
from bottom up, enabling new functions, properties, and behaviors for a very
rich set of applications. However, these advances also require new theoretical
concepts to describe the materials behavior, as it can be quite different from
what is conventionally known from larger scale materials. Atomistic models
are often quite suitable to capture these and other effects.
Assuming an atomistic viewpoint has another quite important aspect: It
allows for the seamless communication of various scientific disciplines with
one another, in particular during the study of deformation and fracture. The
reason is that the notion of the “chemical bond” is a concept with which many
disciplines can be linked. In this sense, chemistry is the most fundamental
language of materials science and related disciplines.
Atomistic models typically contain extremely large number of particles,
even though the actual physical dimension may be quite small. For exam-
ple, even a crystal with dimensions below a few micrometers sidelength has
several tens of billions of atoms. Predicting the behavior of such large many-
particle systems under explicit consideration of the trajectory of each particle
is only possibly by numerical simulation, and must typically involve large
computational facilities.
In the continuum viewpoint, materials are treated without explicitly con-
sidering the underlying inhomogeneous microstructure, that is, the basic
assumption is that matter can be divided indefinitely, without a change of
the material properties. Only a few special cases of continuum models can
be solved analytically and written in closed-form solutions. Many problems
require a numerical solution of the governing partial differential equations,
for instance by finite difference methods, boundary element or finite ele-
ment approaches. Many numerical methods used to solve continuum problems
thereby require the discretization of the domain into integration points (e.g.,
mesh generation in the FEM). A higher density of integration points provides
a more accurate solution of the problem, and higher densities of integration
points must be used in regions of large gradients of stresses and deformation.
In atomistic modeling, the discreteness of matter at the ultimate, atomic
scale is explicitly considered. For example, the discreteness of an atomic
lattice in a metal, where atoms are glued to their equilibrium crystal posi-
tions. Therefore in atomistic modeling no spatial discretization is necessary,
since this is given by the atomic distances, providing a natural measure for
spatial discretization. Each atom is considered to be an individual parti-
cle that cannot be divided further. Atomistic models can rarely be solved
analytically, so that most models are implemented in numerical simulations
that model the motion of each atom in the material over the course of the
simulation time span. The simulation thereby carries out a step-by-step inte-
gration to successively progress the timescale during a study. The collective
behavior of the atoms allows one to understand how the material undergoes
deformation, observe phase changes, or investigate many other phenomena.
1 Introduction 17
The observed phenomena in a larger ensemble of atoms provides the links

between the atomic scale to mesoscopic or macroscopic phenomena. However,
since the basic informations of an atomistic simulation are only atomic posi-
tions, velocities, and forces, the interpretation of these numbers can be quite
challenging.
Continuum approaches provide very powerful methods to obtain the solu-
tion of elastic and sometimes plastic problems. What are their limitations?
Clearly, one limitation is the reduction of characteristic material lengthscales
to the sub-100 nm regime. In this regime, continuum mechanical methods
become questionable, as the material inhomogeneities reach lengthscales com-
parable to the overall size of the structure. And therefore, the basic assumption
of continuum theories that material can be divided without changing the
material properties does not hold any more. Another important area is the
introduction of complex geometrical features, such as nanostructures, and also
the existence of complex chemistry that involves discrete reactions that can
severely influence the behavior of a material, making it impossible to neglect
the discrete nature of molecules.
A fundamental challenge of any model and simulation implementation is
the determination of parameters. This is important for continuum approaches,
as many atomistic or molecular mechanisms (e.g., the generation of dislo-
cations, diffusion, creep) are not explicitly described at the relevant scale
of the underlying physics, but rather treated by empirical relations that
mimic the overall collective effect of these processes. Important parameters
for continuum approaches include elastic coefficients (typically referred to as
constitutive equations), cohesive laws that describe the traction–separation
behavior of interfaces, or the inclusion of parameters to describe the conditions
at which plastic deformation begins. The determination of these parameters
and criteria often involves important challenges. Atomistic modeling, on the
other hand, provides a first principles-based approach to this problem, as it
is capable of capturing the elementary physical mechanisms of failure.
The continuum and atomistic viewpoint provide two fundamentally differ-
ent approaches in treating materials, with different appeal and significance for
specific applications. For many applications, the two views are complementary
and the joint use of the approaches can provide much insight into the behav-
ior of materials. The atomistic models provide a fundamental description of
material properties and processes, which enables one to communicate data of
such studies seamlessly to concepts from other scientific disciplines. Atomistic
methods provide a rather general description of matter, since the same atom-
istic model of a material may be suitable to study elasticity problems as well
to study dissipative materials failure such as fracture. Atomistic simulation is
capable of solving the dynamical evolution of equilibrium and nonequilibrium
processes, providing in particular important detailed insight into the physics
of elementary processes.
1.5 Applications: Experimental and Computational

Mechanics
A key aspect in the analysis of mechanical properties of materials is the ability
to test properties experimentally and to enable proper guidance to interpre-
tation of the results. The combination of experimental testing with computa-
tional and theoretical concepts has proven to be a particularly fruitful strategy,
for both scientific discovery and for technological innovations and applications.
Why do we study mechanical properties of materials? Mechanical proper-
ties are of interest throughout a range of lengthscales and timescales, and for
a variety of applications. At larger scales, the integrity of airplanes, ships, or
space shuttles involves the ability to tolerate large mechanical loads. Bridges,
buildings, or dams must withstand enormous pressures and temperature
ranges and different loading conditions. At smaller scales, reaching microme-
ters and scales much below, mechanical properties are quite important for the
reliability of devices. For instance, integrated circuits in electronic microde-
vices must be able to provide electrical conductivity, but also persist large
ranges of temperatures that lead to large strains and large stresses. Mechan-
ical properties are also significant for the understanding of biological systems
and biological processes. The mechanical properties of biological tissue is crit-
ical in several other areas in biology, in particular at the subcellular scale and
for the behavior of biological tissue, for example during injuries. Biological
systems often use mechanical deformation as signals for cell–cell interaction
or cell–tissue interaction as an important part of the physiologic function.
Figure 1.9 summarizes specific lengthscales and timescales associated with
modeling the mechanical properties of materials for materials and structures
of wide interest. This plot illustrates the grand challenges associated with
modeling and simulation the complex phenomena of mechanical deformation,
spanning many orders of timescales and lengthscales.
1.5.1 Experimental Techniques
Experimental techniques that enable one to probe the mechanical properties

of materials and structures has significantly developed over the past decades.
In particular, a rich set of experimental tools suitable to probe the mechanical
behavior at micro- and nanometer lengthscales is now available. Since many
fundamental deformation mechanisms operate at that scale, these approaches
provide us with the opportunity to understand and test material theories and
simulation results at unparalleled scales.
Mechanical testing at scales between millimeters and meters is often car-
ried out by using tensile test machines or through indentation experiments and
many variations of these approaches. Mechanical tensile test machines provide
a means to apply controlled forces to samples, while measuring the elonga-
tion. This information is then used to produce force–extension or stress–strain
curves as shown in Fig. 1.2. In indentation experiments, the compression of a
1 Introduction 19
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
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
biomechanics to probe the mechanical characteristic of DNA, proteins, cells,

and biological tissue.
Experimental techniques for the analysis of materials is a very wide field.
Due to the scope of this book on modeling and simulation we refer the reader
to the literature for additional details. Throughout this text, we will make ref-
erence to experimental results and experimental techniques that complement
the studies of materials failure reviewed here.
1.5.2 Example Applications: The Significance of Mechanics
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
Fig. 1.11 Nanomechanical experiments of bending deformation of 200-nm gold

nanowires. Subplot (a) depicts a schematic of a fixed wire in a lateral bending
test with an AFM tip. Subplots (b–e) depict AFM snapshots of the mechanical
deformation the nanowire. Subplot (b) depicts results after elastic deformation,
subplots (c) and (d) shows results after successive plastic manipulation, and subplot
(e) shows an SEM image following the bending test. The SEM picture agrees in detail
with the AFM image shown in subplot (d). All scale bars are 1 µm. Reprinted with
permission from Macmillan Publishers Ltd, Nature Materials [3] c 2005
Nanostructures and Nanomaterials
At nanoscale, significant recent interest has been devoted to the mechani-

cal properties of nanowires and tubular structures such as carbon nanotubes.
The miniaturization of materials from macroscales or microscales to the scales
of several nanometers leads to severe changes in the mechanical properties.
Such observations have been made for many nanomaterials that have been
rigorously studied over the past decades. For instance, the strength of nanos-
tructures is often greater than the strength of its macroscopic counterparts
made out of the same material. This is believed to be due to geometric
confinement, the lack of structural defects, or a combination of these effects.
In particular, metallic nanowires have received considerable interest as
novel interconnects for electronic devices or as components of both electrical
and electromechanical devices. Figure 1.11 depicts results of AFM manip-
ulations of a 200-nm gold nanowire. It was found that whereas Young’s
modulus is independent of the diameter of the nanowire, the yield strength is
largest for the nanowires with the smallest diameter. Notably, the strength of
nanowires reaches up to 100 times that of corresponding bulk materials. This
can be explained partly based on the fact that the conventional mechanism
of dislocation nucleation and propagation is hindered due to the geometric
confinement. Thus the material reaches strengths that are much greater than
observed in conventional materials, approaching the strength values compa-
rable to the theoretical strength of materials. The strength of nanowires is
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.
Biomechanics and Biomedical Applications
The fundamental components of life are structures with characteristic dimen-

sions on the order of nanometers as represented by DNA or proteins. The
analysis of their mechanical behavior not only is the key to enable the devel-
opment of new biomaterials (e.g., prostates, material replacements), but also is
crucial for the advancement of the understanding of the physical mechanisms
of biological processes. Injuries such as bone fracture or scars represent severe
plastic deformations and ruptures of elementary components of organisms.
Even though the understanding of associated basic deformation mechanisms
is still in its infancy, the ability to probe mechanical properties and processes
at ultra-small scales has already contributed significantly to the field.
An interesting aspect of mechanical properties is the relation with dis-
eases. Many diseases are associated with a change of mechanical properties of
cells or tissues, which renders mechanics quite important for the study and
diagnosis of human diseases. For example, at the cellular scales, the change of
mechanical properties during disease progression has been studied in the con-
text of malaria, where it was discovered that the disease has a dramatic effect
of making red blood cells much stiffer [74, 75]. The change of the mechanical
properties of individual cells has thereby major consequences for the spread-
ing of malaria in the human body. Cells infected by malaria show more than
ten times increased stiffness, whereas healthy cells are significantly more elas-
tic. Since the cells are so stiff, it is impossible for them to squeeze through
very thin blood vessels and thereby adversely affect other biological tissues.
Figure 1.12 depicts the results of an optical tweezers experiment, illustrating
this effect [75]. Clearly, the cells infected by malaria are much less elastic at
the identical force compared with an uninfected cell. Finite element compu-
tational mechanics has been used to simulate the micromechanical process
of cellular deformation. Figure 1.13 depicts images of a comparison between
experimental snapshots and the results of a three-dimensional finite element
simulation. Further, a combined optical tweezers – molecular dynamics study
of the deformation mechanics of red blood cells has been reported in [76], illus-
trating the potential of combining molecular modeling with nanomechanical
experiments.
Similar observations that illustrate the importance of mechanical proper-
ties as signature of human diseases have been made in the case of cancer, where
1 Introduction 23
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
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
tissue by muscle cells is an important part of the physiological function in our

body, involving the concerted motion of millions of molecules, involving the
continuous rupture and reformation of chemical bonds. Failure of biological
tissues is observed for instance in injuries, when large external forces or rapid
displacements lead to a disruption of the healthy tissue formation.
The fracture of bone is one of the most prominent examples of such
an injury. Many studies in the past years, using advanced mechanical test-
ing methods across a wide range of scales, have contributed to significantly
improved understanding of how bone breaks. Figure 1.15 illustrates a sequence
of microscopic mechanisms that shows how a small crack propagates in bone
tissue. The combination of mechanical testing with new imaging techniques
1 Introduction 25
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.
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
Figure 1.17 depicts the mechanically induced deformation of a protein

structure as reported in [7]. This field was pioneered in the early 1990s,
when molecular dynamics simulations of protein unfolding were first car-
ried out [6]. The combination of atomistic modeling and experiments with
AFM or optical tweezers has been a particularly fruitful combination. For
a review of nanomechanical experiments of biological structures please see
also [74, 77–80].
The examples reviewed in this section illustrate new frontiers of mechanics,
at very small scales, in nanomaterials, or in exceedingly complex biological
structures and tissues. Most methods and results reviewed in this book are
not yet applicable to these complex materials. However, maybe with future
development and research it will be possible to achieve a similar understanding
of the fundamental failure mechanisms as it exists for crystalline systems
today.
1 Introduction 27
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
1.6 Outline of This Book

This book is organized into three main parts: An introduction, a part dis-
cussing the basic concepts of atomistic, continuum, and multiscale approaches,
as well as a large part that reviews examples of modeling of deformation and
fracture phenomena in a variety of materials.
The discussion begins with the introduction of the conventional concepts
of fracture and deformation mechanics, reviewing basic contributions over the
past five and more decades. Along with the presentation of the conventional
theory, we present molecular simulation studies of these phenomena illustrat-
ing how atomistic approaches can be used to model these processes.
The book concludes with a focus on small-scale materials phenomena.
Along the way we will emphasize on analogies and synergies from the interac-
tions between the different fields of studies. This book reviews the techniques
that are introduced in specific examples or case studies. The prime goal of this
approach is to convey to the reader the underlying approach, the methods and
the strategy used to build, use and analyze the computational and theoretical
models. Many examples and case studies are taken from earlier studies of the
author of this book. The discussion should therefore not be considered as a
Fig. 1.17 Large deformation of a protein, here an example of unfolding of the

enzyme lysozyme, result of a reactive force field simulation. The distance between
the ends of the protein (Cα -atom of the terminal residues) is continuously increased
by applying a continuously increasing force [7]. As the force is increased, the pro-
tein molecule undergoes significant structural changes relative to its initial folded
configuration
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
Basics of Atomistic, Continuum and Multiscale

Methods
2
Basic Atomistic Modeling
Atomistic modeling provides a fundamental description of the materials

behavior and materials deformation phenomena. Molecular dynamics simula-
tions represent the numerical implementation to solve the equations of motion
of a system of atoms or molecules, resulting in the dynamical trajectories of
all particles in the system. The purpose of this chapter is to present an intro-
duction into molecular dynamics modeling and simulation approaches. The
discussion includes the physical basics, numerical implementation and exam-
ples of atomistic models for specific materials, as well as a brief introduction
into multiscale simulation methods. We also review analysis methods, in par-
ticular visualization schemes that can be used to extract useful information
from large atomistic systems.
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
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)
is kindly referred to other literature). We briefly review statistical mechan-

ics concepts that provide the theoretical and numerical basis for property
calculation from the results of molecular dynamics simulations. We discuss
the calculation of temperature, measures for the geometry of a particular
atomic system, methods to analyze and display crystal defects, and some
correlation functions that enable one to predict transport properties from
molecular dynamics studies. As illustrated in Fig. 2.1, due to the fundamental
nature of the description of the material behavior, molecular dynamics can
be used to study material properties at the intersection of different scientific
disciplines.
2.2 Modeling and Simulation
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
to the development of a mathematical model of a physical situation, whereas

simulation refers to the procedure of solving the equations that resulted from
model development. Models are often simplifications or idealizations of rather
complex physical systems or phenomena. A key aspect of model development
is the ability to map the essential physics features of a system into a descrip-
tion. M.F. Ashby of Cambridge University used the example of a subway map
to illustrate the concept of model building [85]: A model is an idealization. Its
relationship to the real problem is like that of the map of the London tube trains
to the real tube systems: a gross simplification, but one that captures certain
essentials. The map misrepresents distances and directions, but it elegantly
displays the connectivity. The quality or usefulness in a model may be mea-
sured by its ability to capture the governing physical features of the problem.
Ashby states that all successful models unashamedly distort the inessentials in
order to capture the features that really matter. At worst, a model is a concise
description of a body of data. At best, it captures the essential physics of the
problem, it illuminates the principles that underline the key observations, and
it predicts behavior under conditions which have not yet been studied.
The concept of model building is illustrated in Fig. 2.2, here shown for the
subway system in the Boston area. The comparison of the left and right panels
illustrates how models can facilitate to capture the essential information and
features of a physical system.
The concepts of modeling and simulation are intimately coupled: Without
model development, a simulation can not be carried out. Often, the partial dif-
ferential equations itself that may be a direct result from model development
do not allow to draw significant conclusions for a physical system at hand,
until simulations are carried out. This is nicely summarized in a phrase coined
by Sidney Yip of MIT [86] who stated that Modeling is the physicalization of
a concept, simulation its computational realization.
Both modeling and simulation have their specific challenges. The tasks
associated with modeling requires insight into the physics of the system, its
constituents or the behavior of the particles. The setup of the simulation
requires knowledge in the field of numerical techniques that are suitable to
solve complicated systems of partial differential equations, or to make compu-
tation proceed fast on modern supercomputers. To make efficient use of results
from simulation, strategies need to be used to analyze and interpret this data.
As indicated earlier, the results of atomistic simulations are merely numbers
that represent the position and velocities of atoms at different time steps.
Making sense of this huge amount of information can be a daunting task.
This becomes more evident as the simulation sizes increase to systems with
billions of atoms. Even with today’s largest computers, system sizes with only
a few billion atoms can be simulated, whereas a cubic centimeter of material
already contains more than 1023 atoms.
This book provides illustrates techniques and approaches for both mod-
eling and simulation. The concepts will be applied to specific problems
of describing materials failure phenomena. However, the concepts can be
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
transferred to other applications where similar methods could be fruitfully

applied, such as self-assembly, diffusion studies, or studies of phase transfor-
mation.
2.2.1 Model Building and Physical Representation
Together with data analysis, model building and finding an appropriate

physical representation is probably the most difficult task in computational
materials science. It is imperative that great care must be taken when models
are built, and when results of simulations are analyzed.
Atomistic simulations have proven to be a powerful way to investigate
the complex behavior of dislocations, cracks, and grain boundary processes
at a very fundamental level. Atomistic methods have gained an increasingly
important role and level of acceptance in modern materials modeling. One of
the strengths and the reason for the great success of atomistic methods is its
very fundamental viewpoint of materials phenomena. For instance, the only
physical law that is used to simulate the dynamical behavior is Newton’s law,
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.
2.2.2 The Concept of Computational Experiments
An increasing number of researchers now consider the computer as a tool to

do science, similar as experimentalists use their lab to perform experiments.
Computer simulations have thus sometimes been referred to as “computer
experiments” or “computational experiments.”
The art of a computational experiment is to (1) build an appropriate model

of the physical situation, and to (2) construct it in such a way that the results
obtained by numerical simulation of this model can be utilized to advance the
understanding of a particular phenomenon. The third important step is to (3)
analyze and interpret the results computational experiments to advance the
understanding of the simulated process.
Computational experiments must be set up with care. For instance, if a
model is highly accurate but contains a very large number of numerical param-
eters, it may be impossible to understand how these parameters are related to
a phenomenon of interest. Only after reduction of the large number of param-
eters into a simpler set of reduced variables it is possible to draw significant
conclusions about the behavior of the system. Finding these reduced variables
is a central aspect in designing clever computational experiments.
2.3 Basic Statistical Mechanics

Statistical mechanics provides methods that help us yo analyze molecular
dynamics simulation and to interpret results from these simulations. In par-
ticular, it leads to a direct link between an ensemble of microscopic states and
the corresponding macroscopic thermodynamical properties.
The conversion of the microscopic information to macroscopic observ-
ables such as pressure, stress tensor, strain tensor, energy, heat capacities,
and others requires theories and strategies developed in the realm of sta-
tistical mechanics. Atomistic data (e.g., the pressure tensor) is not valid
instantaneously but needs to be averaged over multiple configurations of the
microscopic system.
One of the most central and probably most useful theorems in the practical
application of atomistic simulation is the Ergodic hypothesis. The Ergodic
hypothesis states that the ensemble average of a property A equals the time
average (the symbol . describes an averaged variable):
AEnsemble = ATime . (2.1)
This is a most useful relation that enables the calculation of thermodynamical

properties by simply averaging over sufficiently long time trajectories (and
thereby measuring the appropriate ensemble properties).
The ensemble average of a property A is defined as

AEnsemble = A(p, r)ρ(p, r)dpdr, (2.2)
p r
with pi = mi vi as the linear momentum of particle i, and p = {pi } being

the set of all linear momentums in the system, for i = 1 . . . N . Similarly,
r = {ri } represents the position vector of all particles. The system state is
uniquely defined by the combination (p, r) since the Hamiltonian H = H(r, p).
In (2.2), the function ρ(p, r) is the probability density distribution, which is

defined as
1 H(r, p)
ρ(p, r) = exp − (2.3)
Q kB T
with
H(r, p)
Q= exp − dpdr. (2.4)
p r kB T
To evaluate these expressions, we would need to known any possible state of
the system, characterized by all possible values of p and r. Obtaining this
information is very difficult, which immediately shows the significance of the
Erdogen hypothesis. The time average from a molecular dynamics simulation
can be calculated by
1
M
ATime = A(p, r), (2.5)
M i=1
where M is the number of measurements taken.
The Ergodic hypothesis further lays the foundation for Monte Carlo tech-
niques. Whereas molecular dynamics generates trajectories over time, Monte
Carlo generates those within the constraint of an ensemble average. The
Ergodic hypothesis states that both viewpoints are equal, allowing one to cal-
culate thermodynamical properties using Monte Carlo. In this sense, Monte
Carlo schemes generate a number of possible microscopic states along with
specific probability densities ρ, which are then summed up discretely to obtain
an estimate for the observed variable.
2.4 Formulation of Classical Molecular Dynamics
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,
Fig. 2.3 Molecular dynamics generates the dynamical trajectories of a system

of N particles by integrating Newton’s equations of motion, with suitable ini-
tial and boundary conditions, and proper interatomic potentials, while satisfying
macroscopic thermodynamical (ensemble-averaged) constraints, leading to atomic
positions ri (t), the atomic velocities vi (t), and their accelerations ai (t), all as a
function of time, for all particles i = 1 . . . N , each of which has a specific mass mi
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.
However, the equation can be solved by discretizing the equations in time. We

note that the spatial discretization for the problem is given by the atom size,
as discussed in Sect. 1.4.
2.4.1 Integrating the Equations of Motion
A simple solution strategy is to develop a stepping method that gives new

coordinates and velocities from the old ones, for each particle i, such as
ri (t0 ) → ri (t0 + ∆t) → ri (t0 + 2∆t) → ri (t0 + 3∆t) . . . (2.10)
A numerical scheme can be constructed by considering the Taylor expan-

sion of the position vector ri :
1
ri (t0 + ∆t) = ri (t0 ) + vi (t0 )∆t + ai (t)∆t2 + . . . (2.11)
2
and
1
ri (t0 − ∆t) = ri (t0 ) − vi (t0 )∆t + ai (t)∆t2 + . . . (2.12)
2
Adding these two equations together yields
ri (t0 + ∆t) = −ri (t0 − ∆t) + 2ri (t0 ) + ai (t)∆t2 + . . . (2.13)
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
In the Leap-frog algorithm, the positions are updated as

1
ri (t + ∆t) = ri (t) + vi (t + ∆t)∆t (2.15)
2
and
1 1
vi (t + ∆t) = vi (t − ∆t) + ai (t)∆t. (2.16)
2 2
Velocity Verlet Algorithm
In the Velocity Verlet algorithm, the positions are updated as

1
ri (t + ∆t) = ri (t) + vi (t)∆t + ai (t)∆t2 (2.17)
2
where
1
vi (t + ∆t) = vi (t) + (ai (t) + ai (t + ∆t)) ∆t. (2.18)
2
2.4.2 Thermodynamic Ensembles and Their Numerical

Implementation
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.
Ensemble Ensemble name

NV E Microcanonical ensemble
NV T Canonical ensemble
NP T Isobaric–isothermal ensemble
µV T Grand canonical ensemble
Table 2.1 Overview of various thermodynamical ensembles (the parameter µ is the
chemical potential)
To illustrate the approach of modifying the equations of motion to obtain a

specific thermodynamical ensemble, here we briefly review a simple algorithm
to enable an N V T ensemble, the Berendsen thermostat. The approach is based
on the idea to change the velocities of the atoms so that the temperature
(which is a direct function of the atomic velocities, as discussed later on in
Sect. 2.8.1) approaches the desired value, mimicking the effect of a heat bath.
This is realized by calculating a rescaling parameter λ

∆t T
λ= 1+ , (2.19)
τ Tset−1
where ∆t is the molecular dynamics time step and τ is a parameter called

“rise time” that describes the strength of the coupling to the hypothetical
heat bath. The velocities are then rescaled according to this parameter, where
the new velocity vectors are given by
vnew,i = λvi (2.20)
for each atom i.

Other approaches to enable the N V T ensemble include methods based
on Langevin dynamics and the Nose–Hoover scheme. For N P T ensembles,
the Parrinello–Rahman approach provides a popular choice for an algorithm.
In this method, in addition to adjusting the temperatures to approach the
desired control value, the pressure is adjusted by changing the cell size of the
simulated system.
States with high energy will occur less often, states with low energy more
often. Each microscopic state has a certain probability associated with the
corresponding energies. During the integration of the equations of motion,
molecular dynamics naturally samples these microscopic configurations and
provides a collection of snapshots that after averaging correspond to the
proper macroscopic state. The obtained trajectories can then be used to
calculate thermodynamical properties by simply averaging over the sampled
configuration without further weighting.
For example, a selection of N particles in a box at pressure P , temperature
T , and volume V has many microscopic configurations (p, r) that all corre-
spond to the same thermodynamical macroscopic state. Molecular dynamics
can for instance be used to calculate the pressure for a given temperature
and system volume, by solving the dynamical evolution of the system over
long time scales. In this spirit, molecular dynamics enables one to sample the
phase space for admissible configurations, and giving them proper weighting.
According to the Ergodic hypothesis (see (2.1)), the molecular dynamics time
average is equal to ensemble average, and system properties can be calculated
with both methods. The results will be identical.
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”)
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
u(t) = u(t) + u (t), (2.21)
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
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.
2.4.3 Energy Minimization
Energy minimization is an approach during which the potential energy of the

system, at zero temperature, is minimized. Energy minimization corresponds
to the physical situation of cooling down a material to the absolute zero
point. Methods in which the deformation behavior of a material or structure is
probed during continuous energy minimization is also referred to as molecular
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. 2.6 Example result of an energy minimization, here an example of minimiz-

ing the structure of a solvated protein (lysozyme). As the number of iterations
progresses, the total potential energy decreases, until it converges and reaches a
constant value (see [8] for further details)
2.4.4 Monte Carlo Techniques
Statistical mechanics provides a theoretical framework to link a number of

microscopic states to macroscopic thermodynamical variables. To achieve this
link, one needs to obtain samples of microscopic states, for instance by using
a dynamical simulation (e.g., molecular dynamics).
An alternative approach to generate this data is to calculate the appro-
priate ensemble averages directly. This is done in Monte Carlo techniques by
sampling phase and state space.
Monte Carlo Techniques: Brief Introduction
The key task in computing the appropriate ensemble average is achieved

by repeated random sampling. This is achieved by generating system states
Fig. 2.7 Illustration of generation of random perturbation from an initial state A

toward a state B, as typically performed in Monte Carlo schemes
according to suitable Boltzmann probabilities [90, 91]. The procedure can be

summarized as follows (Metropolis–Hastings algorithm):
1. Draw random numbers from a random number generator.
2. Advance system according to these random numbers (e.g., for the case of
a molecular structure, move atoms accordingly, as illustrated in Fig. 2.7).
3. Accept or reject new configuration, according to an energy criterion.
4. If N < NA , back to 1. Otherwise, continue with 5.
5. The set of configurations obtained based on this scheme is used to calculate
ensemble properties.
Many Monte Carlo algorithms employ such a procedure to determine a new
state for a system from a previous one. Thereby, the specific moves can be
chosen arbitrarily, which makes this method very widely applicable. However,
a drawback or limitation of this method is that it requires additional knowl-
edge of the system behavior, in particular how the system may evolve as it is
required for the generation of new configurations.
Figure 2.8 summarizes one of the most popular Monte Carlo schemes, the
Metropolis–Hastings algorithm.
Comparison of Monte Carlo and Molecular Dynamics
In contrast to Monte Carlo, molecular dynamics enables one to obtain actual

deterministic trajectories, and thus provides detailed information about the
full dynamical trajectories. Molecular dynamics can model processes that are
characterized by extreme driving forces and that are nonequilibrium processes.
A prominent example for which molecular dynamics is particularly suitable
is fracture. Provided expressions exist for the atomic interactions, molecular
dynamics modeling provides an excellent physical description of the fracture
processes, as it can naturally describe the atomic bond breaking processes.
Other modeling approaches, such as the finite element method, are based on
empirical relations between load and crack formation and/or crack propaga-
tion. In contrast, molecular dynamics does not require such input parameters.
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
In principle, all parameters required for a molecular dynamics simulation can

be derived from first principles, or quantum mechanical calculations (see also
discussion above). Monte Carlo can typically only be used for equilibrium
processes, as it does not provide information about how a system goes from
an unstable state A to a stable state B.
Many materials deformation processes such as fracture are examples for
such phenomena. Thus, in the remainder of this book we focus on molecular
dynamics methods, as they are capable of providing insight into the fracture
mechanisms, which is an important aspect of modeling and understanding
how materials fail under extreme conditions.
2.5 Classes of Chemical Bonding

The behavior of molecules is intimately linked to the interactions of atoms,
which are fundamentally governed by the laws of quantum chemistry. In met-
als, for example, bonding is primarily nondirectional, and can be characterized
by positive ions embedded in a gas of electrons (this is often referred to as the
electron gas model). Other materials show greater chemical complexity – often
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.
For the implementation of molecular dynamics, we must have mathemat-

ical expressions available that provide models for the energy landscape of
these chemical interactions. In other words, reviewing the formula provided
in (2.54), we must know how the potential energy stored in a bond changes
based on the geometry or the position of the atoms. These models are referred
to as force fields or interatomic potentials. The term “force field” is often used
in the chemistry community, whereas the term “potential” is frequently used
in the physics community. Here we will use both terms, as it fits for the
corresponding force field expressions.
2.6 The Interatomic Potential or Force Field:

Introduction
Figure 2.10 depicts a fundamental simplification made in classical molecular
dynamics to replace the atom as a three-dimensional structure by a single
point with a finite mass. This simple picture also illustrates the grand chal-
lenge in developing expressions, to somehow and as accurately as possible
describe the (often complex) effect of the electrons on the atomic interac-
tions. The goal of interatomic potentials or force fields is to give numerical or
analytical expressions that estimate the energy landscape of a large particle
system. This energy landscape is therefore the fundamental input into molec-
ular simulations, in addition to structural information (position of atoms, type
of atoms, and their velocities and accelerations).
In this section, we will review a variety of classical approaches in mod-
eling atomic interactions, ranging through different levels of accuracy and
complexity.
Numerous potentials with different levels of accuracy have been proposed,
each having its disadvantages and strengths. The approaches range from accu-
rate quantum-mechanics-based treatments (e.g., first-principle density func-
tional theory methods [87] or tight-binding potentials [92]), reactive potentials
[93] to multibody potentials (e.g., embedded atom approaches as proposed
in [94]) to the most simple and computationally least expensive pair potentials
(e.g., Lennard-Jones) [9,84]. One of the first molecular dynamics studies was a
Lennard-Jones model of argon in 1964 [83]. Previous studies used hard-sphere
models to describe phase transformations [81,82]. Many potential expressions
are fit carefully so that they closely reproduce the energy landscape predicted
from quantum mechanics methods, while retaining computational efficiency.
The interatomic potential thereby allows to generate a direct link between the
empirical molecular dynamics methods and quantum chemistry.
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
dynamics methods, as well as numerical continuum mechanics methods.

Quantum-mechanical-based treatments are typically limited to very short
time- and length scales, on the order of a few nanometers and picoseconds.
The assumption of empirical interactions in classical molecular dynamics
scheme significantly reduces the computational burden, and the lengthscale
and timescale that can be reached are dramatically increased, approaching
micrometers and several nanoseconds. For comparison, we include also con-
tinuum mechanics-based simulation tools that can treat virtually any length
scale, but they may lack a proper description at small scales, and they are
therefore often not suitable to describe materials failure processes in full detail
(see discussion in Sect. 1.4). Mesoscopic simulation methods such as discrete
dislocation dynamics can bridge the gap between molecular dynamics and
continuum theories by generating an intermediate scale at which clusters of
atoms or small crystals are treated as a single particle [50, 95–101].
The remainder in this section will be focused mostly on empirical potential
expressions that are suitable for the study of mechanical properties of mate-
rials. How do empirical potentials describe the various chemical interactions?
Often, energy contributions from covalent atom–atom interactions, electro-
static interactions, vdW interactions, and others are summed up individually,
so that
U = UElec + UCovalent + UvdW + UH-bonds + . . . . (2.26)
The challenge is how can these individual terms be approximated, most accu-
rately, for a specific material? In the following sections we will describe some
of the most common empirical potentials that provide such approximations.
2.6.1 Pair Potentials

We begin with the simplest atom–atom interactions for which the potential
energy only depends on the distance between two particles. The total energy
of the system is given by summing the energy of all atomic bonds over all N
particles in the system. The total energy is then given by
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
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
Fig. 2.13 Replacing a full-electron representation of atom–atom interaction by a

potential function that only depends on the distance r between the particles

Ni
Ui = φ(rij ) (2.29)
j=1
where Ni is the number of neighbors of atom i (this expression corresponds to

the first summation in (2.27)). Usually, the number of neighbors considered
for inclusion in the potential energy calculation is limited to the second or
third nearest neighbors. Popular pair potentials for the simulation of metals
include the Morse potential [103] and the Lennard-Jones (LJ) potential, which
are described, for instance, in [9, 84, 104]. The LJ 12:6 potential is defined as
6
12
σ σ
φ(rij ) = 40 − . (2.30)
rij rij
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
for copper as reported in [9]. It also illustrates important points in the LJ

potential, such as the equilibrium distance r0 and the point of largest force
Fmax,LJ .
Fig. 2.14 Plot of the LJ potential and its derivative (for interatomic forces) in a
parametrization for copper as reported in [9]
Another popular pair potential is the Morse potential, defined as

2
φ(rij ) = D [1 − exp (−β(rij − r0 ))] . (2.33)
A fit of this potential to different metals (as well as different forms of the

Morse potential) can be found, for instance in [106]. The parameter r0 stands
for the nearest neighbor lattice spacing, and D and β are additional fitting
parameters. The Morse potential is computationally more expensive than the
LJ potential due to the exponential term (however, this is more realistic for
many materials).
An advantage of using pair potentials is the computational efficiency.
Another important advantage is that fewer parameters are involved, which
may simplify parameter studies and the fitting process to different materi-
als. For example, the LJ potential has only two parameters, and the Morse
potential has only three parameters.
The potentials given by (2.30) and (2.33) are strongly nonlinear functions
of the radius r. In some cases it is advantageous to linearize the potentials
around the equilibrium position and define the so-called harmonic potential
(see also Fig. 2.5)
1
φ(rij ) = a0 + k(rij − r0 )2 (2.34)
2
where k is the spring constant r0 the equilibrium spacing, and a0 is a constant

parameter. An important drawback of pair potentials is that elastic properties
of metals cannot be modeled correctly.
2.6.2 Multibody Potentials: Embedded Atom Method for Metals
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
To accurately represent the change in bond properties at a surface, a

description is needed that considers the environment of an atom to determine
the bond strength. Therefore, the bond energy between two particles is no
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].
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
2.6.3 Force Fields for Biological Materials and Polymers
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
Fig. 2.17 Chemical complexity in proteins involves a variety of chemical elements

and different chemical bonds between them. The snapshot shows a small alpha-
helical coiled coil protein domain
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
has the form:

Usystem = Ubond + Uangle + Utorsion + UCoulomb + UvdW + . . . , (2.36)
representing an approach to split up different energy contributions. However,
most of the terms, in particular those describing the bond interactions, are
harmonic expressions.
An expression similar to (2.34) describes the energy contributions to the
bond stretching term Ubond (the total energy due to bond stretching is
obtained by summing over all pairs of atoms). For bond bending between
three atoms i, j, and k,
1
φbend = b0 + kbend (θijk − θ0 )2 (2.37)
2
where θ0 is the equilibrium bond angle (depends on the particular triplet of
atoms considered), and θ is the angle between the three atoms i, j, and k.
For torsional energies between a group of four atoms,
1
φtorsion = t0 + ktorsion (1 − cos(θ1 )) (2.38)
2
where θ1 is the torsional angle and ktorsion is the appropriate spring constant
that describes the magnitude of the resistance to torsional deformation of a
group of atoms.
The contributions from all pairs of atoms (for bond stretching), all triplets
of atoms (for bond bending), and quadruples of atoms (for bond rotation) are
summed up to yield the total potential energy of the system:

Ubond = φbond , (2.39)
pairs

Ubend = φbend , (2.40)
triplets
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

N N
qi qj
UCoulomb = (2.43)
r
j=1 1 ij
j=i=1
The calculation of electrostatic interactions provides a significant computa-

tional burden, as the particle interactions are long range. In particular when
using small, periodic unit cell approaches, the interactions from one cell side
to another must be considered very carefully. Several methods have been
developed to address this problem. The most prominent technique is the
particle-mesh Ewald method (PME), which significantly reduces the computa-
tional effort in accurately calculating these interactions. Most modern molec-
ular dynamics codes have implementation of the PME technique or related
approaches. The vdW terms are typically modeled using Lennard-Jones 6–12
terms. Both the vdW terms and the Coulomb terms contribute to the exter-
nal or nonbonded interactions. H-bonds are often included in the vdW terms.
Some flavors of CHARMM-type potential provide explicit expressions for H-
bonds that involve angular terms to provide a more refined description of the
spatial orientation between H-bond acceptor and donor pairs.
The parameters in such force fields are often determined from quantum
chemical simulation models by using the concept of force field training. Which
specific terms of the force field formulation are considered for a particular
chemical bond are controlled by atom type names. That is, each atom type is
not only specified by its element name but also by a tag that denotes which
type of chemical bond is attached to it. For instance, the tag CA refers to an
aromatic carbon atom, CC to a carbonyl carbon atom, and C to a polar car-
bon atom (e.g., in a protein backbone). Thus, a critical step before a molecular
simulation with a CHARMM-type force field can be carried out is the assign-
ment of these atom types. The information of element types as provided in
the Protein Data Bank, for instance, is insufficient. Automated programs have
been developed to carry out this typing tasks by comparing particular molec-
ular structures with templates in a large database, and assigning appropriate
atom types according to a best fit comparison.
Force fields for protein structures typically also include simulation models
to describe water molecules (e.g., TIP3P, TIP4P, SPC and SPC/E, ST2),
an essential part of any simulation of protein structures. These water force
fields are composed of similar harmonic, bond angle, dispersive, and Coulomb
expressions.
Table 2.2 summarizes several popular choices for organic force fields.
2.6.4 Bond Order and Reactive Potentials
An important class of multibody potentials is based on the concept of bond

orders, a model particularly suitable to describe the forces in covalently
bonded materials (that is, chemical bonds that have a strong directional
dependence). This is important, for instance, in carbon-based materials, such
as carbon nanotubes. The key concept is that the bond strength between two
Force field name Primary application Notes

CHARMM Proteins
AMBER Proteins
CVFF Small molecules
CFF Organic substances
GROMOS Proteins, organic molecules Implemented in
GROMACS code
MM3 Organic molecules
UFF Organic molecules Universal force field
DREIDING Organic molecules
TIP3P, TIP4P Water molecules
SPC, SPC/E Water molecules
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
φ(rij ) = φRepulsion (rij ) − Mij φAttraction (rij ), (2.44)
where φRepulsion (rij ) is a repulsive term and φAttraction (rij ) is an attractive

term. The parameter Mij that multiplies the attractive interactions represents
a many-body interaction parameter. This parameter describes how strong the
attraction is for a particular bond, from atom i to atom j. Most importantly,
the parameter Mij can range from zero to one, and describes how strong this
particular bond is, depending on the particular chemical environment of atom
i. It can thus be considered a normalized bond order, following the concept of
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
hypothetical polytypes of carbon, comparing the prediction from experiment

with the results obtained from the potential.
These equations immediately lead to a relationship between bond length,
binding energy, and coordination, through the parameter Mij . The modula-
tion of the bond strength effectively leads to a change in spring constant as a
function of bond environment,
k(r) ∼ k0 Mij (Z, δ). (2.46)
Note that k0 is a reference spring constant, which is modulated by the

atomic environment that is essentially dependent on the bond radius. This
method has been very successful to describe the interatomic bonding in several
covalently bonded materials, for example the C–C bonds in diamond, graphite,
and even hydrocarbon molecules [125, 126]. The coordination number is a
concept also used in lattice systems, for example crystals. In organic molecules,
the coordination number can be thought of as the amount of covalent bonds
that an atom has made.
Reactive Force Fields
Many attempts have failed to accurately describe the transition energies

during chemical reactions using more empirical descriptions than relying on
purely quantum mechanical (QM) methods.
Reactive force fields represent a strategy to overcome some of the limita-
tions classical force fields, in particular the fact that these descriptions are not
able to describe chemical reactions. In fact, the behavior of chemical bonds at
large stretch has major implications on the mechanical response, as it trans-
lates into the properties of molecules at large-strain, a phenomenon referred
to nonlinear elasticity or hyperelasticity.
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
Reactive potentials are based on a more sophisticated formulation than

most nonreactive potentials. A bond length to bond order relationship is used
to obtain smooth transition between different bond types, including single
bonds, double bonds and triple bonds. Typically, all connectivity-dependent
interactions (that means, valence and torsion angles) are formulated to be
bond-order dependent. This ensures that their energy contributions disap-
pear upon bond dissociation so that no energy discontinuities appear during
reactions (see Fig. 2.20). The reactive potential also features nonbonded
interactions (shielded van der Waals and shielded Coulomb).
Several flavors of reactive potentials have been proposed in recent years.
Reactive potentials can overcome the limitations of empirical force fields and
enable large-scale simulations of thousands of atoms with quantum mechanics
accuracy. The reactive potentials, originally only developed for hydrocarbons,
have been extended recently to cover a wide range of materials, including
metals, semiconductors, and organic chemistry in biological systems such
as proteins. Here we review in particular the ReaxFF formulation [93, 128–
131, 131]. The most important features of the class of ReaxFF reactive force
fields are
• A bond length to bond order relationship is used to obtain a smooth
transition of the energy from a nonbonded to single, double, and triple
bonded molecules.
• All connectivity-dependent interactions (that is, valence and torsion
angles) are made bond-order dependent: Ensures that their energy con-
tributions disappear upon bond dissociation.
• Features shielded nonbonded interactions that include van der Waals and
Coulomb interactions, without discrete cutoff radius to ensure a continuous
energy landscape.
• ReaxFF uses a geometry-dependent charge calculation scheme (similar to
the Charge Equilibration method, QEq [132]) that accounts for polariza-
tion effects and redistribution of partial atomic charges as the molecule
or cluster of atoms changes its shape (e.g., determine the partial atom
charges qi and qj ).
• Most parameters in the formulation have physical meaning, such as
corresponding distances for bond order transitions, atomic charges and
others.
• All interactions in ReaxFF feature a finite cutoff of 10 Å.
The total energy of a system in the ReaxFF model is expressed as the sum
of different contributions that account for specific chemical interactions.
Usystem = Ubond + Uunder + Uover + Uangle + Utors +
Uconj + UH-bonds + UCoulomb + UvdW (2.47)
The term Ubond describes the energy contributions due to covalent bonds.
The terms Eunder and Uover describe energy penalties for under- and over-
coordination. Angular effects are included in Uangle , and contributions due to
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
is used (the parameter γij is a force field parameter that is adapted to

reproduce the orbital overlap contributions).
We explain the approach used in ReaxFF for the example of the term
Ubond . First the bond order is calculated, according to
bππ
b b
rij σ rij π rij
BOij = exp aσ + exp aπ + exp aππ .
r0 r0 r0
(2.49)
The terms ai and bi are fitting parameters that describe the dependence of
the bond order on the bond geometry. The numerical values are adapted for
each bond type. The term rij is the distance between atoms i and j. This
equation yields the graphs shown in Fig. 2.21. Corrected bond orders BOij

are calculated from BOij via correction functions to account for the effect
of over- and under-coordination (the correction functions are a function of
the degree of deviation of the sum of the uncorrected bond orders around an
atomic center from its valency Vali ),
n
bonds

∆i = BOij − Vali . (2.50)
j=1
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,
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
Uangle = f (θijk , BOij , BOjk ). (2.53)
It is noted that this illustrates a distinction to the Tersoff potential. In Ter-

soff’s approach, the angular depedence is included in the multibody term for
pairs of atoms. In the ReaxFF approach, an explicit angular term is included,

similar to the approach used in CHARMM-type force fields.
A crucial aspect of the ReaxFF force field is that all parameters are derived
from fitting against quantum mechanical data (DFT level) [93]. This process
is referred to as force field training. The basic concept is to require a close-
as-possible agreement between the quantum mechanical prediction and the
force field result, for a wide range of properties. Typically these properties
include elastic properties and equation of state, surface energies, dissociation
energies and landscapes, or the interaction energies of organic compounds
with surfaces.
Due to the increased complexities of force field expressions and the charge
equilibration step that is carried out at each force calculation, reactive force
fields are between 50 and 100 times more expensive than nonreactive force
fields, yet several orders of magnitude faster than DFT-level calculations that
would be able to describe bond rupture as well.
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.
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]
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
2.6.5 Limitations of Classical Molecular Dynamics
Atomistic or molecular simulations is a fundamental approach, since it consid-

ers the basic building blocks of materials as its smallest entity, atoms. Stress
singularities at crack tips are handled naturally, thereby avoiding many chal-
lenges associated with continuum methods. Molecular dynamics is also a quite
appropriate tool for describing material deformation under extremely high
strain rates that are not accessible by other methods (FE, DDD, and other
approaches).
However, molecular dynamics simulations feature several limitations. It is
important to remember these limitations during the course of model develop-
ment and data analysis and interpretation. For example, molecular dynamics
simulation models typically only allow one to study materials with dimensions
of several hundred nanometers or below. The time scale limitation is another
serious limit of molecular dynamics, which has prevented many researchers
from studying phenomena of interest or to make rigorous links to labora-
tory experiments that are often carried out at much different time scales.
Please see [134] for a description of these issues. Further associated aspects
and limitations will be discussed in the forthcoming sections.
To treat electrons properly one needs to incorporate quantum mechanical
methods explicitly into the molecular models, which feature degrees of freedom
that describe the structure of electrons. This is particularly significant for
electronic properties, optical phenomena, and magnetic properties, which all

require such quantum mechanical treatments. The physical reason for this is
that these properties are derived from the electronic structure associated with
atoms, molecules, or system of atoms rather than being derived only from the
positions of atoms or the interatomic forces.
Some methods have been developed that utilize simple potential expres-
sions to describe the complex quantum mechanical effects. For instance, the
electron force field, or eFF [135], utilizes a single approximate potential
between nuclei and electrons, and correctly describes many phases relevant
for warm dense hydrogen. The eFF model thereby provides a simplified solu-
tion of the time-dependent Schroedinger equation. The potential formulation
requires only moderate computational effort, since the computational com-
plexity of the terms is similar to those used in traditional force fields that
treat atoms only as particles. Thus it may be possible to use this new force
field to simulate large excited systems that are currently beyond the reach of
quantum mechanical methods.
2.7 Numerical Implementation

Here we summarize a few important numerical and implementation aspects
of molecular dynamics simulation methods.
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.
• Obtain updating method, e.g., the Verlet scheme as described in (2.13),

choose time step and thermodynamical ensemble.
• Obtain an expression for forces, by defining an approximation of the
potential energy landscape U (r).
• Analyze data using statistical methods.
Fig. 2.26 Schematic of the numerical scheme in carrying out molecular dynamics
simulations
2.7.1 Periodic Boundary Conditions
Periodic boundary conditions (PBCs) is a widely used concept in molecu-

lar dynamics. Figure 2.26 depicts a schematic of implementation of periodic
boundary conditions. PBCs allows one to study bulk properties (that means,
there are no free surfaces) with small number of particles (here N = 3 for
three particles).
For a variety of thermodynamical properties it has been demonstrated that
this approach is very efficient. However, for mechanisms or phenomena that
involve inhomogeneous stress and deformation fields, the approach does not
give useful results. To represent such spatially varying fields it is vital to make
the system larger and model the entire collection of atoms that resemble the
physical space.
When periodic boundary conditions are used, all particles are “connected”
and do not sense the existence of a free surface. For numerical reasons, the
original cell is surrounded by 26 image cells (8 in 2D). Image particles in these
image cells move in exactly the same way as original particles. These image
copies are necessary to carry out force calculations in the system, as discussed
in Sect. 2.7.2.
2.7.2 Force Calculation
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 ,
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
Such second-order scaling of the force calculation time is a computational

disaster and would prevent us from solving large systems. Several strategies
are commonly used to avoid this type of scaling, as discussed in the next few
sections.
2.7.3 Neighbor Lists and Bins
Avoiding the second-order scaling of force calculation ∼N 2 is important to

develop numerically feasible strategies. A bookkeeping scheme that is often
used in molecular dynamics simulation is a neighbor list that keeps track of
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).
2.8 Property Calculation

In this section, we summarize important methods to calculate system prop-
erties from molecular dynamics simulations. We introduce definitions and
numerical procedures for the following material properties:
• Temperature T (thermodynamic property)
• Pressure P (thermodynamic property)
• Radial distribution function g(r) (structural information)
• Mean square displacement function (measure for mobility of atoms, relates
to diffusivity)
• Velocity autocorrelation function (transport properties)
• Virial stress tensor (link to continuum Cauchy stress)
2.8.1 Temperature Calculation
Using the kinetic energy of the system K, the temperature is given by
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.
2.8.2 Pressure Calculation
The pressure P is given by

11
N N
dφ
P = N kB T − rij . (2.59)
3 V i=1 j=1,j<i drij
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.
Ω(r + 2r ) considered volume
r+ r r+ 2
r
2
Fig. 2.29 Method to calculate the radial distribution function g(r)
2.8.3 Radial Distribution Function
The radial distribution function (RDF) g(r) is defined as

ρ(r)
g(r) = , (2.60)
ρ0
where ρ(r) is the local density at r, and ρ0 = N/V is the averaged atomic
volume density. A possible numerical strategy to calculate g(r) is depicted in
Fig. 2.29. For such a discrete estimate,
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 ).
Fig. 2.30 Radial distribution function g(r) for various atomistic configurations,
including a solid (crystal), a liquid and a gas
An example is shown in Fig. 2.30. The RDF enables to perform quantita-

tive analysis of atomic structure and can elucidate if a system is in solid, liquid,
or gas state. Moreover, the RDF provides detailed information about the
type of crystal structure or includes precise signatures of the specific atomic
arrangement of a liquid. It also provides a means to compare quantitatively
with experimental results (e.g., obtained using neutron-scattering techniques)
and can thereby assist in developing new potential formulations.
2.8.4 Mean Square Displacement Function
The mean square displacement function is defined as
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).
2.8.5 Velocity Autocorrelation Function
The velocity autocorrelation function (VAF) is defined as
1 1
N N
v(0)v(t) = vi (t0 )vi (t0 + t), (2.64)
N i=1 N j=1
where vi refers to the magnitude of the velocity vector of particle i.

The velocity autocorrelation function gives information about the atomic
motions of particles in the system. Since it is a correlation of the particle
velocity at one time with the velocity of the same particle at another time,
the information refers to how a particle moves in the system, such as during
diffusion. The diffusion coefficient can be calculated as

1 t =∞
D= v(0)v(t) dt (2.65)
3 t =0
Autocorrelation functions can be used to calculate other transport coeffi-

cients such as thermal conductivity or the shear viscosity. These expressions
derived based on the Green–Kubo relations can provide links between correla-
tion functions and material transport coefficients (also applicable to thermal
conductivity, shear viscosity, and other material transport properties).
In addition to the diffusivity D, the VAF also provides information about
structural properties of an atomic systems. For liquids or gases, under rel-
atively weak molecular interactions, the magnitude of the VAF reduces
gradually under the influence of weak forces. This is because the velocity
decorrelates rapidly with time. This is the same concept as stating the atom
loses its memory about what its initial velocity was. For example, in the case
of a gas, the VAF plot is a simple exponential decay, revealing the presence
of weak forces slowly destroying the velocity correlation.
In a solid, under relatively strong interactions, the atomic motion is an
oscillation, vibrating backwards and forwards, reversing their velocity at the
end of each oscillation. Therefore, the VAF corresponds to a function that
oscillates strongly from positive to negative values and back again, while the
magnitude of the oscillations decay in time. This leads to a function resembling
a damped harmonic motion.
Figure 2.31 plots the VAF for different cases.
2.8.6 Virial Stress Tensor
The challenge of defining an atomistic-based stress measure is how to relate

the continuum stress with atomistic stress.
While continuum properties are valid at any random material point, this
is not true for atomistic quantities due to the discrete nature of atomic
microstructure. Figure 2.32 depicts a schematic.
VAF
Ideal gas
1
Dense gas
Liquid
T
Solid
Fig. 2.31 Velocity autocorrelation function (VAF) for a gas, liquid, and solid
Continuous fields ui(x) Discrete fields ui(x)
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
For a pair potential, the virial stress is defined as

⎛ ⎞
1 1 ∂φ(r) ri
σij = ⎝− mα vα,i vα,j + rj | ⎠ , (2.66)
V α
2 ∂r r r=rαβ
α,β,α=β
where ri is the projection of the interatomic distance vector r along coordi-

nate i, and the indices α and β refer to the atom numbers considered in the
calculation. The first term mα vα,i vα,j is the kinetic contribution, and the sec-
ond term is the force contribution. We only consider the force part, excluding
the part containing the effect of the velocity of atoms (the kinetic part), a
measure referred to as the Zhou’s virial stress. It was recently shown that the
stress including the kinetic contribution is not equivalent to the mechanical
Cauchy stress. Then the stress tensor is defined as

11 ∂φ(r) ri
σij = rj | . (2.67)
V 2 ∂r r r=rαβ
α,β,α=β
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
2.9 Large-Scale Computing

Molecular dynamics of mechanics applications can be computationally chal-
lenging, due to
• Complexities of force field expressions (calculation of atomic forces)
• Large number of atoms, leading to huge computational requirements
• Limitations in accessible time scales; due to intrinsic limitations of molec-
ular dynamics, which leads to the requirement to simulate a large number
of molecular dynamics steps
To model macroscopic dimensions of materials with microstructural features,
one needs to consider system sizes with approximately 1023 atoms (corre-
sponding to 1 mole). This computational burden cannot be solved by any
computational equipment as of today. In addition to the shear computational
task, if one were able to carry out such simulations it would also result in
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.
2.9.1 Historical Development of Computing Power
The increase in computational power has largely contributed to the success of

molecular dynamics as a widely used tool in the analysis of small structures
or materials in extreme conditions. Further increase in computational power
will without doubt increase the opportunities of using molecular simulation
for many new applications.
Computational power has increased significantly over the past decades,
now reaching peak performances of several PFLOP. Table 2.3 summarizes the
various performance measures used to characterize computational power, also
indicating when the particular performance measure was reached.
Performance Floating point operations per second (FLOPS)

MFLOPS 106 , Million
GFLOPS 109 , Billion (reached in 1990s)
TFLOPS 1012 , Trillion (reached around 2000)
PFLOPS 1015 , Quadrillion (estimated to be reached in 2010)
Table 2.3 Explanation of the different measures of computing power, as well as
a description when it became available. In comparison a state-of-the art personal
computer (PC, laptop) provides a performance of approximately 30 GFLOPS
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
[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
The organization top500.org publishes lists of the fastest supercomput-

ers in the world twice a year, called “TOP500.” It can be accessed at
http://www.top500.org. Figure 2.35 displays the top 10 of the TOP500 list
as of Spring 2008.
2.9.2 Parallel Computing
On the basis of the concept of concurrent computing, modern parallel com-

puters are made out of hundreds or thousands of small computers (for
example personal computers) working simultaneously on different parts of
the same problem. Information between these small computers is shared by
communicating, which is achieved by message-passing procedures (MPI) [142].
Parallel molecular dynamics is relatively straight-forward to implement
with great efficiency in a message-passing environment. It is important to have
an effective algorithm for handling the summations of N interacting particles.
If summations had to be carried out for each particle over all particles, the
problem would scale with N 2 . This is a computational catastrophe for large
systems! However, if the interactions between particles are short ranged, the
problem can be reduced so that the execution time scales linearly with the
number of particles (that is, execution time scales with N ). The computational
space is divided up into cells such that in searching for neighbors interacting
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
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
(perfect parallel speedup would correspond to a constant execution time for

increasing number of processors).
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.
2.10 Visualization and Analysis Methods

Large-scale atomistic computer simulations can produce terabytes of data,
since the location, velocity, energy, and stress tensor of each atom need
to be stored (an molecular dynamics data set for a billion atoms occupies
around 100 GB of disk space). To interpret and understand the simulation
results, it is essential to have analysis tools available which are capable of
filtering out the useful information. Data processing and visualization is an
important step in the analysis to extract useful information from the simula-
tion. For example, the collective motion and interaction of defects determine
macroscopic properties such as the toughness of a material. Techniques to
extract such information from positions of atoms are critical and yet to be
developed in most cases to achieve such goals. Some of the techniques of post-
processing data and visualization techniques will be discussed in the following
paragraphs.
A long-standing dream of computer simulation scientists is three-dimen-
sional virtual reality to analyze the results. Imagine walking through the data,
as a viewer of atomic-scale size [139,144]! Scientists would then be able to iden-
tify interesting points and study these closer as the simulation proceeds. At
the “Collaboratory for Advanced Computing and Simulations (CACS)” at the
University of Southern California, an “Atomsviewer System” has been devel-
oped which visualizes large-scale data sets from huge computer simulations
[139, 144] (see also http://cacs.usc.edu/). The main feature of this system is
the ability to view materials processes simultaneously from different perspec-
tives. Figure 2.37 shows an example view of the tiled screen view of this system.
To visualize this huge amount of data, new techniques are being developed.
One approach is to process only the data the viewer at the current perspective
will actually see [139, 144]. This is achieved using the octree data structure.
The main idea is that although the data set may be very large, the viewer only
sees a very small portion of the data at any instant in time. The octree data
structure is a data management method to extract the data in the viewer’s
field of view [144]. This method is relatively coarse, and it has been shown
that 60% of all atoms are still invisible since they are hidden behind other
particles. Additional techniques such as a probabilistic approach referred to as
occlusion culling remove hidden atoms and thus further reduce the workload
for visualization. The whole visualization process is set up on a PC cluster
through parallel and distributed computing.
Another important aspect of scientific visualization is the generation of
movies. In recent years, it has been increasingly appreciated that a pic-
ture is worth a thousand words, and a movie is worth a thousand pictures.
Animations of simulations help to guide the eye to discover new scientific
phenomena. Historically, this has been particularly taken advantage of in
the biophysics community, where visualization of complex biostructures (e.g.,
proteins) is key in the understanding and interpreting simulation results. Inter-
estingly, some researchers have started to implement techniques which allow
Fig. 2.37 Rendering of a large molecular dynamics simulation on a tiled display at

USC, showing hypervelocity impact damage of a ceramic plate with impact velocity
15 kms−1 , where one quarter of the system is cut to show the internal pressure
distribution (the projectile is shown in white). This figure illustrates how novel
visualization schemes provide analysis methods for ultra large-scale simulations.
Reprinted from Computational 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
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.
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
Visualization, defined as a method of computing transforming the symbolic

into the geometric, enables to observe simulations and computations. Visual-
ization is also considered an approach for seeing the unseen, which enriches
the process of scientific discovery and fosters profound and unexpected
insights. Visualization of computational results is undoubtedly becoming an
increasingly important and challenging task in supercomputing.
2.10.1 Energy Method
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
{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. 2.39 Application of the energy method to visualize fracture surfaces in a

computational fracture experiment. Only high energy atoms are shown by filtering
them according to their potential energy. This enables an accurate determination
of the geometry of cracks, in particular of the crack tip. Typically, the analysis is
confined to a search region (shown as a dashed line) to avoid inclusion of effects of
free surfaces
Figure 2.39 depicts an application of the energy method to visualize frac-

ture surfaces in a computational fracture experiment. Only high energy atoms
are shown by filtering them according to their potential energy. Similar to that
as shown in Fig. 2.38 for ductile materials, this method enables one to carry
out an accurate determination of the geometry of defects such as cracks, in
particular of the position of the crack tip. Typically, the analysis is confined
to a search region (shown by the dashed line) to avoid inclusion of effects of
free surfaces.
2.10.2 Centrosymmetry Parameter

A more advanced analysis can be performed using the centrosymmetry tech-
nique proposed by Kelchner and coworkers [36]. This method makes use of
the fact that centrosymmetric crystals remain centrosymmetric after homo-
geneous deformation. Each atom has pairs of equal and opposite bonds with
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.
Defect ci /a20 Range ∆ci /a20
Perfect lattice 0.00 ci < 0.1

Partial dislocation 0.1423 0.01 ≤ ci < 2
Stacking fault 0.4966 0.2 ≤ ci < 1
Surface atom 1.6881 ci > 1
Table 2.4 Centrosymmetry parameter ci for various types of defects, normalized by
the square of the lattice constant a20 . In the visualization scheme, we choose intervals
of ci to separate different defects from each other
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
An example using this centrosymmetry technique is shown in Fig. 2.40.

This plot shows the same section as in Fig. 2.38b. Unlike in the analysis
with the energy method, stacking fault regions can be visualized with the
centrosymmetry technique.
2.10.3 Slip Vector
Although the centrosymmetry technique can distinguish well between dif-

ferent defects, it does not provide information about the Burgers vector of
dislocations. The slip vector approach was first introduced by Zimmerman
and coworkers in an application of molecular dynamics studies of nanoinden-
tation [147]. This parameter also contains information about the slip plane
and Burgers vector. The slip vector of an atom α is defined as
nα
1
i = −
sα xαβ
i − X αβ
i , (2.72)
ns
α=β
where ns is the number of slipped atoms, xαβ i is the vector difference of

atoms α and β at the current configuration, and Xiαβ is the vector differ-
ence of atoms α and β at the reference configuration at zero stress and no
mechanical deformation. The slip vector approach can be used for any mate-
rial microstructure, unlike the centrosymmetry parameter which can only be
used for centrosymmetric microstructures.
Figure 2.41 shows the result of a slip vector analysis of a single dislocation
in copper [39]. The slip vector s is drawn at each atom as a small arrow.
The Burgers vector b is drawn at the dislocation, where its actual length is
exaggerated to make it better visible. The dislocation line can be determined
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.
2.10.4 Measurement of Defect Speed
Accurate determination of the propagation speed of defects is crucial in the

analysis in particular of rapid materials failure phenomena.
For instance, the crack tip velocity is an important measure in the analysis
of dynamic fracture of brittle materials. The crack tip velocity is determined
from finding the crack tip position. The geometry of the crack tip is deter-
mined by finding the surface atom at the tip of the crack. This is achieved
by considering the surface atom with the highest y position, for instance, for
the case of crack propagation in the y-direction. This search is carried out
in the interior of a search region inside the slab. This quantity is averaged
over a small time interval to eliminate very high frequency fluctuations. To
obtain the steady state velocity of the crack, the measurements of the crack
speed must be taken within a region of constant stress intensity factor (see, for
instance [114] for an additional discussion on this issue). Figure 2.39 illustrates
this approach based on the energy filtering method.
2.10.5 Visualization Methods for Biological Structures
For the visualization of organic molecules (such as proteins), specific tools

have been developed. Many visualization tools exist that are capable of dis-
playing biological protein molecules and molecular clusters. A rather versatile,
powerful and widely used visualization tool is the Visual Molecular Dynamics
(VMD) program [148]. This software enables one to render complex molecular
geometries using particular coloring schemes.
It also enables one to highlight important structural features of proteins by
using a simple graphical representation, such as alpha-helices, or the protein’s
backbone. The simple graphical representation is often referred to as cartoon
model. These visualizations are often the key to understand complex dynam-
ical processes and mechanisms in analyzing the motion of protein structures
and protein domains, and they represent a filter to make useful information
visible and accessible for interpretation. Figure 2.42 shows the visual analysis
of a simple alpha-helix protein structure, with different visualization options.
Fig. 2.42 Analysis of a simple alpha-helix protein structure, with different

visualization options, plotted using VMD [148]
2.10.6 Other Methods
Other researchers have used a common neighbor analysis to analyze the

results of molecular dynamics simulations of crystalline structures [149–151].
In this method, the number of nearest neighbors is calculated, which allows
one to distinguish between different defects. Additional analysis to analyze
more complex structures such as grain boundaries is possible based on the
medium-range-order (MRO) analysis. This method is capable of determining
the local crystallinity class. The MRO analysis has been applied successfully
in the analysis of simulations of nanocrystalline materials, where an exact
characterization of the grain boundary structure is important (e.g., [152–154]).
2.11 Distinguishing Modeling and Simulation

The specific meaning of modeling versus simulation has been introduced in
Sect. 2.2. Here we briefly explain the application of these terms for the exam-
ple of a generic molecular dynamics model. Table 2.5 provides a list of tasks
associated with molecular dynamics and its classification into modeling and
simulation.
2.12 Application of Mechanical Boundary Conditions

The application of boundary conditions in atomistic and molecular systems
is essential, in particular for studies of the mechanical behavior.
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
Most straightforward are approaches to apply displacement boundary con-

ditions. In these applications, the dynamics of corresponding atoms (or groups
of atoms forming a physical domain whose displacement is prescribed) is
altered so that this group of atoms follows a prescribed motion rather than
following the dynamics according to the equations of motion.
To model weak layers or different interatomic interactions in different
regions of the simulation domain, one can assign a virtual type to particu-
lar groups of atoms. On the basis of the type definition of interacting pairs
of atoms, it is possible to calculate different interatomic interactions. For
instance, this enables one to model an atomically sharp crack tip by removing
any atomic interaction across a material plane (effectively describing this part
as a free surface). An example for this type of domain decomposition is shown
in Fig. 2.43. To apply mechanical load by controlling the displacement, a lin-

ear velocity gradient is established prior to simulation to avoid shock wave
generation from the boundaries (see also Fig. 2.43). To strain the system, a
few layers of atoms at the boundary of the crystal slab are not moved accord-
ing to the natural equations of motion (e.g., by the Verlet algorithm), but are
instead displaced in each step, according to the prescribed displacement, with
the velocity that matches the initial velocity field. This procedure has been
used in several atomistic studies of dynamic fracture [28, 146, 155, 156]. It is
possible to stop the increase of loading after a prescribed loading time, from
which on the boundaries are kept fixed. In an alternative method, the system
can be strained prior to the beginning of the simulation (according to the
particular loading direction), and the outermost material layers are then kept
fixed during the simulation (that is, atoms in this group do not move at all).
The application of stress or pressure boundary conditions can be more
challenging, but it can be achieved by utilizing appropriate ensemble schemes,
as for instance the Parinello–Rahman method. This approach enables one to
prescribe a particular stress tensor to the system. By changing the prescribed
value of the stress tensor one can simulate slow loading conditions.
To apply the forces to the molecule that induce deformation, steered
molecular dynamics (SMD) has evolved into a useful tool. Steered molecu-
lar dynamics is based on the concept of adding a harmonic moving restraint
to the center of mass of a group of atoms. This leads to the addition of the
following potential to the Hamiltonian of the system:
1
U (r1 , r2 , ..., t) = k (vt − (r(t) − r0 ) · n) , (2.73)
2
where r(t) is the position of restrained atoms at time t, r0 denotes origi-
nal coordinates and v and n denote pulling velocity and pulling direction,
respectively. The net force applied on the pulled atoms is
F (r1 , r2 , ..., t) = k (vt − (r(t) − r0 ) · n) . (2.74)
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).
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
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
Molecular dynamics is a useful method to

• Perform virtual experiments, that is, computational experiments
• Implement a computational microscope to visualize and analyze micro-
scopic processes
• Gain fundamental understanding about behavior of materials
Further, molecular dynamics:
• Has an intrinsic length scale, given by the distance of atoms in the material
(typically on the order of the length of a chemical bond), that is, between
1 and 5 Å
• Handles stress singularities (e.g., at crack tips) intrinsically
• Is ideal for deformation under high strain rate and extreme conditions,
which are not accessible by other methods (such as the finite element
method, discrete (mesoscale) dislocation dynamics, and other simulation
approaches)
3
Basic Continuum Mechanics
The purpose of this chapter is to provide an introduction into basic continuum

mechanical concepts, including stress and strain tensor and equilibrium con-
ditions. Using the concepts introduced in this chapter, we will derive simple
solutions to elastic beam bending problems to demonstrate how problems can
be addressed in the realm of continuum mechanics. The objective of solving
such elasticity problems is to provide a link between applied forces, boundary
conditions, and the stresses, strains, and displacements inside the material.
Such information is critical for instance to find out if a designed structure can
operate safely, away from its failure condition. There are several additional
applications and significant components of elasticity problems in many other
fields, as will be discussed later in this chapter. Topics covered in this chapter
include definition of elasticity, elastic response, energetic vs. entropic elastic-
ity, Young’s modulus, stress and strain tensor, as well as the mechanics of a
beam. The continuum mechanics concepts introduced in this chapter will be
used later on, when a direct comparison between atomistic and continuum
methods is carried out.
3.1 Newton’s Laws of Mechanics

Sir Isaac Newton (1642–1727) proposed three fundamental laws of mechanics,
laying the foundation for the continuum mechanical framework. All continuum
mechanical governing partial differential equations can be derived from these
three concepts.
• First law: An object in a state of rest or uniform motion tends to remain
in that state of motion unless an external force is applied to it. That is, as
long as the sum of forces acting on it is zero

F= Fi = 0, (3.1)
the direction and magnitude of velocity of it does not change.

• Second law: The change of motion is proportional to the applied force

to an object. That is,
d(mv)
= F. (3.2)
dt
For constant mass m, this law simplifies to
dv
m = ma = F, (3.3)
dt
where a is the acceleration (for the application in the molecular dynamics
formulation, please see Sect. 2.4). It is noted that the second law as stated
above is strictly valid only for a single point mass that cannot undergo rota-
tion. For a collection of N point masses, the equation has to be extended
to include
d
N N N
(xi × mi vi ) = (xi × Fi ) = Mi . (3.4)
dt i=1 i=1 i=1
• 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
3.2 Definition of Displacement, Stress, and Strain

We begin with a brief introduction into basic concepts. We consider a beam
under axial, tensile loading, as shown in Fig. 3.1. The globally applied bound-
ary conditions (shear force P and axial force N ) lead to a local section stress
distribution.
This problem features several fundamental aspects of mechanics of mate-
rials. Considering the force–extension curve shown in Fig. 3.1, several regimes
of deformation can be recognized. In the first regime, the F –∆u curve reveals
a linear relationship between applied force F and extension ∆u,
∆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)
where ks is a proportionality factor that can be regarded similar as a spring

constant. It describes the stiffness of the mechanical element.
The formulation put forward in (3.7) is dependent on the size of the
mechanical element considered. For instance, making the beam longer results
in a smaller factor ks . To be able to characterize material properties indepen-
dent of size of the structure, (3.7) can be rewritten in a renormalized way,
by dividing by the volume V of the element considered. The volume can be
expressed as V = AL, where A is the cross-sectional area and L is the length
of the beam. Then,
F 1 ks ∆u
= . (3.8)
AL A L
The quantity
F
σ= (3.9)
A
is defined as stress, and
∆u
ε= (3.10)
L
as strain. Rewriting (3.8) yields
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.
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.
The goal of the continuum mechanical framework is to derive a relationship

between applied forces and displacements, strains, and stresses in the struc-
ture. In a sense, such solutions represent a simple class of multiscale problems,
as they provide a link between local forces and the global structural response.
Figure 3.3 illustrates this concept. The goal of solving this type of problem is
to connect the global scale (scale on the order of L where boundary condi-
tions are applied that is, the applied load P , N , constraints of displacements
at x = 0) with the local scale (that is, the section of the beam). In partic-
ular, here the stress state inside the beam σxx and how it varies over the
cross-sectional area is of interest.
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)
3.2.1 Stress Tensor
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
The concepts introduced in the previous section for a one-dimensional case

can easily be generalized for a three-dimensional situation. Figure 3.4 depicts
a sectioned body with internal forces (the plot on the right shows a split of
the internal force vector into its components ∆Px , ∆Py , and ∆Pz ).
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).
3.2.2 Equilibrium Conditions
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.
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
First, the moment equilibrium has to be satisfied. The condition that

no net torque acts on an infinitesimal element (conservation of the angu-
lar momentum) yields the condition that the stress tensor is symmetric (see
schematic in Fig. 3.7),
σij = σji . (3.14)
This condition reduces the nine components of the second-order tensor to six.
Further, based on Newton’s first law (3.1), a differential material volume

dV = dxdydz is in equilibrium if the net force applied to this element is zero,
fi = 0.
The differential equilibrium equations can be derived by applying New-
ton’s laws. Carrying out a balance of forces in the x-direction (note that the
equation is expressed per unit depth of the infinitesimal cube) yields

∂σxx ∂σyx
σxx + dx − σxx dy + σyx + dx − σyx dx + fx (dx × dy) = 0,
∂x ∂y
(3.15)
resulting in
∂σxx ∂σyx
+ + fx = 0. (3.16)
∂x ∂y
This balance is carried out also for the y-direction and a similar result is
obtained, leading to
∂σyx ∂σyy
+ + fy = 0. (3.17)
∂x ∂y
For the three-dimensional case, this approach results in three equations:
∂σxx ∂σxy ∂σxz
+ + + fx = 0, (3.18)
∂x ∂y ∂z
∂σyx ∂σyy ∂σyz
+ + + fy = 0, (3.19)
∂x ∂y ∂z
∂σxz ∂σzy ∂σzz
+ + + fz = 0. (3.20)
∂x ∂y ∂z
These equations can be written more generally for all directions and in three
dimensions as
∂σij
+ fi = 0. (3.21)
∂xj
This equation is sometimes also written as
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
σij,j + fi = ρ∂tt ui . (3.24)
Solving these partial differential equations is generally quite difficult, and

it can only be achieved in specific situations. Often, numerical approaches
such as the finite element method are used to approximate the solutions.
Fig. 3.8 Schematic to illustrate the definition of the deformation tensor
3.2.3 Strain Tensor
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
In general, the displacements ui can be expressed in terms of the defor-

mation tensor eij . The deformation tensor for small deformation is defined as
∂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
3.3 Energy Approach to Elasticity

How are stresses and strains related? A thermodynamical view of the physical
process of elastic deformation provides insight into this problem.
The key physical concept in describing elastic deformation is the fact that
it refers to reversible deformation. Thus, here we focus on the first regime
of the example shown in Fig. 3.1 (the elastic regime in which deformation is
indeed reversible, that is, upon release of the applied force, the beam returns to
its undeformed configuration). For the discussion in this section, we recall that
the relationship between stretching force and extension is given by F = ks ∆u.
The first law of thermodynamics states that
∂U
= δW + δQ, (3.35)
∂t
where δW = ẋFe (Fe is the applied force, and δQ is the heat provided to the
system.
The second law of thermodynamics states that
∂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
∂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
The elasticity tensor is then given by
∂2Ψ
cijkl = . (3.48)
∂εij dεkl
This reveals that as soon as an expression for F = U − T S is known

for a specific material, a rigorous link between stress and strain is provided
within the framework of linear elasticity. This is a core component of the
concepts described here, as it provides a link between traditional and new
fields of mechanics. The quantity Ψ can be calculated from first principles
atomistic calculations, for example. We will utilize this concept later when we
make a link between atomistic and continuum mechanical elasticity, or the
stress–strain behavior.
Hooke’s law is an approach to link the measure of deformation with the

corresponding stress necessary to induce this deformation. The stress and
strain tensor – both second-order field tensors – are related by a fourth-order
tensor, the elasticity tensor cijkl via the generalized Hooke’s law:
σij = cijkl εkl . (3.49)
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
cijkl = clkij , (3.52)
which is a consequence of the independence of the order of differentiation in

(3.48), resulting in 21 independent elastic coefficients.
3.4 Isotropic Elasticity

In the most general case, 21 independent elastic coefficients cijkl exist,
as discussed in the previous section. In isotropic elasticity, the number of
independent elastic coefficients is significantly reduced, and there are only
two independent coefficients that describe the relation between stress and
strain.
In isotropic materials, the force–load response, or equivalently, the stress–
strain relationship does not depend on direction of applied load. Examples for
isotropic materials are polycrystalline metals, in which texture formation leads
to anisotropy, or amorphous materials that have no preferred orientation even
at atomistic scales. In contrast, in anisotropic elasticity, mechanical properties
depend on direction of the applied load. Examples for anisotropic materials
include single crystals of materials, since different symmetry directions in
crystals lead to different elasticity.
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)
3.5 Nonlinear Elasticity or Hyperelasticity

The type of elasticity discussed in the previous sections assumed that the
elastic stiffness does not change under deformation. Here we briefly intro-
duce a concept referred to as nonlinear elasticity or hyperelasticity, that
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.
Fig. 3.10 Illustration of the concept of nonlinear elasticity or hyperelasticity. Sub-

plot (a) shows the stress–strain relationship, and subplot (b) depicts the tangent
modulus as a function of strain. Linear elasticity is based on the assumption that
the modulus is independent of strain. However, most real materials do not show this
behavior. Instead, they show a stiffening effect (e.g., rubber, polymers, biopolymers)
or a softening effect (e.g., metals, ceramics)
3.6 Elasticity of a Beam
Fig. 3.11 Geometry of the beam
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 .
3.6.1 Reduction Formulas
A key ingredient in exploiting the properties of beam problems is the intro-

duction of reduction formulas, that is, the calculation of equivalent force and
moment values per section of the beam. These section quantities are then only
a function of the x-coordinate (beam axis), as the variation of the stresses with
y and z within a section is lumped into a single value. Thereby the dimen-
sionality of the problem can be reduced from 3D to 1D. By calculating the
force vector at a cross-section of the beam, at a given location x,

Nx = σxx (y, z)dS, (3.69)
S

Qy = σyx (y, z)dS, (3.70)
S

Qz = σzx (y, z)dS. (3.71)
S
The total force vector is
FS = Nx ex + Qy ey + Qz ez . (3.72)
In addition to the force expressions, we can also calculate section moments:

Mx = (yσzx (y, z) − zσyx (y, z)) dS (3.73)
S

My = zσxx (y, z)dS (3.74)
S

Mz = − yσxx (y, z)dS (3.75)
S
3.6.2 Equilibrium Equations
At each section of the beam, the differential equilibrium conditions in (3.21)

have to be satisfied. These equilibrium equations are derived in analogy to
the considerations done at the differential element, which lead to the more
general form of the governing equations.
The equilibrium conditions can be written as (note that the force density
qi has units of force per unit length) for the forces,
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
These equations can be solved by integration while considering proper bound-

ary condition at x = 0 and x = L to solve for the resulting unknown
integration constants. For example, moments or forces at the end of the beam
may be specified.
3.6.3 Example: Solution of a Simple Beam Problem
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
The equilibrium equations for this particular problem are

dNx
=0 (3.82)
dx
dQy
=0 (3.83)
dx
dQz
− ρAg = 0. (3.84)
dx
Integration yields Nx = C1 , Qy = C2 , and Qz = C3 + ρAgx. The moment
equilibrium conditions yield
dMx
=0 (3.85)
dx
dMy
− (C3 + ρAgx) = 0 (3.86)
dx
dMz
+ C2 = 0 (3.87)
dx
Integration yields Mx = C4 , My = C5 + C3 x + 12 ρAgx2 , and Mz = C6 − C2 x.
The application of boundary condition now allows us to find the solution to
the problem: The forces F(x = L) = 0 and moments M(x = L) = 0 since
there is a free end at x = L. Therefore, C1 = C2 = C4 = C6 = 0 and
C3 = −ρgLA. Since M (x = L) = 0, C5 = 12 ρgL2 A. Overall the only nonzero
force and moment distributions are given by,
Qz (x) = ρgA(x − L) (3.88)

1
My (x) = ρgA(x − L)2 . (3.89)
2
Figure 3.12 shows the solution field of this problem.
3.6.4 Calculation of Internal Stress Field
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:
u = u0 (x) + us (y, z) (3.90)

ε = ε0 (x) + εs (y, z) (3.91)
Fig. 3.13 Demonstration of the concept of the Navier–Bernouilli assumption
According to the Navier–Bernouilli assumption, an initially plane beam

section that is perpendicular to the beam’s reference axis remains plain after
deformation, and is also perpendicular in the deformed state (see Fig. 3.13).
Then, the displacements in the section can be written as
us = u0 (x) + ω(x) × xs (3.92)
where xs = yey + zez , and ω(x) as the infinitesimal rotations:
∂uy,0
ωz,0 = , (3.93)
∂x
∂uz,0
ωy,0 =− . (3.94)
∂x
This leads to the following expressions for the strains
εxx = εxx,0 + θy,0 z − θz,0 y. (3.95)
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
My = zσxx dS, (3.99)

S
Mz = −yσxx dS. (3.100)
S
Now we assume an isotropic material behavior (with constant E along the

length x), which results in
σxx = Eεxx , (3.101)
where εzz = εyy = −νεxx .
Considering (3.95) and (3.101), we find that
σxx = E (εxx,0 + θy,0 z − θz,0 y) . (3.102)
Then, the normal force and section moments are

Nx = E (εxx,0 + θy,0 z − θz,0 y) dS, (3.103)
S

My = E zεxx,0 + θy,0 z 2 − θz,0 yz dS, (3.104)
S

Mz = E −yεxx,0 − θy,0 yz − θz,0 y 2 dS. (3.105)
S
This set of equations can be written in simpler form as a matrix product

⎡ ⎤ ⎛ # dS # #
zdS − S ydS
⎞⎡
εxx,0
⎤
Nx # S
# 2
S
#
⎣ My ⎦ = ⎜ ⎝ S zdS
⎟⎢
z dS − S zydS ⎠ ⎣ θy,0 ⎦ .
⎥
(3.106)
Mz # #S
# 2
− ydS − zydS − y dS
S S S
θ z,0
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)
For a rectangular cross section (defined by b × h), the second-order area

moments are given by
b3 h
Iyy = , (3.108)
12
bh3
Izz = , (3.109)
12
and
Izy = 0. (3.110)
In fact, it can be shown that for any cross-section with at least one symmetry
around one of the section axis Izy = 0.
In this case, the moments can be expressed as
My = EIzz θy,0 , (3.111)

Mz = EIyy θz,0 . (3.112)
These expressions can now be used in (3.102) to find an expression for

the stress σxx (x; y, z), in terms of the section quantities Nx (x), My (x), and
Mz (x):
Nx (x) z y
σxx (x; y, z) = + My (x) − Mz (x). (3.113)
A Izz Iyy
Similar expressions can be developed also for cases when there is no geometric
symmetry in the cross–section.
Finally, we briefly outline how the stress distribution in the section can be
obtained based on these equations. Recall the solution of the problem solved
above,
1
My (x) = ρgA(L − x)2 . (3.114)
2
The largest section moment appears at x = 0, being My = 12 ρgA. Therefore,
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.
3.6.5 Differential Beam Equations
The beam equilibrium equations can be expressed as
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,
∂ 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
Using the differential beam equations, the same solution as constructed

earlier can be obtained, with the following boundary conditions: The dis-
placement uz (x = 0) = 0 and Qz (x = L) = 0 as well as My (x = L) = 0,
noting that qz = −ρAg.
Integration of these equations yields for the geometry shown in Fig. 3.12a
Qz (x) = ρgA(x − L) (3.126)
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:
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,
uz (L) ∼ ρgA (3.131)
and it is inversely proportional to the parameter EIzz (EIzz is also referred

to as bending stiffness),
1
uz (L) ∼ . (3.132)
EIzz
Note that typically the indexes “0” are dropped, as the solution always refers
to the result of the beam reference axis. The rotation and displacement
functions are shown in Fig. 3.12c and d.
Figure 3.14 depicts the solution of another case, for a beam subject to a
point load P applied at the tip of the beam. For this case, qz = 0, the shear
force
Qz (x) = −P, (3.133)
the moment
My (x) = P (L − x), (3.134)
the rotation

P x2
ωy (x) = − − xL , (3.135)
EIzz 2
and the displacement

P x3 x2 L
uz (x) = − . (3.136)
EIzz 6 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
3.7 The Need for Atomistic Elasticity: What’s Next

Elastic deformation is an important aspect of engineering design. Structures
such as buildings, microelectronic components, displays, airplanes, and many
others are typically designed for elastic deformation, that is, to specify the
amount of deformation under typical loads load. Elastic properties are also
important for plastic flow, since the critical conditions for the onset of perma-
nent deformation are related to the material behavior at small bond stretches.
Further, as we shall see in Chap. 6, the driving force for brittle fracture is also
related to the elastic energy stored in a solid.
For both geometries of beam problems reviewed here (Figs. 3.12 and 3.14),
the solution of the displacement of the beam reference axis depends on two
classes of material parameters. First, geometric parameters that are captured

in Izz and A. Second, a material parameter E, Young’s modulus. Contin-
uum mechanics is very powerful in providing us with a relation between
displacements, moments, stress, and other quantities, and the applied load,
as a function of material parameters.
However, continuum mechanics does not provide us with a description of
how the material parameters are determined – E remains an unknown param-
eter. One source of determining this quantity is to measure it by experimental
studies. Another possibility is to calculate the value of E from the atomistic
structure of the material. This will be done in the forthcoming chapter, when
we will explore methods that enable us to use atomistic simulation to predict
the values of Young’s modulus.
4
Atomistic Elasticity: Linking Atoms
and Continuum
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.
4.1 Thermodynamics as Bridge Between Atomistic

and Continuum Viewpoints
The link between atomistic models and the concept of elasticity used in con-
tinuum mechanical theories can be most directly established by considering
concepts of thermodynamics: By calculating how the free energy of a sys-
tem changes due to deformation, one can calculate the stress state. How is
this possible? In the spirit of handshaking, thermodynamics provides this link
since both a statistical mechanics approach and a continuum theory can be
related to thermodynamical concepts. In this sense, thermodynamics is the
glue between atomistic methods and continuum theories, linking microscopic
states with a macroscale system. In the following sections, we will describe how
this concept can be used to link atomic interactions with concepts of stress
and strain. Such handshaking concepts also lay the foundation for some mul-
tiscale simulation techniques, as they allow to transcend throughout distinct
scales. However, it is emphasized that during the transition from microscale to
macroscale, information is lost: Whereas the atomistic scale provides a deter-
ministic description of the atomistic state, an averaged, continuum description
does not. Some of the most fundamental concepts of thermodynamics tell
us that many microscopic states refer to the same macroscopic state. This
represents an important issue in the case when a transition is to be made
from microscale to macroscale and vice versa. Whereas statistical concepts

can be used in the case of equilibrium phenomena (e.g., by assigning random
atomic velocities due to a specific temperature, following an appropriate Boltz-
mann distribution); this is a tremendously challenging task for nonequilibrium
systems.
The need to advance methods that combine atomistic and continuum anal-
ysis is becoming increasingly compelling with rapid advance in computational
resources. Many supercomputer facilities now provide peak performances of
several TFLOPs per second, and system sizes with billions of particles can
readily be simulated. In order to compare atomistic simulations with con-
tinuum analysis level, it is necessary to use methods that allow transition
between the two levels of descriptions [157–159]. Of particular interest is the
relationship between interatomic potentials and associated elastic properties.
Here we limit the scope of the discussion primarily to the calculation of
properties at thermodynamical equilibrium.
4.2 The Atomic and Molecular Origin of Elasticity:

Entropic vs. Energetic Sources
Elasticity stems from the change of the free energy due to the collective inter-
actions of atom and their bonds under deformation, and thus it is intimately
linked to the chemistry that defines the atomic interaction energies. The elas-
tic properties of materials can be expressed as the partial derivative of the
free energy density with respect to the strain tensor that characterizes the
resistance to deformation, as discussed in Sect. 3.3.
We briefly review the main steps delineated above, based on the free
energy F that is composed out of energetic U and entropic contributions
S (temperature is denoted by T ). The free energy is
F = U − T S, (4.1)
we can define the free energy density (V denotes the volume),

F
Ψ= . (4.2)
V
The (scalar) stress is then is given by
∂Ψ
σ= (4.3)
∂ε
and the elastic modulus can be expressed as
∂ 2Ψ
E= , (4.4)
∂ε2
since the stress and strain are related by Hooke’s law, σ = Eε.
4 Atomistic Elasticity: Linking Atoms and Continuum 123
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.
4.3 The Virial Stress and Strain

The virial stress and strain provide important concepts to couple deformation
behavior of atomistic systems with continuum theories and concepts.
The virial stress is defined in Sect. 2.8.6. As most atomistic quantities,

the stress tensor coefficients must be averaged in space and time in order to
be compared with continuum concepts. Thus, the virial stress needs to be
averaged over space and time to converge to the Cauchy stress tensor.
The strain field is a measure of geometric deformation of the atomic
lattice [112]. The local atomic strain is calculated by comparing the local
deviation of the lattice from a reference configuration. Usually, the reference
configuration is taken to be the undeformed lattice. In the atomistic simula-
tions, the information about the position of every atom is readily available,
either in the current or in the reference configuration and thus calculation of
the virial strain is relatively straightforward. We limit the consideration to
the two-dimensional case.
We define the following tensor for atom l

1 ∆xkl
N kl
l i ∆xj
qij = , (4.9)
N r02
k=1
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.
4.4 Elasticity Due to Energetic Contributions

A possible method of linking atomistic and continuum concepts is to use the
Cauchy–Born rule [160, 161], which provides a relation between the energy
created from a macroscopic strain field and the atomistic potential energy
found in a stretched crystal lattice [157–159, 162].
4.4.1 Cauchy–Born Rule

The central assumption of the Cauchy–Born rule is that the energy of an
atomic system can be expressed as a function of a macroscopically applied
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
This approach is particularly simple for numerical evaluation in periodic

systems, in which the atomistic system corresponds to the unit cell of a crystal
lattice. In this case, the expression given in (4.13) is typically expressed as
energy contributions of individual bonds in all possible directions as a function
of the strain tensor.
Further, in crystal lattices with pair wise interactions, there is typically a
finite number of bonds, so that the integral is replaced by a summation over all
bonds. The summation typically involves the pair potential function φ, evalu-
ated at the particular bond length r(εij ), which is then directly summed up to
yield the total potential energy. The sum over all these energy contributions
resulting from all α bonds in the system yields the SED:
1
Ψ (εij ) = φ(rα (εij )). (4.14)
Ω α
In summary, the typical steps in applying the Cauchy–Born rule involve:

• Define lattice and corresponding unit cell, along with cell volume.
• Define interatomic interactions (that is, choose a potential function).
• Express the energy of the unit cell (denoted by Ψ ) as a function of all
atomic coordinates, as a function of the applied strain tensor. In the simple
case of pair-wise interactions, the calculation typically reduces to consid-
ering the distances between all pairs of atoms, which can then be used to
determine the energy contribution of all pairs of atoms. The final result of
this step is an expression Ψ (εij ).
• Use the expression of Ψ (εij ) to calculate the stress tensor and elasticity
coefficients, according to (4.5) and (4.6).
4.4.2 Elasticity of a One-Dimensional String of Atoms

As a first example, we consider a one-dimensional string of atoms, as shown
in Fig. 4.1.
Fig. 4.1 Example to illustrate Cauchy–Born rule in a one-dimensional geometry
In its undeformed configuration, the distance between pairs of atoms is

r0 . This one-dimensional crystal is defined by a unit cell that contains one
bond and one atom. The volume of the unit cell is Ω = r0 × 1 × 1 (assuming
unit lengths in the out-of-plane directions). Assuming existence of a function
describing the relation of bond energy and bond length, the SED is given by
1
Ψ (ε) = φ(r), (4.15)
r0 Ω
where D = 1 × 1 denotes the out-of-plane cross-sectional area. The function
φ(r) could, for example be a Lennard-Jones 12 : 6 potential, defined in (2.30).
The bond length r in the deformed state, as a function of the strain ε, is given
by
r = (1 + ε) r0 . (4.16)
Therefore,
1 1
Ψ (ε) = φ(r) = φ ((1 + ε) r0 ) . (4.17)
r0 Ω r0 Ω
Since we are interested in obtaining elastic coefficients within the concept
of linear elasticity, the second derivatives of Ψ (ε) with respect to ε should be
constant. Thus we consider a second-order Taylor expansion of the interatomic

potential around the equilibrium bond length r0 :
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.
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
4.4.3 Elasticity and Surface Energy of a Two-Dimensional

Triangular 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]
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
√
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.
Nonlinear Tethered LJ Potential
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 ,
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
where the parameter 0 can be chosen to change the small-strain elastic

properties (here we assume σ = 0 = 1) [156, 165].
Figure 4.4 plots the elastic properties associated with this potential for a
two-dimensional triangular lattice. The upper two subplots show the stress vs.
strain behavior under uniaxial stress loading. The left refers to uniaxial stress
loading in the x-direction, and the right plot shows the stresses for uniaxial y
loading. The Poisson ratio is calculated to be around ν ≈ 0.33. The nonlinear
nature of this potential can clearly be identified in these plots. The tangent
moduli stiffen strongly with strain, and agree with the small-strain elastic
properties of the LJ potential. Therefore, the wave velocities assuming small
perturbation from the equilibrium position are the same in both the LJ and
the tethered LJ potential.
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
We introduce a harmonic potential with the objective to mimic linear elastic

material behavior, as assumed in most theories of fracture. The linear spring
potential given by (2.34) corresponds to the “ball-spring” model of solids and
yields a plane-stress elastic sheet for a triangular lattice.
Using expressions similar to (4.35), Young’s modulus E and shear modulus
µ can be shown to be
√
2 3
E = √ k, µ = k. (4.37)
3 4
√
When we assume k = 72 3 2 ≈ 57.14 so as to match the small-strain elastic
properties of the LJ potential, E ≈ 66 and µ ≈ 24.8. Equation (4.35) (expres-
sion for infinitesimal strains) can be used to show that Poisson’s ratio ν = 1/3.
Using the above given values for elastic properties, the wave speeds can be
obtained straightforwardly. The longitudinal wave speed can be calculated
from the elastic properties to be

3µ
cl = (4.38)
ρ
√
with ρ = 2/21/3/ 3 ≈ 0.9165 (assuming mass m = 1). The shear wave speed
is given by the square root of the ratio of the shear modulus µ to the density
ρ thus
µ
cs = . (4.39)
ρ
Finally, the speed of elastic surface waves, the Rayleigh speed, is given by
cR ≈ βcs . (4.40)
The value of β can be found by solving the following equation Γ = cs /cl :

β 6 − 8β 4 + 8 3 − 2Γ 2 β 2 − 16 1 − Γ 2 = 0. (4.41)
This solution is found to be β ≈ 0.923.
Spring Young’s Shear Poisson’s cl cs cR

constant, k modulus, E modulus, µ ratio, ν
√
3
36 √ 2 ≈ 28.57 33 12.4 0.33 6.36 3.67 3.39
72 3 2 ≈ 57.14 66 24.8 0.33 9 5.2 4.8
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
The wave speeds are given by
cl = 9, cs = 5.2, and cR ≈ 4.8. (4.42)
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.
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.
Harmonic Bond Snapping Potential
In contrast to the elastic properties of a solid that never breaks as reported

in Sect. 4.4.3, here we discuss the elastic properties of a triangular lattice
with harmonic interactions where the bonds break upon a critical separation
r > rbreak . The interatomic potential is defined as

a0 + 12 k(rij − r0 )2 if rij < rbreak ,
φij (rij ) = (4.43)
a0 + 12 k(rbreak − r0 )2 if rij ≥ rbreak .
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
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.
Loading direction Poisson relaxation εbreak

x Yes 0.08
y Yes 0.1
x No 0.26
y No 0.09
Table 4.2 Failure strain of the two-dimensional solid associated with the harmonic
potential with snapping bonds under different modes of uniaxial loading for rbreak =
1.17
We start with a discussion of stretching in the x-direction (Fig. 4.7a), and

consider stretching with and without Poisson relaxation. For uniaxial tension
without Poisson contraction, the length of both bonds r1 and r2 increases. In
contrast, for uniaxial tension with Poisson contraction, the length of bonds r1
increases, whereas the condition that σyy = 0 requires that |r2 | = r0 . There-
fore, with the assumption of Poisson relaxation, for arbitrarily large strains in
the x-direction, only two bonds r1 break while the other four bonds r2 never
break. However, these bonds do not contribute to the stress. In contrast, if
no Poisson relaxation is assumed, these bonds do indeed contribute to the
stress and fail at much higher strain than the first two bonds, this increasing
the critical strain for bond breaking. Such behavior is indeed observed in the
numerical calculation of the elastic properties. Figure 4.8 (left) shows uniax-
ial tension with Poisson relaxation. Under stretch in the x-direction, the solid
fails at about 8% strain. As could be verified in Fig. 4.9 (left), the solid fails at
about 26% strain when no Poisson relaxation is assumed. A reduced modulus
E ≈ 10 is found between the failure of the first two bonds and the failure of
the remaining four bonds. Note that in the figures, the number of bonds is
also indicated by the dotted line. The reason for the huge difference in the
two cases is, as outlined above by theoretical considerations, the contribution
of the four bonds r2 which only break at very high strains. Note that for any
stress state not equal to uniaxial tension, there will be a contribution to the
stress from the two remaining bonds.
We continue with a discussion of stretching in the y-direction (Fig. 4.7b),
and also consider stretching with and without Poisson relaxation. For uniaxial
tension without Poisson contraction, the length of both bonds r2 increases
while the bond length of bonds r2 remains r0 and does not contribute to the
stress. Under unixial tension with Poisson relaxation, the length of all bonds is
adjusted to satisfy the condition that σxx = 0. In both cases, upon a critical
strain the number of bonds drops to two (since the four bonds r2 break)
and the remaining bonds r1 do not contribute to the stress in both cases
with and without Poisson relaxation. This is a significant difference to the
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
behavior in the other loading direction. As a consequence, the critical strain

for failure is comparable under both loading conditions in the y-direction.
Such behavior is indeed observed in the numerical calculation of the elastic
properties. Figure 4.8 (right) shows uniaxial tension with Poisson relaxation.
Under stretch in the y-direction, the solid fails at about 10% strain. As could
be verified in Fig. 4.9 (right), the solid fails also at about 10% strain when no
Poisson relaxation is assumed.
In summary, there is a strong dependence of the failure strain on the
loading condition. Table 4.2 summarizes the failure strains for different modes
of loading in the x- and y-direction.
Under large stretching, harmonic lattices behave differently than solids
defined by the LJ potential since bonds contribute little to the stress as they
weaken strongly with strain. Therefore, the direction with lowest breaking
strain is associated with loading in the x-direction, which is the direction with
highest fracture surface energy. The consequence of this is that crack propa-
gation is stable along the direction of high fracture energy [146, 166]. Solids
associated with harmonic bond snapping potentials show a different behav-
ior: In the harmonic bond snapping systems, the failure strain is larger in
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
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
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
Fracture Surface Energy

The fracture surface energy γs is an important quantity for the nucleation and
propagation of cracks. It is defined as the energy required to generate a unit
distance of a pair of new surfaces (cracks can be regarded as sinks for energy,
where elastic energy is converted into surface fracture energy). The Griffith
criterion predicts that the crack tip begins to propagate when the crack tip
energy release rate G reaches the fracture surface energy 2γs , G = 2γs [62].
The fracture surface energy can be expressed as
∆φ
γs = − , (4.47)
d
where d is the crack advance and ∆φ is the energy necessary to break atomic
bonds as the crack advances a distance d. The bond breaking process is
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
, -
2a1 + k1 (r1 − rbreak )2 E0 (ron − r0 )2 − E1 (ron − r1 )2 − (r1 − rbreak − r0 )2
γsbi,l = = √ .
2r0 4/ 3r0
(4.51)
4.4.4 Elasticity and Surface Energy of a Three-Dimensional FCC

Lattice
Thus far, we have focused attention to one- and two-dimensional models. We

dedicate this section to the discussion of mechanical and physical properties
of three-dimensional solids associated with a face-centered cubic lattice. The
aim is to determine the elastic properties and the fracture surface energy, to
be used for computational experiments of mode III cracks (see section 6.10
and 6.11). We will discuss results for harmonic potentials, as well as LJ and
EAM solids.
Harmonic Potential
Here we focus on harmonic interactions between atoms as defined in (2.34).

As in the two-dimensional models, atoms only interact with their nearest
neighbors. We begin the discussion with the elastic properties and wave speeds
for cubical crystal orientation.
We assume that nearest√neighbor distance
√ is r0 = 21/6 ≈ 1.12246, so that
the lattice constant a0 = 2r0 = 2 1/6
2 ≈ 1.5874. For mass m = 1, the
density is given by ρ ≈ 1, since the volume of one unit cell is V = 4, and there
are four atoms per unit cell with mass unity. The atomic volume is Ω0 = 1.
In a FCC crystal with pair potential atomic interactions,
c1111 = b1111 /Ω0 , c1122 = c1212 = b1122 /Ω0 . (4.52)
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
4φ (a0 /2) 2φ (a0 /2)

4 4
b1111 = √ 2 , b1122 = b1212 = √ 2 . (4.53)
a0 / 2 a0 / 2
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
Crystal orientation Spring constant k E µ ν cl cs cR

√
[100] 36 3 2 ≈ 28.57 48 18 0.33 8.48 4.24 3.86
√
[100] 72 3 2 ≈ 57.14 96 36 0.33 12 6 5.56
Table 4.3 Elastic properties and wave speeds associated with the harmonic potential
(see (2.34)) in a 3D solid for different choices of the spring constant k, cubical crystal
orientation
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)
This yields wave velocities

(1 − ν) E µ
cl = ≈ 8.48, cs = ≈ 4.24, (4.57)
(1 + ν)(1 − 2ν) ρ ρ
and finally the Rayleigh wave speed is given by
cR ≈ 0.91cs ≈ 3.86. (4.58)

√
For k1 = 2k0 ≈ 57.14, the wave speeds are a factor of 2 larger. The results
are summarized in Table 4.3.
Figure 4.13 shows the numerically estimated elastic properties associated
with the harmonic potential with k0 = 28.57 in the [100] crystal orienta-
tion with Poisson relaxation. The values for the elastic properties show good
agreement. Additional results are shown in Fig. 4.14 for uniaxial loading with-
out Poisson relaxation, and in Fig. 4.15 for triaxial loading. Note that under
uniaxial loading as shown in Figs. 4.13 and 4.14, Young’s modulus increases
slightly with strain, while it decreases with strain under triaxial loading as
shown in Fig. 4.15.
Now we discuss some additional, numerical results of elastic properties for
uniaxial tension with Poisson relaxation in the [110] and the [111] direction.
Figure 4.16a plots the results for uniaxial tension with Poisson relaxation
in the [110] direction. Young’s modulus is approximately E ≈ 72. It is notable
that Poisson’s ratio is different in the y- and the z-direction. The relaxation in
the z-direction is νz ≈ 0.5, and in the y-direction there is no relaxation. This
result, as well as the values for Young’s modulus can also be obtained from
continuum mechanics theories based on generalized Hooke’s law (calculation
not shown here). Unlike the two-dimensional triangular lattice, the three-
dimensional FCC lattice is not isotropic.
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
orientation, without Poisson relaxation. The plot shows the elastic properties as a
function of strain
Lennard-Jones and EAM Potentials

Figure 4.17 shows the stress–strain curves for a pair potential and a multibody
potential. Figure 4.17a shows the results for a LJ potential with nearest
neighbor interaction, and Fig. 4.17b shows the results for an EAM potential
for nickel. In both cases, the [110] direction is very weak and fails at about
12% strain in the case of an LJ potential, and it fails at a strain of only 8%
in the case of an EAM potential. In contrast to this, the cohesive strain in
the [100] direction is largest in the EAM potential and smaller when the LJ
potential is used. The critical cohesive strains are summarized in Table 4.5.
An important difference to the harmonic potentials studied in the previous
section is that Young’s modulus significantly decreases with strain, leading to
a strong softening effect.
For the small-strain elastic properties of a LJ potential describing a three-
dimensional FCC lattice, a direct relation between the LJ potential parameters
σ and 0 can be found. For instance, the bulk modulus can be expressed as
0
K = 64 3 , (4.59)
σ
by assuming only nearest neighbor interactions. There exists a direct relation-
ship between σ and the lattice constant a0 ,
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].
[100] [110] [111]

Potential type Cohesive strain εcoh εcoh εcoh
Lennard-Jones (LJ) 0.25 0.17 0.13
Embedded atom method (EAM) 0.35 0.23 0.08
[100] [110] [111]
Table 4.5 Cohesive strains εcoh , εcoh , and εcoh for the LJ potential and the EAM
potential. In all potentials, the weakest pulling direction is the [110] directions
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
Fracture Surface Energy
Potential γsh γsl

Harmonic (with rbreak = 1.17, k ≈ 28.57) 0.033 0.029
Harmonic (with rbreak = 1.17, k ≈ 57.14) 0.066 0.057
Lennard-Jones 1.029 0.891
Tethered LJ (with rbreak = 1.17) 0.119 0.103
Table 4.6 Summary of fracture surface energies for a selection of different potentials
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.
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
The fracture surface energy of a three-dimensional FCC system can be

expressed as
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.
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.
4.4.5 Concluding Remarks

The studies shown in this section suggest that by designing the interatomic
potential different elastic properties can be obtained.
The LJ system shows a strong softening of Young’s moduli with strains
(see Fig. 4.3). In contrast, the tethered LJ system yields a solid whose elastic
properties stiffen with strain (see Fig. 4.4). The harmonic potential serves as a
reference that yields approximately linear elastic properties (see Fig. 4.5). The
LJ and tethered LJ potential yields continuously changing Young’s moduli,
which may complicate the analysis of materials failure phenomena. Therefore,
we propose a simplistic potential to describe these hyperelastic effects, the
biharmonic potential. The biharmonic potential is composed of two harmonic
potentials and yields bilinear elastic properties (see Fig. 4.11). An important
feature of the biharmonic potential is that it allows to define unique wave
speeds for small and large strains.
The approach described in this chapter exemplifies the development of
model materials for computer simulations. The potentials defined in this
chapter will be used to study specific aspects of the dynamics of brittle fracture.
4.5 Elasticity Due to Entropic Contributions

The dominance of entropic behavior is a well-known phenomenon from
many polymers, including biological structures. The contributions to the
entropic term to elasticity can be described in several ways, including classical
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.
4.5.1 Elasticity of Single Molecules: Worm-Like-Chain Model
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
We first focus on entropic elasticity of a single molecule, the building block

of a polymeric material.
The persistence length is defined as the molecular length at which entropic
contributions to elasticity become important, as the molecule shows signifi-
cant bending purely due to its thermal energy. A molecule with length far
beyond the persistence length will bend, even without application of forces,
and assume a conglomerated, wiggly shape (see Fig. 4.18 for a series of snap-
shots of a single molecule with increasing length, at constant temperature).
With the bending stiffness of a molecule denoted as EI, the persistence length
is defined as
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
A schematic of a convoluted molecule is shown in Fig. 4.19, also showing

the change in molecular configuration as the molecule is stretched. Entropic
effects become important and appear in measurements, for example when one
stretches a convoluted molecule.
Assuming that the initial point-to point distance is x < L (expressing the
fact that the molecule is convoluted) the force that resists stretching is given
by
kB T 1 1 1 x
F (x) = − + . (4.64)
ξp 4 (1 − x/L)2 4 L
This model is called the Worm-Like-Chain Model (WLC) or Marko-Siggia
equation (see, for instance [167,169,170]). The molecular properties enter this
equation in form of the persistence length, which is a function of the bending
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)
4.5.2 Elasticity of Polymers

Similar considerations as those reviewed above have been used to understand
the macroscopic elastic response of rubber – a class of material whose elasticity
is largely dominated by entropic effects. Figure 4.21 depicts a sample unit
cell of an entropic polymer that undergoes deformation, characterized by the
extension ratios λi = rr,new /ri (see Fig. 4.21).
We briefly review how to link the molecular properties to the overall elas-
ticity, using an approach that is similar to the Cauchy–Born rule, following an
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
example reviewed in [168]. Considering a freely jointed Gaussian chain with

links of length each, the entropy of such a chain is
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
By averaging over all chains in the system,

∆S = −kB b2 (λ21 − 1) < x2 > + (λ22 − 1) < y 2 > + (λ23 − 1) < z 2 > . (4.68)
Assuming an isotropic solid, that is, the end-to-end distances of the Nb

chains are directed equally in all directions, we find that
1
< x2 > = < y 2 > = < z 2 > = < rb2 >. (4.69)
3
Since
√ '
rRMS = nl = < rb2 >. (4.70)
and < rb2 > = nl2 , we arrive at

1 2 1
< x2 > = < y 2 > = < z 2 > = nl = 2 . (4.71)
3 2b
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)
linking uniaxial stress σ and uniaxial extension ratio λ through

E 1
σ= λ2 − , (4.76)
3 λ
linking uniaxial stress σ and uniaxial extension ratio λ through E. This
provides a link of microscopic parameters (N ∗ = N/V , the density of molec-
ular chains per unit volume) with a macroscopic measurable, E. From this
equation, we also note that the stiffness is proportional to the temperature,
E ∼ T.
Despite the significant progresses made in advancing the understanding
of brittle or ductile materials, the atomistic and molecular mechanisms of
deformation of natural and biological materials in which entropy governs are
often poorly understood. The study of elastic and plastic deformation of such
materials using large-scale computer simulations poses great challenges and
thus great opportunities.
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
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.
Unit/physical constant Conversion result

1 eV 1.60217646 × 10−19 Nm (1 Nm = 1 J)
1 kcal 4184 Nm (1 Nm = 1 J)
1 kcal mol−1 4184 m2 kgs−2 mol−1
1 mol 6.02214 × 1023 particles (=Avogadro’s number)
1 Pa 1 Nm−2
1 amu 1.6605402 × 10−27 kg
kB , Boltzmann constant 1.3806503 × 10−23 m2 kgs−2 K−1
1J 1.439325215 × 1020 kcal mol−1
1J 0.0002390057361 kcal
1 kcal mol−1 6.947700141 × 10−21 J
1 cal 4.184 J
Table 4.7 Summary of frequently used units to SI units and/or definition of
constants
5
Multiscale Modeling and Simulation Methods
This chapter is dedicated to a discussion of multiscale modeling and simulation

methods. These approaches are aimed to provide a seamless bridge between
atomistic and continuum approaches, sometimes by introducing intermediate
“mesoscopic” methods of simulation. This chapter gives a brief overview over
available models and approaches, along with a discussion of the historical
development. The quasicontinuum method and a hybrid ReaxFF method are
discussed in more detail. Several case studies are reviewed.
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
coupling among different models. This applies to different geometric regions

in a concurrent approach and to different simulation methods in a hierarchi-
cal approach. There are many computational challenges associated with these
schemes, as some of the computational engines may require more computa-
tional effort than others, so that load balancing becomes an important issue.
There exist several review articles on multiscale modeling of materials in the
literature [18, 176, 177], including a mathematical perspective. In addition to
methods aimed to extend accessible lengthscales, several methods have been
developed to cover larger timescales.
In the following sections, we provide a discussion of selected multiscale
approaches. First, we focus on methods to span vast lengthscales, and second,
we focus on methods to span vast timescales.
5.2 Direct Numerical Simulation vs. Multiscale

and Multiparadigm Modeling
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
Material deformation is a phenomenon that cannot be understood at a

single scale alone. It requires the consideration of multiple scales to capture
the progression of the elementary physical mechanisms.
How can one capture multiple scales? One possibility is to simply simulate
all particles in a system. Another possibility is to use a combination of methods
with different accuracy or resolution. This idea is motivated by the fact that
5 Multiscale Modeling and Simulation Methods 159
in many problems, a high resolution and a high accuracy is only required in

regions that are small compared with the overall specimen size. This is for
instance the case in the example of a crack-like defect, where only atoms close
to the crack tip experience large stresses, whereas atoms further away undergo
only small deformation.
The connection of multiple simulation techniques, coupled by handshaking
or parameter passing, is illustrated in Fig. 5.1. We will discuss this figure
later in much more detail and also illustrate in selected case studies how the
handshaking between various computational techniques can be achieved.
The systematic integration of models that range from the quantum mechan-
ical to macroscopic scales can enable one to make quantitative predictions of
complex phenomena with few (or without, in some cases) empirical parame-
ters. Figure 5.2 depicts an overview over the process of predictive multiscale
modeling. Quantitative predictions are enabled via the validation of key prop-
erties, which then enables to extrapolate and predict the behavior of systems
not included in the initial set of parameters used to develop the model.
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
5.3 Differential Multiscale Modeling
Computational modeling can be used to predict quantitative numbers of

material properties. Since parameter or geometry changes can typically be
implemented quite easily, it can also be used to perform a procedure referred
to as differential multiscale modeling. Differential multiscale modeling, in con-
trast to predictive multiscale modeling focuses on the differential aspects of
how macroscopic properties change due to variations of microscopic prop-
erties and microscopic structures. In this approach, it is not required that
each individual simulation provides a quantitative predictive capability of the
phenomena. Rather, it is the change of properties that is predicted. It has
been argued that this approach, in some ways a weak form of the predic-
tive method, may provide a robust set of results and can thereby provide
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.
5.4 Detailed Description of Selected Multiscale Methods

to Span Vast Lengthscales
In this section, the aim is to provide a brief review of selected multiscale
methods used for modeling mechanical deformation in crystalline materials –
in particular metals – along with their advantages, potentials, and drawbacks.
5.4.1 Examples of Hierarchical Multiscale Coupling
An example for hierarchical multiscale modeling is a study of the shear

strength of crystals [178]. The authors investigate the shear strength of crys-
tals based on a multiscale analysis, incorporating molecular dynamics, crystal
plasticity, and macroscopic internal state theory applied to the same system.
The objective of the studies was to compare different levels of description and
to determine coupling parameters. Further studies of climb dislocations in
diffusional creep in thin films [50] and mesoscopic treatment of grain bound-
aries during grain growth processes [179, 180] have also employed hierarchical
simulation approaches.
The development of hierarchical multiscale approaches has also con-
tributed to new methods and simulation strategies in a variety of fields.
For example, mesoscopic dislocations dynamics simulations treat disloca-
tions as particles embedded in a linear elastic continuum (see, for instance
[95, 96, 181, 182]). Dislocation particles interact according to their linear-
elastic fields and move according to empirical laws for dislocation mobility. All
Fig. 5.3 Example for implementation of a hierarchical multiscale method, where

parameters are passed through various lengthscales
nonelastic reactions between dislocations that may occur have to be included

in the simulation setup as interaction rules. Thus, an important issue in these
approaches is to identify proper coupling variables to transition between the
different scales. This is typically achieved by hierarchically coupling to full
atomistic methods.
Multiscale methods have also been particularly useful for polymers and
biological materials. Figure 5.3 depicts a schematic that shows a systematic
coarse graining of the molecular structure of a tropocollagen molecule. The
concept here is to divide a larger molecular structure into a representation of
super-atoms, or beads, which as a whole represent the entire tropocollagen
molecule. A similar approach will be discussed in Sect. 8.7.1 for modeling of
large carbon nanotube systems.
The concept of hierarchical integration of computational approaches has
also been used in the design of new materials, from bottom up, as illustrated
in the design of Cybersteel [18, 19]. Figure 5.4 illustrates the hierarchical inte-
gration of computational tools in the design process. Such new material design
methods are based on the bottom-up multiscale design, using multiscale mod-
eling as the basic engineering tool. In the study reviewed in Fig. 5.4, the design
process begins at the quantum scale, where the particle–matrix interface
decohesion is analyzed by using first principles methods, in order to deter-
mine the appropriate traction-separation law. This traction-separation law is
then embedded into the simulation of the submicron cell that contains sec-
ondary particles. The sequential information flow across multiple hierarchies
is repeated several times until the required macroscopic behavior is achieved.
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
5.4.2 Concurrent Integration of Tight-Binding, Empirical Force

Fields and Continuum Theory
Rigorous multiscale approaches for mechanical properties of solids were first

reported in the 1990s. One of the first approaches in this field included an inte-
gration of atomistic simulation with finite element models [183]. The authors
discuss simulations performed with a hybrid atomistic-finite element (FEAt)
model, and compared the results with the continuum-based Peirls–Nabarro
model for different crack orientations in a nickel crystal. The researchers
demonstrated the basic assumptions of the continuum model for disloca-
tion nucleation, that is, stable incipient slip configurations are formed prior
to dislocation nucleation, and found relatively good agreement of the FEAt

model with the Peierls model for critical loading associated with dislocation
nucleation. In the FEAt model, the region with atomistic detail is determined
prior to computation and cannot be updated during the simulation.
Another concurrent modeling approaches spanning scales from quantum
methods to continuum was developed as a method to model fracture dynamics
of brittle materials [144, 174, 184]. Fracture dynamics is governed by pro-
cesses over a range of interconnected lengthscales, all of which are vital in
deciding how a crack propagates in crystals. This multiscale methodology
links lengthscales ranging from the atomistic scale, treated with a tight bind-
ing (TB) approximation, through microscale, treated by classical molecular
dynamics, to the mesoscale, handled by finite element methods in the con-
text of continuum elasticity. The method was applied to study dynamical
fracture in silicon, a covalently bonded brittle material whose fracture mechan-
ics is intrinsically complicated and cannot easily be captured by empirical
approaches.
The formation and breaking of covalent bonds at and in the vicinity of
the crack tip is treated by a near quantum mechanical approach, a semi-
empirical nonorthogonal tight-binding formulation, which describes the bulk,
amorphous, and surface properties of Si well. Somewhat further away from the
crack tip region, where strains are large, the chemical bonds are not broken
but deviate from their ideal bulk bonding arrangements. In this region, the
Stillinger–Weber [124] empirical interatomic potential is used to describe the
material with classical molecular dynamics. A continuum finite element region
is used in the far-field, where the atomic displacements from ideal positions
and strain gradients are small. The use of three different regions requires two
different hand-shaking zones: FE/MD and MD/TB. The total Hamiltonian
for the system is written as
Htot = HFE + HTB + HFE/MD + HMD/TB . (5.1)
The degrees of freedom of the Hamiltonian are the atomic positions r and
atomic velocities v for the TB and MD regions, and nodal displacements and
their velocities for the FE region.
To achieve FE/MD coupling, the FE mesh dimensions are brought down
to atomic dimensions at the interface between the two regions. Moving away
from the interface and into the continuum the mesh size is expanded. The
FE and MD regions share atoms and nodes on either side of the interface
where they are given half weights each to the Hamiltonian. Since the FE/MD
interface is far from the crack and any plastic deformation zone, the mapping
is unambiguous and gives correct forces near the interface.
For the MD/TB coupling, the dangling bonds on the TB side are saturated
by pseudohydrogen atoms. The Hamiltonian matrix of these pseudohydrogen
atoms are carefully constructed to tie off a single Si bond and ensure absence
of any charge transfer when the atoms are in a perfect Si lattice position.
The TB terminating atoms called “silogens” are fictitious monovalent atoms
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
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
appropriate boundary conditions for the atomistic simulation domain that

mimics the fine atomistic-scale vibrations of atoms. Figure 5.5 depicts a mul-
tiscale analysis of a 15-walled CNT by a bridging scale method. The method
has found many other applications, for instance to model intersonic crack
propagation as reported in [188].
5.4.3 The Quasicontinuum Method and Related Approaches

A quasicontinuum (often also referred to as QC) model for quasistatic
atomistic-continuum simulations was developed by Tadmor, Miller, Ortiz,
Phillips, Shenoy, and coworkers [20, 175, 189] beginning in the mid 1990s.
The quasicontinuum method is based on the observation that in many
large-scale atomic simulations a large section of atomic degrees of freedom
can be described by effective continuum models and only a small subset of
atomic degrees do something different.
The chief objective is to systematically coarsen the atomistic description
by judicious introduction of kinematic constraints on the full atomistic rep-
resentation. These kinematic constraints allow full atomic resolution to be
preserved where required, for example in the vicinity of defects and inter-
faces, and to treat large number of atoms collectively in regions where the
deformation fields vary slowly on the scale of the lattice. For example, in
the simulation of dislocation mechanisms, the method enables to describe
the dislocation core region fully atomistically, while most of the bulk region
surrounding the dislocation is treated as a continuum.
The method begins from a conventional atomistic approach that computes

the energy as a function of atomic positions. The configuration space of the
solid is then reduced to a subset of representative atoms. The position of the
remaining atoms are obtained by piecewise linear interpolations of the repre-
sentative atom positions. Effective equilibrium equations are then obtained by
minimizing potential energy of the solid over the reduced configuration space.
The energy of the system is defined as the weighted sum of the representative
atom energies.
N
Utot ≈ ni U i , (5.2)
i=1
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
Fig. 5.6 Results of a simulation of a crack in a thin film constrained by a rigid

substrate, exemplifying a study using a concurrent multiscale simulation method,
the quasicontinuum approach [20]
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
Fig. 5.7 Application of the quasicontinuum method in the simulation of a nanoin-

dentation experiment. Subplots (a) and (b) depicts a cross-sectional view of the test
sample used in the nanoindentation simulations for increasing indenter penetration
(part of the indenter is also shown). Subplot (c) plots the dislocation structure at the
indenter penetration corresponding to the indentation depth shown in subplot (b).
Subplot (d) shows a load vs. displacement curve predicted by full atomistic (LS) and
quasicontinuum (QC) simulations, illustrating that the two methods show excellent
agreement. Reprinted from Journal of the Mechanics and Physics of Solids, Vol.
49(9), J. Knap and M. Ortiz, An analysis of the quasicontinuum method, copyright
to three-dimensional problems is the representation of three-dimensional

dislocation loops that would extend across the atomistic/continuum interface.
5.4.4 Continuum Approaches Incorporating Atomistic Information
Recently, a virtual internal bond (VIB) model has been proposed as a

bridge of continuum models with cohesive surfaces and atomistic models with
interatomic potentials [157]. The VIB method differs from an atomistic model
in a sense that a phenomenological “cohesive force law” is adapted to act

between material particles, which are not necessarily atoms. A randomized
network of cohesive bonds is statistically incorporated into the constitutive
response of the material based on the Cauchy–Born rule (see Sect. 4.4.1). This
is achieved by equating the strain energy function on the continuum level to
the potential energy stored in the cohesive bonds due to an imposed deforma-
tion. The basic idea of this method is to compute the constitutive equation
directly from the interatomic potential. Other features of the VIB model [157]
can be found elsewhere.
The method has been used to study crack propagation in brittle materi-
als, and is able to reproduce many experimental phenomena such as crack tip
instabilities or branching of cracks at low velocities. An important implica-
tion of the VIB method is that it provides a direct link between the atomic
microstructure and its elastic properties, for any given potential. The method
was recently extended to model viscoelastic materials behavior [193]. The
fact that this method is able to perform simulations on entirely different
lengthscales makes it interesting for numerous applications particularly in
engineering, where more complex situations have to be modeled.
5.4.5 Hybrid ReaxFF Model: Integration of Chemistry

and Mechanics
In this section, we review a method to integrate different force fields within a

single computational domain. This method is not a true multiscale method,
but it rather is a multiparadigm method.
The integration of chemistry and mechanical properties remains a challeng-
ing issue, in particular in systems at finite temperature and when thousands
and more reactive atoms are present. Chemistry at the atomic scale is well
handled by quantum mechanical methods such as QC and DFT methods.
However, the number of reactive atoms in these approaches remains limited
to a few hundred at most, which constitutes a severe limitations in modeling
defect structures and deformation phenomena that emerge at larger scales.
The combination of the ReaxFF force field, which is capable of treating chem-
ically complex materials, with nonreactive potentials is a possible strategy to
overcome these limitations.
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].
However, ReaxFF has a wider range of applicability to other types of atomic

structures, such as hydrocarbons, proteins, and semiconductors.
It has been demonstrated that ReaxFF reproduces quantum mechanical
results for both reactive and nonreactive systems, including hydrocarbons,
nitramines, ceramics, metal alloys, and metal oxides. Due to the complex-
ity of the underlying mathematical expressions in ReaxFF, and the necessity
to perform a charge equilibration (QEq) [132] at each iteration, ReaxFF is
approximately 10–50 times more expensive computationally than simple FFs
such as CHARMM, DREIDING, or covalent force fields such as Tersoff’s force
field. However, ReaxFF is several orders of magnitude faster than quantum
mechanics-based ab initio methods. For details about the ReaxFF method-
ology and development we refer the reader to Sect. 2.6.4, where some of the
important aspects are reviewed.
EAM-based models, on the other hand, have emerged as a well-established
methodology to study mechanical deformation of metals. Fitted to experi-
mental and ab initio data, they reproduce a wide range of properties such
as equations of state, surface energies, stacking fault energies, and others
quite accurately, which are important in correct representation of deformation
behavior. The computational expense of EAM models is on the same order as
that of empirical potentials, and systems of sizes of multibillion atoms have
been investigated using EAM potentials.
Thus a multiparadigm integration of ReaxFF and empirical force fields
such as EAM and Tersoff, to capture chemistry at the regions of interest,
for example at crack surfaces, using ReaxFF, and bulk elastic and plastic
deformation using the EAM or Tersoff model is a quite promising approach
to take advantage of the best of both approaches.
Formulation of the Hybrid ReaxFF Model
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):
Htot = HReaxFF + HEAM + HReaxFF−EAM, (5.3)
where
HReaxFF−EAM = wReaxFF (x)HReaxFF + (1 − wReaxFF (x))HEAM (5.4)
describes the hybrid Hamiltonian formulation in the transition region shown

in Fig. 5.8. The assignment of weights is based on the concept to specify a
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).
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.
FReaxFF−EAM = wReaxFF (x)FReaxFF + (1 − wReaxFF (x))FEAM . (5.7)
In summary, this force handshaking algorithm requires a large transition

region and therefore a slowly varying interpolation function, and if the hand-
shaking is done in a regime in which both potentials provide an identical or
very similar energy landscape. It works well when a transition region of larger
than 7 Å width with a linear or sinusoidal interpolation of the force contri-
butions is used. Figure 5.9 depicts the energy landscape of two force fields,
illustrated here for interactions around the equilibrium position of atoms.
Typically, for small deviations from the equilibrium position both potentials
provide a similar description and as long as the handshaking is done in this
regime, (5.7) provides a suitable approximation.
This hybrid model is in principle not limited to two methods, but it can
be generalized to NC different computational methods:

NC
Htot = wi Hi , (5.8)
i=1
where the weight of NC computational engines add up to unity,

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
where Fj,i is the force contribution on atom j due to computational engine i

and Fj is the resulting force vector on atoms j.
The width of the transition region Rtrans depends on the nature of the
system, but it is generally a few times the typical atomic distance in a lattice or
in an organic molecule. The width of the buffer layer Rbuf describing the ghost
atoms is about 10% larger than any long-range cutoffs to rule out possible
boundary effects. The important point here is that the atoms at the interface
to the ghost atoms (which still contribute a small amount to the total force)
should not sense the existence of the boundary of the buffer layer at any point,
thus Rbuf > Rcut .
Numerical Implementation: The Computational Materials Design

Facility (CMDF)
Multiscale methods often feature great computational complexity, and the

development, application, and modification of numerical implementations may
require significant effort. Thus several techniques have been developed that
aid in making it easier to integrate different computational tools by providing
a modular structure.
For example, the computational materials design facility (CMDF) is a
Python [196]-based simulation framework allowing multiparadigm multiscale
simulations of complex materials phenomena operating on disparate length-
scale and timescale [197]. Individual computational engines are wrapped
using the “Simplified Wrapper and Interface Generator” (SWIG) for rapid
integration of low-level codes with scripting languages [196]. This frame-
work enables complex multiscale simulation tasks encompassing a variety of
simulation paradigms, such as quantum mechanics, reactive force fields, non-
reactive force fields, coarse grain mesoscale, and continuum descriptions of
materials. The CMDF framework enables the combination of ReaxFF to cap-
ture the QM description of reactions with classical nonreactive potentials to
describe nonreacting regions providing the means for describing many complex
materials failure processes as reported herein.
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.
Example of CMDF Model of Oxidation
Figure 5.11 depicts an example study of a nanoscale elliptical penny-shaped

crack in nickel filled with O2 . The system is under 10% tensile strain loading
in the x-direction (orthogonal to the long axis of the elliptical defect).
Oxidative processes leading to formation of an oxide layer are competing
with extension of the crack. In later stages of the simulations, the oxide layer
still remains, keeping the Ni half spaces together, indicating that it involves
strong Ni–O bonds. The reactive region can expand or shrink during the
simulation and is determined by the positions of the oxygen atoms. Failure
initiates by formation of nanovoids in the Ni bulk phase. Classical model-
ing schemes, for example based on the EAM method, cannot describe such
complex organic–metallic systems.
Fig. 5.10 Example CMDF script (upper part) and schematic of the structure of
CMDF (lower part)
5.5 Advanced Molecular Dynamics Techniques

to Span Vast Timescales
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
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
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 )
The simulation is carried out at a temperature of T ≈ 400 K with N = 385

atoms. The high temperature in the TAD method is chosen to be 950 K. The
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
integration time step is δt = 2 × 10−15 s. The diffusivity is then calculated to

be
DsMD = 7.53 × 10−14 m3 /s−1 . (5.12)
This value is comparable to experimental data Dsexp ≈ 11×10−14 m3 /s−1 [204].
The activation energy of all state transitions is found to be 0.57 eV.
We further show an example of how the temperature accelerated method
could be used. Here we consider the atomic activities near a surface step in
a [100] copper surface. We find that atoms at the surface step tend to hop
away from the perfect step. This defines two states (A), the perfect step, and
(B), when the atom is hopped away from the step. The simulation suggests
that over time, the two states A and B interchange. Figure 5.13 shows the
time-averaged stability of the two states as a function of temperature.
It can be observed that for low temperatures, the living time of state (B)
is much smaller compared to that of state (A). State (A) is observed to be
State transition (from to) Activation energy (eV)

A→B 0.609
B→A 0.217
Table 5.1 Activation energy for different state transitions
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.
Fig. 5.15 Hybrid CADD-Parallel Replica study of mechanical twinning of a metal.

Subplot (a) shows the simulation domain, illustrating the continuum/discrete dis-
location regime and the full atomistic domain (blow-up in right part). Subplot (b)
shows a comparison of atomistic simulation results with the predictions of an ana-
lytical model. The plot shows the time to nucleation of a trailing or twinning partial
versus applied load in Al at a temperature of 300 K. The circles refer to the multiscale
simulation results covering many orders of magnitudes in timescales. The dashed
lines correspond to the predictions of the analytical. Reprinted with permission
from Macmillan Publishers Ltd, Nature Materials [21] c 2007
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.
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
Material Deformation and Failure

6
Deformation and Dynamical Failure
of Brittle Materials
In this chapter we review applications of molecular dynamics simulation to

study the fracture behavior of brittle materials. Starting with a review of the-
oretical concepts of dynamic brittle fracture at the continuum scale, we move
on to discuss a simple one-dimensional model of the dynamics of brittle frac-
ture. We proceed with a discussion of two-dimensional and three-dimensional
models, followed by a case study of multiparadigm modeling of fracture of
silicon. This chapter illustrates the use of model potentials in a systematic
application to computational experiments to elucidate how the interatomic
potential properties and the chemistry is linked with macroscopic observables
in dynamic fracture (crack initiation conditions, crack speed, crack surface
structure). This chapter deals with pure brittle fracture (for both elastic
and hyperelastic conditions). The effects of dislocations and plasticity will
be covered in subsequent chapters.
Fig. 6.1 Picture of shattered glass, a model for a brittle material
Figure 6.1 shows a picture of shattered glass. Glass is a very brittle material
that typically shatters into many pieces upon fracture.
6.1 The Nature of Brittle Fracture

As schematically visualized in Fig. 6.2, brittle fracture is a complex mul-
tiscale process. At the scale of several atomic distances (that is, several
Ångstroms), interatomic bonding and the atomic microstructure determine
important material properties for fracture, as for instance the fracture surface
energy. At this scale, the chemistry of atomic interaction and therefore quan-
tum mechanics can play an important role. Breaking of atomic bonds occurs
in the fracture process zone at length scales of several nanometers [22, 61].
In a region around the crack tip extending a few tens of nanometers, the
material experiences large deformation and nonlinearities between stress and
strain become apparent. The macroscopic fracture process on a scale of sev-
eral micrometers can only be understood if the mechanisms on smaller length
scales are properly taken into account.
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
Modeling of dynamic fracture can be quite challenging. A variety of numer-

ical tools have been developed over the last decades. Modeling attempts
focused on cohesive surface models [205], for instance. However, since these
methods are based on continuum mechanics theories, a priori knowledge about
the failure path must be known. Continuum models incorporating atomistic
information like the VIB method [157, 206] overcome this limitation, as they
include information about the underlying atomic lattice (see also discussion
in Sect. 5.4.4), for instance. In contrast, atomistic methods require no a pri-
ori knowledge about the failure. Studying rapidly propagating cracks using
6 Deformation and Dynamical Failure of Brittle Materials 187
atomistic methods is particularly attractive, because cracks propagate at

speeds of km s−1 , which corresponds to nm ps−1 . This scale is readily acces-
sible with classical molecular dynamics methods. Much of the research of
dynamic fracture focused on understanding the atomic details of crack prop-
agation and its relation to macroscopic theories [22] as well as experiments
of fracture [207]. The first part of this review describes simulation work that
mostly treats generic “brittle model materials” rather than specific materi-
als. Afterward we will focus on simulations that discuss fracture in specific
materials.
The studies in the area of dynamic fracture will be focused on the following
points:
• How can one build appropriate atomistic models of brittle materials to
study their fracture behavior?
• How do atomistic simulations results compare with continuum mechanics
theory predictions?
• What is the role of material nonlinearities (that is, hyperelasticity as shown
in Fig. 3.10) in dynamic fracture?
• What is the effect of geometric confinement and crack propagation along
material interfaces?
• What is the effect of the details of chemical interactions, bond rearrange-
ments, or chemically aggressive environments on fracture?
This chapter begins with a discussion of some theoretical aspects and a
review of continuum theory of fracture mechanics, presented in Sect. 6.2.
We then proceed with a discussion of a one-dimensional model of fracture
in Sect. 6.4. We discuss a linear elastic continuum theory serving as a basis for
the extension of the analytical model to the nonlinear case. We report an atom-
istic model of one-dimensional fracture and show that the continuum theories
agree reasonably well with the atomistic simulation results. It is shown that
hyperelasticity (see Fig. 3.10) can significantly alter the dynamics of fracture,
in agreement with the analytical model. The one-dimensional model allows to
study some of the phenomena that also appear in higher dimensional models
in a mathematical and numerical simple framework.
We recall that Sect. 4.4.3 was devoted to a discussion on mechanical and
physical properties of two-dimensional solids. A good understanding of these
material properties is critical to compare the atomistic simulation results
with continuum mechanics theories. We present methods to calculate elas-
tic properties and wave speeds from the interatomic potential. Several choices
of interatomic potentials are discussed. We also address the issue of calculating
the fracture surface energy.
In Sect. 6.5, we report joint continuum-atomistic studies of the deformation
fields near a moving mode I crack in a harmonic lattice. It will be shown
that in harmonic lattices corresponding to linear elastic material, continuum
mechanics theory is a reasonable model. We compare the stress and strain
fields, particle velocity distribution, potential energy field, and energy flow.
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.
6.2 Basics of Linear Elastic Fracture Mechanics

The mechanics of fracture of materials has interested scientists and engineers
for several hundred years. One of the earliest scientists to work on fracture was
Galileo Galilei. He studied the strength of materials approximately 350 years
ago, in 1635, when he stated that one cannot reason from the small to the
large, because many mechanical devices succeed on a small scale that cannot
exist in great size. This statement poses one of the most fundamental questions
related to the strength of materials, relating how defects, or flaws, influence
the strength of materials, and how the size of materials could influence their
strength. It was not until the early twentieth century that scientists had a
good physical understanding about the origins of these effects.
How strong are materials? How does the strength depend on the size of
structures? Such questions have always played a critical role in engineering
science. In particular, with the new paradigms of materials science that were
developed in the second half of the twentieth century, relating structure–
function property, the science of fracture – in particular its theoretical and
analytical, and also experimental investigation – has received tremendous
attention.
Today, the physics of fracture remains a highly active research field. In
particular, with the increased importance of new nanoscale materials, or the
analysis and synthesis of biological materials, fracture theories have received
much attention. Theories of fracture, in particular the continuum mechanics
approach, have proven to be very successful and widely applicable. In fact,
some of the most powerful predictions have been made by the continuum
mechanical treatments of fracture, such as crack limiting speeds or the stress
fields surrounding a moving crack. Some of these theories and models will be
reviewed in the following sections, focusing on important ideas and concepts
of linear elastic fracture mechanics.
6.2.1 Energy Balance Considerations: Griffith’s Model of Fracture
Theoretical fracture mechanics, in particular solutions based on linear elas-

ticity theories applied to crack problems has been extremely successful over
the past 100–150 years. Here we review some of the basic concepts, in par-
ticular the underlying mathematical approaches to derive solutions for crack
problems (for a review of fracture mechanics, please see [209]).
Griffith (1921) [62] formulated a criterion for unstable crack extension
by considering a balance of energies, including mechanical strain energy and
surface energy – the energy necessary to create new surfaces. When a crack
grows, the potential energy stored in the material – at the atomic level, these
energetic contributions come from the distortions of atomic bonds – decreases.
This can be visualized straightforwardly by considering that after fracture
and crack propagation some piece of the material that was previously strained
is now completely relaxed. Its potential energy was transformed into another
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
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)
Fig. 6.4 Thin strip geometry. The gray arrows indicate the mode I (tensile loading),
by a stress σ0
Thus, the energy stored in this geometry – as a function of crack length a –

is given by
σ 2 (1 − ν 2 )
Wp = Wp(2) − Wp(1) = − ξãB (6.11)
2E
with strip width ξ, the energy release rate can be expressed as (note that here
we take the partial derivative with respect to A = aB, which is the area of
one crack surface)
σ 2 (1 − ν 2 )ξ
G= . (6.12)
2E
From classical fracture mechanics, the critical stress for crack nucleation in
this perfectly brittle material is given by the Griffith condition G = 2γs . 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
two new surfaces. Using the Griffith condition we arrive at

4γs E
σf = . (6.13)
ξ(1 − ν 2 )
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
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
Undeformed Stretching=store Release elastic energy

elastic energy dissipated into breaking
chemical bonds
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
It is noted that the characteristic length dimensions in the various cases

considered here (e.g., a finite size crack embedded in a continuum and a thin
strip geometry) have a different meaning, referring to the crack size in the first
case and to the material dimension in the second case. Notably, in the thin
strip case, the fracture stress actually does not depend on the size of the crack
(which is infinity due to its semi-infinite nature), but instead on the width
of the material strip, given by ξ. Solutions for many other geometries can be
derived with similar methods.
To account for plasticity and other mechanisms at the crack tip, the surface
energy is sometimes replaced with γs + γp , where the second term γp can be
much larger than γs . This allows to treat materials that are not perfectly
brittle within the same theoretical framework, including metals. Others have
generalized this idea to include other dissipative effects.
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.
6.2.2 Asymptotic Stress Field and Stress Intensity Factor
Fig. 6.6 Schematic of cracks under mode I, mode II, and mode III crack loading
It is commonly distinguished between different modes of crack loading,

depending on how the load is applied. The geometry of mode I, mode II, and
mode III loading is shown schematically in Fig. 6.6. At least two-dimensional
atomistic models of dynamic fracture are required to describe the behavior
of mode I and mode II cases, and three-dimensional models are required for
mode III cracks.
The stress intensity factor is an extremely useful concept to calculate the
energy release rate G for different geometries and loading conditions. The
stress intensity factor describes the impact of the geometry on the stress field
in the vicinity of a crack. The stress field in the vicinity of the crack tip is
given by an asymptotic solution [22, 210, 211]. With KI as the stress intensity
factor for a mode I crack,
KI (1)
σij (Θ) = √ Σij (Θ) + σij + O(1). (6.18)
2πr
The functions Σij (Θ) represent the variation of stress components with angle
Θ [22]. For mode I loading, the functions Σij are given by

Θ Θ 3Θ
Σxx = cos 1 − sin sin , (6.19)
2 2 2

Θ 3Θ Θ
Σxy = cos cos sin , (6.20)
2 2 2

Θ Θ 3Θ
Σyy = cos 1 + sin sin . (6.21)
2 2 2
The asymptotic field has a universal character because it is independent of

the details of the applied loading, characterized by the stress intensity factor
(1)
KI . The values of σij and the first-order contribution O(1) are determined
from the boundary conditions, and can be neglected in areas very close to the
crack tip. It is important to note that the crack tip represents a singularity
for stresses, as
1
σij ∼ √ . (6.22)
r
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
Some insight can already be gained by studying some scaling relationships.

We note that
KI
σ22 ∼ √ (6.24)
r
and
KI √
u2 ∼ r (6.25)
E
so that
KI2
Gδa ∼ . (6.26)
E
Specifically,
KI
σ22 = , (6.27)
2π(δa + x1 )
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)
For the case of a crack included in an elastic medium,

√
KI = σ0 πa. (6.31)
For a small crack at a surface of an infinite elastic medium loaded in tension

(mode I), √
KI = 1.12σ0 πa. (6.32)
Many other expressions for KI , KII , and KIII can be found in the literature
(e.g., in stress intensity factor handbooks [61, 64, 212]).
6.2.3 Crack Limiting Speed in Dynamic Fracture
After nucleation of a crack in a brittle material, its propagation speed typically

increases significantly. The crack propagation speed is defined as the derivative
of the crack tip position with respect to time,
∂a
v= . (6.33)
∂t
Larger applied mechanical loads generally lead to larger propagation speeds.
However, the maximum speed that a crack can attain is limited by a maximum
value that is related to the speed of waves in the elastic media in which the
crack propagates. This is similar to the speed of light, which provides the
upper bound of the velocity at which light can travel.
The specific limiting speed of cracks depends on their mode of loading

(that is, whether the applied load is mode I, mode II, or mode III). Mode I
cracks are limited by the Rayleigh wave speed cR , mode II cracks are limited
by the longitudinal wave speed cl , and mode III cracks are limited by the
shear wave speed cs . A particular notable feature of mode II cracks is that
even though the limiting speed is cl , velocities between cR and cs are not
admitted, which leads to a velocity gap. Specific crack mechanisms have been
discovered that enable the crack to overcome this gap and “jump” from sub-
Rayleigh speeds directly to intersonic crack speeds (this jump occurs via the
“Burridge–Andrews mechanism” in mode II interfacial cracks).
The physical reason for the limitation of the maximum crack propagation
speed is the dependence of the energy release rate G on the crack velocity, as
given here for a mode I crack:
v
G(v) ∼ 1 − , (6.34)
cR
indicating that the energy release approaches zero when v → cR , and would
assume negative values for velocities in excess of cR (this is a result of the fact
that the stress intensity factor is a function of the crack speed [22]). In this
situation, the crack would represent a source of energy, which is physically
impossible.
The aspect of the crack limiting speed will be discussed in great detail in
the subsequent chapters.
6.3 Atomistic Modeling of Brittle Materials

The earliest molecular dynamics simulations of fracture were carried out more
than 30 years ago by Ashurst and Hoover [213]. Many features of dynamic frac-
ture were described in that paper, although their simulation size was extremely
small (only 64 × 16 atoms with crack lengths around ten atoms). In one of the
first papers on large-scale atomistic modeling of fracture [155], the authors
reported molecular dynamics simulations of fracture in systems up to 500,000
atoms, which was a significant number at the time. In these atomistic cal-
culations, a Lennard-Jones potential as described in (2.30) was used. The
results in [146, 155] were striking because the molecular dynamics simula-
tions reproduced phenomena that were discovered in experiments a few years
earlier [207]. The most important observation was the so-called mirror-mist-
hackle transition. It was observed that the crack face morphology changes as
the crack speed increases. The phenomenon is also referred to as dynamic
instability of cracks. Up to a speed of about one-third of the Rayleigh wave
speed, the crack surface is atomically flat (mirror regime). For higher crack
speeds the crack starts to roughen (mist regime) and eventually becomes very
rough (hackle regime), accompanied by dislocation emission. Such phenomena
were observed at similar velocities in experiments [207]. Since the molecular
dynamics simulations are performed in a perfect lattice, it was concluded that

these dynamic instabilities are a universal property of cracks. The instabilities
were subject to numerous other studies (e.g., [214]) in the following years.
Material cR (in m s−1 ) cs (in m s−1 ) cl (in m s−1 )

Steel 2,940 3,200 6,000
Al 2,850 3,100 6,300
Glass 3,030 3,300 5,800
PMMA 920 1,000 2,400
Table 6.1 Overview over wave speeds in a variety of materials, indicating the
longitudinal wave speed cl , the shear wave speed cs , and the Rayleigh wave speed cR
A question that has attracted numerous researchers is that of the limiting

speed of cracks [22]. The crack speed is limited by an impenetrable barrier
that is related to the speed of sound in the material. The limiting speed
for mode I cracks is the Rayleigh wave speed. For mode II cracks, velocities
below the Rayleigh speed, and those between the shear wave speed and the
longitudinal wave speeds are admissible. Between these two regimes, there is
an impenetrable velocity gap, which led to the uncertainty that mode II cracks
may also be limited by the Rayleigh wave speed. The review clearly points
out that the wave speeds are critically important in the study of cracks in
brittle materials. An overview of wave speeds in several materials is provided
in Table 6.1.
Despite the existence of this velocity gap, experiments have shown that
shear-loaded (mode II) cracks can move at intersonic velocities through a
mother–daughter mechanism [215, 216]. The experiments reported in [215]
provided the first unambiguous evidence that mode II loading drive the the
crack to intersonic speeds, even in purely homogeneous systems with only one
distinct set of wave speeds. Figure 6.8 depicts a velocity analysis of such an
intersonic crack, propagating in Homalite-100. Figure 6.9 shows isochromatic
fringe patterns near the intersonic crack, clearly showing the existence of the
shock Mach cone.
Molecular dynamics simulations reproduced this observation, and provided
a quantitative continuum mechanics analysis of this mechanism [165]. A short
distance ahead of the crack, a shear stress peak develops that causes nucle-
ation of a daughter crack at a velocity beyond the shear wave speed. This
topic is an example where atomistic simulations could immediately be cou-
pled to experiments. This also led to the development of the fundamental
solution of intersonic mode II cracks [217]. The fundamental solution was
then used to construct the solution describing the dynamics of a suddenly
stopping intersonic crack [208].
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
Other researchers [165] reported simultaneous continuum mechanics and

atomistic studies of rapidly propagating cracks. The main objective of the
studies was to investigate if the linear continuum theory can be applied to
describe nanoscale dynamic phenomena. The studies included the limiting
speed of cracks and Griffith analysis [165]. The results suggest that contin-
uum mechanics concepts could be applied to describe crack dynamics even at
nanoscale, underlining the power of the continuum approach.
Materials in small dimensions have also attracted interest in the area of
dynamic fracture. Studies of such kind involve crack dynamics at interfaces
of different materials (e.g., in composite materials). Since interfaces play an
important role in the dynamics of earthquakes, cracks at interfaces have been
significantly studied in recent years [218]. Some investigations revealed that
shear-dominated cracks at interfaces between dissimilar materials can move
at intersonic and even supersonic velocities [219, 220]. If shear-dominated
cracks propagate along interfaces between two dissimilar materials, multiple
mother–daughter mechanisms have been observed, and they were referred to
as mother–daughter–granddaughter mechanisms [219].
Other studies of brittle fracture were based on lattice models of dynamic
fracture [214,221]. These models have the advantage that crack dynamics can
be solved in closed form for some simplified cases [221]. In contrast to the
large-scale molecular dynamics models described above, lattice models are
usually small and do not rely on big computers.
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
In [222], the authors report an overview over atomistic and continuum

mechanics theories of dynamic fracture, emphasizing the importance of the
atomic scale in understanding materials phenomena. They discuss scaling
arguments allowing to study crack dynamics in small atomic systems and
scaling it up to larger length scales comparable to experiment. A study of
fracture in tetravalent silicon based on the Stillinger–Weber (SW) potential
is discussed. The authors state that the SW potential has problems describ-
ing brittle fracture in silicon well, since the experimentally preferred fracture
planes (111) and (110) could not be reproduced. The authors further discuss
other possible potentials for silicon in terms of their applicability to model
fracture of silicon. A velocity gap is discussed implying that at zero tem-
perature there is a minimum speed at which cracks can propagate. Various
simulations of fracture of silicon are summarized [223]. In [221], further issues
of atomic brittle fracture are discussed, such as lattice trapping. Also, the
author showed a relation of crack velocity and loading indicating that there
are regimes of forbidden velocities, so that the crack speed increases discretely
with increase of loading. In another publication they compared the crack veloc-
ity as a function of energy release rate calculated by molecular dynamics to
experimental results [224]. Further discussion on the role of the potential in
dynamic fracture can be found in a review article [220].
Very large-scale atomistic studies of dynamic fracture involving 10–100
million atoms have been reported in the late 1990s [225]. These researchers
studied fracture of silicon nitride, fracture of graphite, and fracture in gal-
lium arsenide. They also report studies of fiber-reinforced ceramic composites
(silicon nitride reinforced with silica-coated carbide fibers). More recently, the
research group reported molecular dynamics simulations with up to one billion
atoms [139,140]. In a review, further approaches of modeling dynamic fracture

are summarized [137].
Others reported a series of molecular dynamics simulations to evaluate
the influence of several aspects on the dynamic crack tip instability based
on various potentials [105]. The authors also report a velocity gap for crack
speeds. They use a particular type of boundary conditions leaving the crack
in an elliptical-shaped boundary with viscous damping at the outside to avoid
reflection of waves from the boundary. Due to its shape similar to a stadium,
it was referred to as “stadium damping” [105]. The crack propagates within
an N V E ensemble in an elliptical “stadium” that is characterized by center
and stadium. Outside this inner ellipse viscous damping or N V T temperature
control is applied. This setup is chosen because stress waves reflecting from
the boundaries can severely influence the dynamics of cracks, leading to crack
arrest. The authors found that the limiting speed of cracks is between 30
and 40% depending on the potential. It was reported that cracks release the
excess energy by emitting strong acoustic waves during breaking of every single
atomic bond. Further, the authors did not observe crack branching since the
velocity was too low for this phenomenon to be observed.
Other research focused on mechanical behavior of quasicrystals [226, 227].
Quasicrystals, for the first time observed in 1984, show a symmetry “between
crystal and liquids” and cannot be described as a Bravais lattice [228]. They
are metallic alloys whose positions of atoms are long range translationary
ordered. Research in this field focused on dislocation motion and crack propa-
gation. Unlike in crystals where dislocations leave the lattice undisturbed after
they have passed, in quasicrystals they leave a phason-wall that weakens the
binding energy and may serve as paths for crack propagation [226, 227, 229].
Atomic studies helped to clarify the fracture mechanism in such materials.
6.4 A One-Dimensional Example of Brittle Fracture:

Joint Continuum-Atomistic Approach
In the following sections, we will discuss a simple one-dimensional model to
illustrate some key aspects of dynamic fracture, in particular focusing on
crack limiting speed, inertia properties of cracks, and the effect of material
nonlinearities on dynamic fracture.
In this first, simple case study it will be illustrated by joint continuum-
atomistic studies of a one-dimensional model of fracture that hyperelasticity,
the elasticity of large strains, plays the governing role in the dynamics of
fracture in brittle materials and that linear theory is incapable of capturing
all phenomena, such as the speed of crack propagation in real materials.
The first part of the section is dedicated to a systematic comparison of the
linear elastic continuum model with molecular dynamics simulations featuring
harmonic interatomic potentials. The results for wave propagation velocities,
the critical condition for fracture, inertia properties of the crack as well as
Fig. 6.10 Geometry of the one-dimensional model of fracture
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
Fig. 6.11 One-dimensional atomistic model of dynamic fracture
that the 1D model has many of the features of higher-dimensional models

of dynamic fracture. For instance, there exists a limiting speed for the one-
dimensional crack associated with the wave velocity, and a critical condition
for fracture initiation similar to the Griffith criterion can be formulated.
Due to its simplicity, the one-dimensional model of fracture seems an ideal
starting point for analyses of the complex dynamics of fracture in nonlinear
materials, rather than immediately relying on two-dimensional models. It will
be shown that it is possible to extend the linear continuum model to describe
the nonlinear dynamics of cracks for a bilinear stress–strain law. This elastic
behavior is characterized by two distinct Young’s moduli, one for small strains
and one for large strains, and provides the most simple constitutive law of
hyperelasticity. The new continuum model predicts that the crack propagates
supersonically, if there exists a local zone around the crack tip with stiffer
elastic properties than in the rest of the material (which is elastically softer).
On the basis of the continuum model, we construct an atomistic model as
illustrated in Fig. 6.11. The model features a one-dimensional string of atoms.
Part of the atoms are bonded to a rigid substrate by a “weak potential,” whose
bonds snap early leading to a finite fracture energy. Bonds between the atoms
never break. Using harmonic interatomic potentials, the elasticity of a string
of atoms corresponds to a straight linear elastic bar of homogeneous material.
Using nonlinear interatomic potentials, the atomistic model is readily able to
model a nonlinear material response. A bilinear stress–strain law as assumed
in the continuum model can be mimicked at the atomic scale by using a
biharmonic potential. The new continuum model of one-dimensional fracture
in nonlinear materials in conjunction with the nonlinear atomistic simulations
allow to carry out simultaneous atomistic-continuum studies of the nonlinear
dynamics of fracture.
The plan of this chapter is as follows. After a review of the linear contin-
uum theory of one-dimensional dynamic fracture, we present the continuum
model for one-dimensional fracture in the nonlinear case. In joint continuum-
atomistic studies, we investigate the predictions of both linear and nonlinear
continuum theories with atomistic simulation results. We find reasonable
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.
6.4.2 Linear-Elastic Continuum Model
The analytical continuum solution is discussed in detail elsewhere [231–233].

We only summarize the main results here. With particle displacement u, par-
ticle velocity u̇ = ∂u/∂t, density ρ, coordinate system x, and stress σ, the
equation of motion in the absence of body forces is
∂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
∂2u ∂2f ∂2u 2

2∂ f
= , = c 0 . (6.38)
∂x2 ∂ξ 2 ∂t2 ∂ξ 2
This solution represents a signal travelling in the positive or negative
x-direction. Also, it follows that a stress wave
∂f
σ=E = E H(ξ) = E H(x ∓ c0 t) (6.39)
∂ξ
is moving with the sound velocity c0 , and the particle velocity
∂f c0
u̇ = ∓c0 = ∓c0 H(x ∓ c0 t) = ∓ σ. (6.40)
∂ξ E
In (6.39) and (6.40), the function H(s) is the unit step function (H(s) = 0
for s < 0, and H(s) = 1 for s ≥ 0). In the model of one-dimensional fracture
(as shown in Fig. 6.10), we assume that the left part of the string of atoms
(which is free and not attached to the substrate) is loaded with stress σ0 . We
assume that the crack front moves at propagation velocity ȧ in the positive
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)
The emitted stress wave is related to the particle velocity

c0
u̇ = σe . (6.43)
E
Equations (6.41) through (6.43) can be solved for the three unknowns σt , σe
and u̇. We define α = ȧ/c0 as the ratio of crack propagation velocity to the
sound velocity. The particle velocity behind the crack tip is given by
ȧ σ0
u̇ = − , (6.44)
1+α E
and the local stress wave behind the crack tip carries
1
σt = σ0 . (6.45)
1+α
The emitted stress wave is
α
σe = − σ0 . (6.46)
1+α
The ratio of local to initial strain is
1
εt /ε0 = , (6.47)
1+α
where ε0 is the initial strain prior to crack propagation. Also,
1 σ0
εt = . (6.48)
1+α E
In these equations, the crack speed ȧ remains an unknown. However, we can
make use of the energy balance
dT dφ
G=W − − = R(α), (6.49)
da da
where W is the external work, dT is the increment of kinetic energy, dφ is the

increment of potential energy, and R(α) is the dynamic fracture resistance.
Balancing kinetic and potential energy using (6.43), (6.44), and (6.46) we
arrive at G = G0 g(α) = R(α) with G0 = σ02 /(2E), and
1−α
g(α) = . (6.50)
1+α
We emphasize that the crack driving force vanishes for α → 1, independent of
how large we may choose G0 , because g(ȧ) → 0 in this case, and therefore the
sound velocity provides an upper bound for the crack propagation velocity.
An energy balance for fracture initiation (ȧ = 0) in the spirit of Griffith’s
analysis leads to an expression for crack initiation
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
Fig. 6.12 Bilinear stress–strain law as a simplistic model of hyperelasticity (mim-

icking the behavior shown in Fig. 3.10). The parameter εon determines the critical
strain where the elastic properties change from local (El ) to global (Eg )
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.
6.4.3 Hyperelastic Continuum Mechanics Model

for Bilinear Stress–Strain Law
If the stress–strain dependence is not linear as assumed in (6.36), the theory

discussed in the last paragraph does not hold. However, the linear theory can
be extended employing the concept of local elastic properties and local wave
velocities, in the spirit of the work discussed in [27]. It was hypothesized that
hyperelastic effects become important in the dynamics of cracks because of
the strong deformation gradients in the vicinity of the crack [27, 219]. Within
a relatively small region, elastic properties may change drastically due to
hyperelastic effects. The term “local” is hereby referred to as the region very
close to the crack tip, and “global” refers to regions far away from the crack tip.
A bilinear stress–strain law serves as a unique tool to study the nonlinear
dynamics of cracks: This model features two Young’s moduli, El associated
with small perturbations from the equilibrium position (strain smaller than
εon ), and Eg associated with large deformations (strain larger or equal than
εon ). The parameter εon allows tuning the strength of the hyperelastic effect.
The bilinear stress–strain law is shown schematically in Fig. 6.12.
There is a conceptual difference between the higher-dimensional models of
fracture and the one-dimensional model of fracture. In the higher-dimensional
models of fracture, the zone of large deformation is local to the crack tip, with
large deformation gradients. In the one-dimensional model of fracture, the
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 .
In the remainder, we confine the investigations to the choice of El /Eg = 4.

Equation (6.54) is then simplified to

σ0 /El if ε0 < εon ,
ε0 = (6.55)
4σ0 /El − 3εon 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
which yields αg = 2αl in the case considered here.

In the following, we derive expressions for the local strain field near the
crack tip for a crack moving in a hyperelastic material. We distinguish two
cases:
• Case 1: The local strain near the crack tip εt , is smaller than the onset
strain of the hyperelastic effect, εon . The crack dynamics is governed by
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).
Case 1: Expanding Region of Local Elastic Properties
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
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
The particle velocity u̇ is given by

αl c0,l σ0
u̇ = − . (6.59)
1 + αl El
Case 2: Local Hyperelastic Region
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)
independent of the crack speed. The second wave has magnitude

(2) σ0 /2 El
εt = − εon + εon + εon (6.61)
Eg Eg
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
Summary of the Predictions of the Hyperelastic Continuum Model
We summarize the major predictions of the hyperelastic continuum model for

a bilinear stress–strain law. We distinguish two cases, case 1 when the local
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.
6.4.4 Molecular Dynamics Simulations of the One-Dimensional

Crack Model: The Harmonic Case
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
example, when choosing electron volt as reference energy ( = 1 eV), Angstrom

as reference length (l∗ = 1 Å), and the atomic mass unit as reference mass
(m∗ = 1 amu), the reference time unit corresponds to t∗ = 1.0181 × 10−14 s.
In the simulation procedure, we distinguish an equilibration phase and
a fracture simulation phase. In the equilibration phase, we initialize the free
part of the bar with a prescribed homogeneous strain, given by ε0 = σ0 /E and
let the system equilibrate for a longer time. During that time, we introduce
a viscous damping force fd,i = −u̇i η into the system with η = 0.3 to damp
out waves generated during equilibration, so that the particle velocities (and
strain gradients) are damped out relatively fast. During equilibration, atomic
bonds glued to the substrate can never break, and the total energy of the
system is given by
1
1

Utot = k(rij − a0 )2 + H(i − Nf )kp r̂i2 (6.63)
i,j
2 i
2
where k is the spring constant for interatomic interaction, and kp is the spring
constant of the pinning potential. The variable
r̂i =| x0,i − xi |, (6.64)
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)
The wave velocity in a string of atoms is given by

E
c0 = (6.68)
ρ
with density ρ = m/a0 for the present one-dimensional lattice. For k =

28.5732, Young’s modulus E = 32.07, and c0 ≈ 6. The elastic properties
are determined numerically as a check if the assumptions are valid.
We define an atomic strain of atom i which is directly related to the con-
tinuum mechanics concept of strain [112], considering only nearest neighbors
in a one-dimensional system
xi−1 − xi+1
εi = . (6.69)
2 a0
In the remainder of this chapter, we preferably use the atomic strain to analyze
the simulation results, since it provides a useful way to study the state of
deformation in the atomic lattice.
We start with a comparison of the theoretical prediction of the elastic
properties of the one-dimensional string of atoms with atomistic simulations.
The numerically estimated elastic properties agree well with the theory. The
measurements of applied stress σ0 vs. strain, and the numerically estimated
local modulus in the string of atoms match the theoretical predictions given by
(6.67) well. Additional studies of wave propagation velocity show good agree-
ment of the predicted wave velocity with the measured wave velocity. Simula-
tions with other spring constants and consequently other wave velocities pro-
vide evidence that the agreement of theory and simulation is generally good.
Griffith criterion predicts that fracture initiates when the elastic energy
released per unit crack advance equals the energy to create free surface per unit
crack advance. The fracture energy is given by (6.66). Setting this quantity
equal to the energy release rate allows to determine the critical load to initiate
fracture. The computational results are compared to the theory prediction in
Table 6.2. A result of the simulations is that the results converge to the theory
prediction as kp becomes much smaller than k, but we find larger disagreement
with the theory prediction if kp is large. This could be due to the fact that
the fracture process zone becomes very small when kp is large, leading to very
large strain gradient at the crack.
Equation (6.46) predicts that the local stress wave depends on the crack
propagation velocity. Figure 6.14 plots the magnitude of the local stress wave
for different propagation velocities from atomistic simulations, in comparison
with the theory prediction.
The dynamic fracture toughness is a function of α and σ0 , and is given by
(6.52). Atomistic simulations provide an ideal tool to provide information on
this curve. Figure 6.15 plots the dynamic fracture toughness for different crack
propagation velocities. As can be verified, the assumption that R0 = const. is
reasonable as long as the crack velocity is below 80% of the wave velocity. For
Ratio kp /k Predicted initiation load R0pred Measured initiation load R0
10.0 0.0039 0.00014 × 10−4

1.0 0.0039 0.0015
0.1 0.0039 0.0030
0.01 0.0039 0.0040
0.003 0.0039 0.0040
Table 6.2 Critical load R0 for fracture initiation, for different values of the spring
constant kp of the pinning potential. The results are in good agreement with the
theory prediction when kp becomes much smaller than k
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
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
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
energy. A variation in fracture energy is achieved by varying the bond snapping

distance r̂ of the pinning potential (see (6.66)) according to
r̂ = r̂0 + ∆r̂ sin(x/p), (6.72)
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,
Fig. 6.17 Prescribed fracture toughness and measured crack velocity as the crack
proceeds along x
v = v̂0 + ∆v sin(x/p), (6.73)
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.
6.4.5 Molecular Dynamics Simulations of the One-Dimensional

Crack Model: The Supersonic Case
This section is dedicated to molecular dynamics simulations of supersonic

cracks. To achieve a bilinear stress–strain law according to (6.53), the total
potential energy of the nonlinear system is given by
1 1

Utot = k(rij − a0 ) + βk H(r − ron )(rij − ron )
2 2
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
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.
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.
6.4.6 Discussion and Conclusions
We have used a simple one-dimensional model of dynamic fracture to investi-

gate fundamentals of the nonlinear dynamics of fracture. On the basis of the
continuum model of one-dimensional dynamic fracture, we have proposed an
atomistic model of a string of atoms. We have verified that the continuum
model of one-dimensional dynamic fracture can be successfully applied at the
atomistic level, if harmonic interactions are assumed between atoms. It was
shown that the one-dimensional crack carries no inertia, a phenomenon that is
also found in higher dimensions [23,208,234]. The fact that we find good agree-
ment of the one-dimensional atomistic model featuring harmonic interactions
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
with the continuum theory corresponds to work on the comparison of the

atomistic level with continuum theory [165] for mode II cracks.
Finding analytical solutions for dynamic fracture in hyperelastic materials
in higher dimensions is very difficult, if not impossible in many cases. However,
analytical understanding of the nonlinear dynamics becomes possible based
on the simple one-dimensional model. We have proposed a continuum model
based on the local elastic properties to predict the elastic fields around the
crack tip, when a bilinear stress–strain law is assumed. The major prediction
of the continuum model is supersonic crack propagation, if there exists a
local elastically stiff region confined to the crack tip. By molecular dynamics
simulations, it was shown that the local elastic properties at the crack tip
indeed govern the dynamics of fracture, in agreement with the predictions of
the model. If there is an elastically stiff zone close to the crack tip, the crack
can propagate supersonically through the material. We emphasize that this
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
Fig. 6.22 Particle velocity field near a supersonic crack, comparison between theory
and simulation
one-dimensional molecular dynamics simulations as reviewed here, and found

similar results.
This study illustrates the potential of an atomistic appraoch in studying
brittle fracture. According to conventional continuum-type theories of frac-
ture, it has been widely 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 con-
cept of local wave speeds. In materials where the large-strain elasticity differs
significantly from the small-strain elasticity, the concept of global wave veloc-
ities cannot be used any more to describe the dynamics of the crack. Instead,
the concept of local elastic properties, and associated local wave velocities gov-
ern the dynamics of the crack. Since “real” materials all show strong nonlinear
effects, this suggests that hyperelasticity is crucial for dynamic fracture.
The one-dimensional model has found useful applications in addressing
other fundamental questions of mechanics of materials. A potential applica-
tion is strain gradient effects in elasticity, and its possible implications on
dynamic fracture. The mechanics of one-dimensional structures could also
be important in the emerging bio-nanotechnologies, often involving function-
alization of single molecules. An objective of future studies could be the
development of experimental techniques based on the one-dimensional model
of fracture.
6.5 Stress and Deformation Field near Rapidly

Propagating Mode I Cracks in a Harmonic Lattice
In this section, we review large-scale atomistic simulations to study the near-
crack elastic fields, focused on mode I dynamic fracture from both atomistic
and continuum mechanics viewpoints. The static solution was discussed in
Sect. 6.2.2. Here we review the more general solution that also includes the
case of moving cracks. In the continuum theory, the stress field in the vicinity
of the crack tip is given by an asymptotic solution [22,210,211]. With KI (t, v)
as the dynamic stress intensity factor,
KI (t, v) (1)
σij (Θ, v) = √ Σij (Θ, v) + σij + O(1). (6.76)
2πr
The functions Σij (Θ, v) represent the variation of stress components with
angle Θ for any value of crack speed v [22]. These functions only depend on
the ratio of crack speed to wave speeds. The functions Σij (Θ, v) that appear
in (6.76) are defined as follows [22].
. /
1 cos(1/2Θl ) cos(1/2Θs )
Σxx (Θ, v) = (1 + α2s )(1 + 2α2l − α2s ) √ − 4α α
s l √ ,
D γl γs
. / (6.77)
2αl (1 + αs )2 sin(1/2Θl ) sin(1/2Θs )
Σxy (Θ, v) = √ − √ , (6.78)
D γl γs
and
. /
1 2 2 cos(1/2Θl ) cos(1/2Θs )
Σyy (Θ, v) = − (1 + αs ) √ − 4αd αs √ . (6.79)
D γl γs
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].
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
prediction of stress and strain fields by continuum mechanics is reproduced

quantitatively. An important observation is that the hoop stress becomes
bimodal at about 73% of Rayleigh speed, in agreement with the continuum
theory. In addition, we report comparison of continuum theory with molecular
dynamics simulation of the strain energy field near the crack tip as well as
the energy transport field near rapidly moving cracks.
6.5.1 Stress and Deformation Fields

In this section, we compare stress and deformation field near a rapidly moving
crack tip with continuum mechanics theories.
Angular Variation of Stress

We analyze the angular variation of the principal stress and hoop stress close
to the crack tip and compare the results of the simulation to the continuum
mechanics solution given by (6.76). Atomic quantities are evaluated in a small
region around a constant radius of r ≈ 11 centered at the crack tip. The
continuum theory solution and the simulation results are both normalized
with respect to the dynamic stress intensity factor.
We find that if the stress field measurements are taken while the crack
accelerates too rapidly, the agreement of measured field and continuum theory
prediction can be poor. Acceleration effects can severely change the resulting
stress fields. Although the crack tip is regarded as inertia-less since it responds
immediately to a change in loading or fracture surface energy, it takes time
until the elastic fields corresponding to a specific crack speed spread out! In
fact, the fields spread out with the Rayleigh velocity behind, and with the
shear wave velocity ahead of the crack. In other regions around the crack tip,
the fields are reached in the long-time limit (t → ∞) [22, 23]. Therefore, we
choose a moderate strain rate ε̇xx = 0.000 01.
As a consequence of the relatively low strain rate and the finite slab size,
the crack only achieves about 87% of Rayleigh wave speed. We calculate the
stress for different crack speeds ranging from 0 to 87% of the Rayleigh speed.
Figures 6.24–6.26 show the angular variation of σxx , σyy as well as σxy .
Figure 6.27 shows the angular variation of the hoop stress σΘ . Figure 6.28
shows the angular variation of the maximum principal stress σ1 near the
crack tip. In all figures, the continuous line is the corresponding analytical
continuum mechanics solution [22]. It can be observed from the plots that the
hoop stress becomes bimodal at a velocity of about 73% of the Rayleigh wave
velocity. This is in agreement with the predictions by continuum mechanics
theories [22].
Elastic Fields near the Crack Tip

Here we use a higher strain rate of ε̇xx = 0.0005 to drive the crack close to
the Rayleigh wave speed.
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-
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
Fig. 6.26 Comparison between σxy from molecular dynamics simulation with har-
typical trimodal structure of the asymptotic principal strain and principal

stress field develops close to the Rayleigh velocity, in contrast to the bimodal
structure at low crack speeds.
The stress fields σxx , σyy , and σxy for a crack propagating close to the
Rayleigh wave velocity are shown in Fig. 6.30(a–c). As before, the upper plot
is the simulation result, while the lower part plots the prediction by continuum
mechanics.
Finally, Fig. 6.29 plots the particle velocity near the crack tip for a
crack propagating close to the Rayleigh velocity. Figure 6.31a shows u̇x , and
Fig. 6.31b shows u̇y . The continuum theory prediction and the atomistic sim-
ulation result match well. The particle velocity field behind the crack tip is
found to be smeared out more in the simulation results due to thermalization
effects not accounted for in the continuum theory.
6.5.2 Energy Flow near the Crack Tip
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
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
Pj = σij u̇i + (U + T )v δ2j , (6.86)
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]
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.
6.5.3 Limiting Velocities of Mode I Cracks in Harmonic Lattices
We also study the dependence of crack dynamics in harmonic materials with

different spring constants. Linear elastic fracture mechanics predicts that the
limiting crack speed for mode I cracks should only depend on the elastic prop-
erties, and therefore, in case of harmonic potentials, on the spring constant k.
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
Simulations of cracks propagating along a confined fracture path in a harmonic

lattice show that continuum mechanics theory of fracture can be successfully
applied even at the atomistic level. We compared the virial stress and strain
from atomistic simulation results with the continuum mechanics solution of
the asymptotic field for different propagation velocities.
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
The results suggest that the agreement of molecular dynamics simula-

tions and continuum mechanics is generally good, as it is shown for the stress
tensor components σxx , σyy , and σxy . It is observed that there is some dis-
agreement at larger angles Θ > 150◦ , perhaps due to surface effects in the
atomistic simulations. In Fig. 6.24 it is observed that for σxx , the shape of
σxx (Θ) is qualitatively reproduced well over the entire velocity regime between
0 and 87% of Rayleigh speed. However, the angles of the local maxima
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
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.
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 Crack Limiting Speeds of Cracks: The Significance

of Hyperelasticity
The elasticity of a solid clearly depends on its state of deformation. Metals
will weaken or soften, and polymers may stiffen as the strain approaches
the state of materials failure. It is only for infinitesimal deformation that
the elastic moduli can be considered constant and the elasticity of the solid
linear. However, many existing theories model fracture using linear elasticity.
Certainly, this can be considered questionable since material is failing at the
tip of a dynamic crack because of the extreme deformation, as illustrated in
Fig. 6.35. We review studies that show by large-scale atomistic simulations
that hyperelasticity, the elasticity of large strains, can play a governing role
in the dynamics of fracture and that linear theory is incapable of capturing
all phenomena. We introduce the concept of a characteristic length scale for
the energy flux near the crack tip and demonstrate that the local hyperelastic
wave speed governs the crack speed when the hyperelastic zone approaches
this energy length scale.
Large-scale atomistic simulation studies reviewed here show that hyper-
elasticity, the elasticity of large strains, can play a governing role in the
dynamics of brittle fracture. This is in contrast to many existing theories
of dynamic fracture where the linear elastic behavior of solids is assumed
sufficient to predict materials failure [22, 61, 240].
Some experimental work [241, 242] as well as many computer simula-
tions [146, 155, 205] have shown a significantly reduced crack propagation
speed in comparison with the predictions by the theory. In contrast, other
experiments indicated that over 90% of the Rayleigh wave speed can be
achieved [25, 224, 243–247]. Such discrepancies between theories, experiment,

and simulations cannot always be attributed to the fact that real solids feature
a variety of imperfections such as grain boundaries and microcracks (either
preexisting or created during the crack propagation), as similar discrepancies
also appear in molecular dynamics simulations of cracks traveling in perfect
atomic lattices. Earlier studies have independently led to the conclusion that
hyperelastic effects at the crack tip may play an important role in the dynam-
ics of fracture [27, 146, 155, 248]. Their suggestions have been used to help
explaining phenomena related to crack branching and dynamic crack tip insta-
bility, as well as explaining the significantly lower maximum crack propagation
speed observed in some experiments and many computer simulations. How-
ever, it is not generally accepted that hyperelasticity should play a significant
role in dynamic fracture.
One reason for this belief stems from the fact that the zone of large defor-
mation in a loaded body with a crack is highly confined to the crack tip, so
that the region where linear elastic theory does not hold is extremely small
compared to the extensions of the specimen [22, 61]. In this study, we use
large-scale molecular dynamics simulations [219] in conjunction with contin-
uum mechanics concepts [22, 61] to prove that hyperelasticity can be crucial
for understanding dynamic fracture.
The study reviewed here shows that local hyperelasticity around the crack
tip can significantly influence the limiting speed of cracks by enhancing or
reducing local energy flow. This is true even if the zone of hyperelasticity is
small compared to the specimen dimensions. The hyperelastic theory com-
pletely changes the concept of the maximum crack velocity in the classical
theories. For example, the classical theories clearly predict that mode I cracks
are limited by Rayleigh wave speed and mode II cracks are limited by longi-
tudinal wave speed. In contrast, both super-Rayleigh mode I and supersonic
mode II cracks are allowed by hyperelasticity and have been seen in computer
simulations [156, 219].
In the simulations, it is found that there exists a characteristic length
scale associated with energy flow near the crack tip such that hyperelastic-
ity completely dominates crack dynamics if the size of hyperelastic region
approaches this characteristic length. In earlier simulations [156, 219], a non-
linear interatomic stiffening was assumed, and there was no sharp distinction
between the linear and nonlinear elastic regimes for the stretched solid. In con-
trast, the model is based on a biharmonic potential composed of two spring
constants, one associated with small deformations and the other with large
deformations (see discussion in Sect. 4.4.3). This serves as a simplistic model
material for hyperelasticity, allowing us to investigate the generic features of
hyperelasticity common to a large class of real materials.
6.6.1 Modeling
We consider propagation of a crack in a two-dimensional simulation geometry

shown in Fig. 6.23. 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. To avoid crack branching, a weak fracture layer
is introduced by assuming that atomic bonds across the prospective crack path
snap at a critical atomic snapping distance rbreak while those in the rest of
the slab never break. As outlined in Sect. 4.4.3, the snapping distance can be
used to adjust the fracture surface energy 2γ.
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”)
For a systematic study of hyperelastic effects in dynamic fracture, we adopt

the biharmonic potential defined in (6.53). This potential is composed of two
spring constants k0 and k1 . Here 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 perturba-

tions from the equilibrium distance r0 , and the second spring constant k1 is
associated with large bond stretching for r > ron . The role of√k0 and k1 is
reversed in the elastic softening system (k0 = 2k1 and k1 = 36/ 3 2). The elas-
tic properties associated with this potential are shown in Fig. 4.11. Figure 6.36
shows how the biharmonic potential is used to model a continuous stiffening
effect, and how this model can be used to control the size of the hyperelas-
tic region near the crack. A similar approach is used to model hyperelatic
softening material behavior.
To strain the system, we use two approaches. The first is using a constant
strain rate applied over a loading time by displacing the outermost rows of
atoms. After the loading time, the boundaries are kept fixed. In the second
method, we strain the system prior to simulation in the loading direction, and
keep the boundary fixed during simulation. In either way, the crack starts to
move once a critical strain is applied. It can be shown that the stress intensity
factor remains constant in a strip geometry inside a region of [114]
3/4lx < a < (ly − 3/4lx). (6.88)
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].
6.6.2 Crack Speed and Energy Flow
Molecular dynamics simulations suggest that a localized, small hyperelastic

region around the crack tip can have significant effects on the dynamics of
crack propagation. In all simulations, the slab is statically loaded with 0.32%
strain in mode I. The strain energy density far ahead of the crack tip is given
by
ε2 lx E
S = xx , (6.89)
2(1 − ν)2
where E is the Young’s modulus at small strain. The linear elastic expression
of strain energy density is valid because material far ahead of the crack is
strained always below the onset threshold of the bilinear law, that is it remains
in the linear elastic regime of material response. The strain and strain energy
density both vanish far behind of the crack. For a unit distance of crack
propagation, a strip of material with energy density S ahead of the crack is
replaced by an identical strip with zero strain energy behind the crack.
Fig. 6.37 Hyperelastic region in a (a) softening and (b) stiffening system
According to the linear elastodynamic theory of fracture [22], the crack

speed should satisfy the dynamic energy release rate equation
2γ
A(v/cR ) = (6.90)
S
where the function A(v/cR ) is a universal function of crack velocity v for a
given material. Assuming that the small-strain elasticity completely governs
the dynamics of fracture, the linear theory predicts that crack velocity should
depend only on the ratio S/γ.
During crack propagation, the energy stored ahead of the crack tip is
partly converted by the bond breaking process into fracture surface energy,
and partly dissipated into atomic motion. In the purely harmonic case, the
fracture surface energy γ depends on rbreak and E. In the biharmonic case, the
fracture surface energy depends on rbreak , ron , E0 , and E1 . The strategy is to
focus on the prediction from linear theory that crack velocity depends only on
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.
6.6.3 Hyperelastic Area
A geometric criterion based on the principal strain is used to characterize the

area with hyperelastic material response close to the crack tip. The region
occupied by atoms having a local maximum principal strain
ron − r0
ε1 ≥ εon = (6.91)
r0
defines the hyperelastic area AH by an integral over the whole simulation
domain Ω
AH = H(ε1 − εon ) dΩ. (6.92)
Ω
Figure 6.37a shows the hyperelastic area in the case of a stiffening mate-
rial, and Fig. 6.37b shows the hyperelastic area in the case of an elastically
softening material, indicating that the hyperelastic effect is highly localized
to the crack tip (these pictures show a portion of the simulation slab near the
crack tip). However, the effect of hyperelasticity on crack velocity is significant,
independent of the slab size.
Enhancement or Reduction of Energy Flow
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)
Fig. 6.38 Hyperelastic region and enhancement of energy flow in the (a) softening
and (b) stiffening system
may be identified as a measure for the local energy flow.

A measure for the enhancement or reduction of energy flow is obtained by
subtracting the magnitude of the dynamic Poynting vector in the harmonic
case from that in the biharmonic case at every point in the slab
∆P = Pbiharm − Pharm . (6.94)
If the difference is negative, energy flow is reduced, and if the difference is

positive, energy flow is enhanced. The steady-state fields are averaged over
space as well as time to obtain good statistics.
Figure 6.38 shows the energy flow enhancement and reduction in the vicin-
ity of the crack tip for the elastically stiffening bilinear system (a) and for the
elastically softening system (b). In each plot, the local hyperelastic zone is
indicated by a dotted line. The energy flow in the vicinity of the crack tip is
enhanced in the bilinear stiffening case and reduced in the softening case. In
these plots, we also indicate the direction of energy flow with arrows and note
that in the softening case, the energy flow ahead of the crack almost vanishes.
The plots show that the local hyperelastic effect leads to an enhancement
(stiffening system) or reduction (softening system) in energy flow. The small
hyperelastic regions enhance the energy flow around the crack tip. The higher
crack velocity in the stiffening system and the lower velocity in the softening
system are due to enhancement or reduction of the energy flow in the vicinity
of the crack tip. Table 6.3 summarizes change of net energy flow, as well as
change of energy flow toward and away from the crack tip, in comparison to
the harmonic system. The results quantify those depicted in Fig. 6.38a, b and
show that the net energy flow as well as the flow of energy toward and away
from the crack tip are all enhanced in the stiffening case, and reduced in the
softening case.
Change of net Change of Change of Change of

energy flow to energy flow energy flow away limiting
crack tip to crack tip from crack tip speed
Stiffening +19% +20% +25% +20%
Softening −32% −32% −35% −30%
Table 6.3 Change of energy flow to the crack tip, due to a bilinear softening or
stiffening interatomic potential
Fig. 6.39 J-integral analysis of a crack in a harmonic, softening and stiffening

material, for different choices of the integration path Γ . The straight lines are a
linear fit to the results based on the calculation of the molecular dynamics simulation
studies
J-Integral Analysis
The integral of energy flux, or path-independent dynamic J-integral is defined

as
F (Γ ) = (σij nj u̇i + (U + T )v n2 ) dΓ. (6.95)
Γ
It can be shown that its value is path-independent for steady-state crack
motion [22].
We find that F (Γ ) around the crack tip increases by 19% in comparison
with the harmonic case for the stiffening system, while decreasing by 32%
for the softening system. The results are shown in Fig. 6.39. To calculate
the integral, we choose a circular shape of Γ centered around the crack tip.
Atomic quantities like stress and particle velocity are averaged spatially and
over time and then the line integral is computed. The plot shows the value
of the dynamic J-integral is independent of the shape of Γ which proves that
crack motion is in steady state.
6.6.4 How Fast can Cracks Propagate?
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.
Intersonic Mode I Cracks
Figure 6.40 plots the crack velocity as a function of the hyperelasticity

onset strain εon . The crack speeds shown in Fig. 6.40 are determined during
steady-state propagation. It is observed that the earlier the hyperelastic effect

is turned on, the larger the limiting velocity. Measuring the hyperelastic area
AH using the principal strain criterion, we find that AH grows as εon becomes
smaller. A correlation of the square root of the hyperelastic area with the
achieved limiting speed of the crack is shown in Fig. 6.40. In Fig. 6.41, we
depict the shape of the hyperelastic area near the crack tip for different choices
of εon . The shape and size of the hyperelastic region is found to be indepen-
dent of the slab width lx . In all cases, the hyperelastic area remains confined
to the crack tip and does not extend to the boundary of the simulation.
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.
Supersonic Mode II Cracks
We have also simulated a shear-dominated mode II crack using the biharmonic

stiffening potential. We define rbreak = 1.17, and ron is chosen slightly below
to keep the hyperelastic region small. The dynamic loading is stopped soon
after the daughter crack is nucleated [156, 165, 219]. The result is shown in
Fig. 6.43. The daughter crack nucleated from the mother crack propagates
supersonically through the material, although the hyperelastic zone remains
localized to the crack tip region.
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
Supersonic mode II crack propagation as shown in Fig. 6.44 has been

observed in other molecular dynamics studies [219] using an anharmonic stiff-
ening potential. However, a clearly defined hyperelastic zone could not be
specified in these simulations. Our result proves that a local hyperelastic stiff-
ening effect at the crack tip causes supersonic crack propagation, in clear
contrast to the linear continuum theory. The observation of super-Rayleigh
and intersonic mode I cracks, as well as supersonic mode II cracks, clearly
contradicts the prediction by the classical theories of fracture.
6.6.5 Characteristic Energy Length Scale in Dynamic Fracture
The problem of a super-Rayleigh mode I crack in an elastically stiffening

material is somewhat analogous to Broberg’s [249] problem of a mode I crack
propagating in a stiff elastic strip embedded in a soft matrix.
The geometry of this problem is shown in Fig. 6.45. Broberg [249] has
shown that, when such a crack propagates supersonically with respect to
the wave speeds of the surrounding matrix, the energy release rate can be
expressed in the form
σ2 h
G= f (v, c0 , c1 ) (6.97)
E
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
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.
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.
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.
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].
6.7 Crack Instabilities and Hyperelastic Material

Behavior
Cracks propagating in homogeneous materials show a very interesting dynam-
ics: From experiment [207] and computer simulation [146, 155, 157, 206] it is
known that cracks propagate straight for low speeds with perfect cleavage
(“mirror”), and become unstable at higher speeds. The onset of instability
results in an increasingly rough crack surface (‘mist”), which becomes more
intense when the crack speed increases further (“hackle”). This phenomenon
was referred to as the “mirror-mist-hackle” transition. Computer simulation
played an important role in this area [155], since it showed that the crack
tip instability also occurs in perfect atomic lattices and is therefore not due
to material imperfections. It was proposed [27, 155, 207] that the mirror-mist-
hackle transition is due to an intrinsic dynamic crack tip instability.
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)
The dynamic crack tip instability can nicely be observed in LJ systems

[146, 155]. A simulation result of such a study is shown in Fig. 6.48. After an
initial phase where cleavage is mirror-like, the crack surface starts to roughen
at about 30% of the Rayleigh wave speed. Eventually, the crack surface turns
into a hackle region accompanied by emission of dislocations. The final speed
of the crack is around 50% of the Rayleigh speed. These observations are in
agreement with the results discussed in [146, 155].
Fracture instabilities have already been predicted in the classical literature
based on linear elastic fracture mechanics theories, where it has been proposed
that the governing stress measure determining the direction of crack propa-
gation is the circumferential or hoop stress [22, 210]. If this view of crack tip
instabilities is adopted, the instability should occur at speeds around 73% of
Rayleigh wave speed. In contrast to this, in many experiments [207] as well as

several computer simulations [146, 155] the crack tip instability was observed
at speeds as low as 1/3 of the theoretical limiting speed. In this section, we
will show that this discrepancy may also be due to hyperelasticity.
The outline of this section is as follows. We begin with a discussion of
the direction of stable crack propagation in harmonic lattices. In agreement
with the analysis described in Sect. 4.4.3, crack propagation is stable along the
direction of lowest fracture surface energy for low velocities. We then continue
with a discussion of the crack-speed dependent tip instability in harmonic
solids and show that in agreement with Yoffe’s analysis [210], cracks become
unstable at a speed of about 73% of Rayleigh wave speed. In contrast, by
using a softening LJ potential it will be shown that the instability occurs at
crack velocities around 1/3 of Rayleigh wave speed [29]. We then continue with
results using stiffening and softening potentials and show that hyperelasticity,
the elasticity of large strains, plays a governing role in the stability of cracks.
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
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.
6.7.2 Design of Computational Model
Figure 6.49 illustrates the concept of hyperelastic softening in contrast to lin-

ear elastic behavior. In the spirit of “model materials” as introduced earlier,
we develop a new, simple material model which allows a systematic transition
from linear elastic to strongly nonlinear material behaviors, with the objec-
tive to bridge different existing theories and determine the conditions of their
validity. By systematically changing the large-strain elastic properties while
keeping the small-strain elastic properties constant, the model allows us to
tune the size of hyperelastic zone and to probe the conditions under which
the elasticity of large strains governs the instability dynamics of cracks. In
the case of linear elastic model with bond snapping, we find that the insta-
bility speed agrees well with the predicted value from Yoffe’s model. We then
gradually tune up the hyperelastic effects and find that the instability speed
increasingly agrees with Gao’s model. In this way, we achieve, for the first
time, a unified treatment of the instability problem leading to a generalized
model that bridges Yoffe’s linear elastic branching model to Gao’s hyperelastic
model.
Fig. 6.49 The concept of hyperelastic softening close to bond breaking, in

comparison to the linear elastic, bond-snapping approximation
Although simple pair potentials do not allow drawing conclusions for

unique phenomena pertaining to specific materials, they enable us to under-
stand universal, generic relationships between potential shape and fracture
dynamics in brittle materials; in the present study we use a simple pair poten-
tial that allows the hyperelastic zone size and cohesive stress to be tuned. The
potential is composed of a harmonic function in combination with a smooth
cut-off of the force based on the Fermi–Dirac (F–D) distribution function to
describe smooth bond breaking. We do not include any dissipative terms. The
force vs. atomic separation is expressed as
0 1−1
dφ Ξ
(r) = k0 (r − r0 ) exp r +1 . (6.102)
dr rbreak − Ξ
Assuming that the spring constant k0 is fixed, the potential has two additional
parameters, Ξ and rbreak . The parameter rbreak (corresponding to the Fermi
energy in the F–D function) denotes the critical separation for breaking of the
atomic bonds and allows tuning the breaking strain as well as the cohesive
stress at bond breaking. It is further noted that
σcoh ∼ rbreak . (6.103)
The parameter Ξ (corresponding to the temperature in the F–D function)
describes the amount of smoothing at the breaking point. In addition to
defining the small-strain elastic properties (by changing the parameter k0 ),
the present model allows one to control the two most critical physical param-
eters describing hyperelasticity, (1) cohesive stress (by changing the parameter
rbreak ), and (2) the strength of softening close to the crack tip (by changing
the parameter Ξ).
Figure 6.50 depicts force vs. atomic separation of the interatomic potential
used in the study reviewed here. The upper part shows the force vs. separation
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 ):

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.
6.7.3 Computational Experiments
We carry out systematic numerical studies based on continuously varying

potential parameter. The investigations will focus on the predictions of Gao’s
model vs. those of Yoffe’s model. We begin with harmonic systems serving as
the reference and then increase the strength of the hyperelastic effect to study
the dynamics of crack tip instability in hyperelastic materials.
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)
Harmonic Potential – The Linear Elastic Reference System
We find that cracks in homogeneous materials with linear elastic properties

(harmonic potential, achieved by setting Ξ to infinity) show a critical instabil-
ity speed of about 73% of the Rayleigh wave speed, independent of the choice
of rbreak .
The crack surface morphology is shown in Fig. 6.51. This observation is in
quantitative agreement with the key predictions of Yoffe’s model.
It is observed that the occurrence of the instability can be correlated with
the development of a bimodal hoop stress as proposed by Yoffe, as is shown
in the sequence of hoop stress snapshots as a function of increasing crack
speed depicted in Fig. 6.52. We conclude that in agreement with Yoffe’s pre-
diction, the change in deformation field governs the instability dynamics in
the harmonic systems.
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
Hyperelastic Materials Behavior in Real Materials
It is observed that crack dynamics changes drastically once increasingly

stronger softening is introduced at the crack tip and linear elastic Yoffe model
fails to describe the instability dynamics.
The predictions by Yoffe’s model are included in Fig. 6.53 as the red line,
and the predictions by Gao’s model are plotted as the blue points. As seen
in the plots, it can be observed that for any choice of rbreak and Ξ, the
instability speed lies in between the prediction by Gao’s model and that by
Yoffe’s model. Whether it is closer to Gao’s model or to Yoffe’s model depends
on the choice of rbreak and Ξ. For small values of rbreak and Ξ, we find
that the instability speed depends on the cohesive stress, which is a feature
predicted by Gao’s model. The instability speed seems to be limited by the
Yoffe speed (as can be confirmed for large values of rbreak and Ξ). Whereas the
observed limiting speeds increase with rbreak , they saturate at the Yoffe speed
of 73% of Rayleigh wave speed for larger values (see Fig. 6.53, bottom curve
for Ξ = 300). In this case, the instability speed is independent of rbreak and
Ξ. This behavior is reminiscent of Yoffe’s deformation controlled instability
mechanism and suggests a change in governing mechanism for the instability
speed. Overall, the results indicate that the instability speed depends on the
strength of softening (parameter Ξ) and on the cohesive stress close to bond
breaking (parameter rbreak ).
Stiffening Materials Behavior: Stable Intersonic Mode I

Crack Propagation
A generic behavior of many rubber-like polymeric materials is that they stiffen

with strain. What happens to crack instability dynamics in such materi-
als? Recent experiments have shown intersonic mode I crack propagation in
rubber-like materials with elastic stiffening characteristics. According to the
existing theories, such high-speed crack propagation should not be possible in
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
homogeneous materials. We hypothesize that local stiffening near the crack

tip may lead to a locally enhanced Yoffe speed and enhanced energy flow, so
that the onset of crack tip instability is shifted to higher velocities.
The simulations are based on a simple model in which we change the
large-strain spring constant and small-strain spring constant, similar to stud-
ies described in the previous section except for the smoothing part near
bond snapping. The model is depicted schematically in Fig. 6.54a. Upon a
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.
6.7.4 Discussion and Conclusion
In this section we reviewed studies of the dynamics of crack tip instabilities.

We find that the onset of the instability is governed by a critical crack speed,
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.
• In stiffening materials, the instability is increased to much higher values

than the 73% mark of the Rayleigh wave speed, even reaching super-
Rayleigh velocities.
Hyperelasticity, the elasticity of large strains, can control dynamic crack
tip instabilities. This explains the discrepancy of measured instability speeds
in “real materials” [207] and predicted instability speeds by linear elastody-
namic theory [22]. The reason is that virtually all real materials show a strong
softening close to materials failure.
6.8 Suddenly Stopping Cracks: Linking Atomistic

Modeling, Theory, and Experiment
This section addresses the following question: “What happens if a crack prop-
agating at very high velocities suddenly comes to rest?” We all know that if we
try to stop a heavy object like a car, we feel considerable resistance due to its
inertia. How does a crack stop? Does the crack carry properties like “inertia”
or mass? The research in the last decades has shed light on these fundamental
questions about the nature of fracture. It was found that the stress field at
the crack immediately responds to changes in loading condition or fracture
surface energy for velocities lower than or equal to the Rayleigh wave speed.
This result led to the terminology of the crack being “mass less,” because it
responds instantaneously to a change in crack driving force. If we turn back
to the analogy of stopping a heavy object, this implies that we could stop
it instantly from any velocity without feeling any resistance, and the object
would react to any applied force immediately without delay. In this section,
we carry out large-scale atomistic simulations to focus on the atomic details
of the dynamics of suddenly stopping cracks.
6.8.1 Introduction
Large-scale atomistic simulations are used to study suddenly stopping cracks

under mode I (tensile) as well as mode II (shear) loading conditions. The crack
velocity, denoted as v, is related to the crack tip position a by v = ȧ = da/dt.
The time history of the crack tip velocity is arranged to be
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
important for constructing solutions for nonuniform crack growth [22,234,252],

and for understanding crack propagation in materials with changing resistance
to fracture.
The limiting speed of a mode I crack is the Rayleigh wave velocity,
therefore
v ≤ cR . (6.106)
In mode II, the allowed velocities are sub-Rayleigh (6.106) as well as intersonic
crack propagation speeds [165, 216, 219, 253]
c s < v ≤ cl (6.107)
There is a forbidden velocity regime
c R < v ≤ cs (6.108)
which can be overcome by a mother–daughter mechanism involving nucleation

of a secondary crack (daughter crack) at some distance away from the primary
crack (mother crack) [165,237,254]. Intersonic crack propagation has also been
reported in earthquakes since 1982 [255] and has led to active research in this
field. The discovery of intersonic crack propagation has almost doubled the
limiting crack velocity from Rayleigh to longitudinal wave speed. In the case of
nonlinear materials, the limiting velocities can be lifted to even higher speeds!
This allows for supersonic crack propagation as discussed in Sect. 6.6. The
allowed velocity regimes are depicted in Fig. 6.56.
Fig. 6.56 Allowed velocities for mode I and mode II crack propagation, linear and
nonlinear stiffening case
The stress field around a crack propagating rapidly (also referred to as

the dynamic field) is very different from the field around a static crack, as
discussed in Sect. 6.5. In mode I, the static field is expected to arise as soon
as the crack stops and is emitted with the shear wave speed. For sub-Rayleigh
mode II cracks, the situation is the same. For intersonic crack propagation in
mode II, it has been shown [208] that the static field spreads out with the
shear wave speed as soon as the mother crack has “reached” the daughter
crack tip. In any loading case, the corresponding static field is established
instantaneously on a line ahead and behind the crack propagation direction
(“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.
6.8.2 Theoretical Background of Suddenly Stopping Cracks

The suddenly stopping crack is important for studies related to nonuniform
crack growth. Solutions to this problem are often denoted as fundamental solu-
tions for crack growth. The core idea is to construct the solution of nonuniform
crack growth from the solution for uniform crack growth at constant veloc-
ity [22,234]. This becomes possible because the dynamic stress intensity factor
can be written as a product of the static stress intensity factor for given geom-
etry and a universal function which depends only on the propagation velocity.
The dynamic stress intensity factor can be expressed by [22]
KI,II (a, ȧ) = kI,II (ȧ)KI,II (a, ȧ = 0). (6.109)
The universal functions kI and kII may be approximated by [22]
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]
G(a, ȧ, loading , . . . ) = Γ (ȧ), (6.112)
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
The right-hand side can be shown to be [22, 114, 221]
g(ȧ) = 1 − ȧ/cR . (6.114)
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
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.
6.8.3 Atomistic Simulation Setup
Figure 6.57 shows the simulation geometry which consists of a 2D atomic

lattice with dimensions lx and ly . The suddenly stopping crack is modeled
by a finite length weak layer (similar as in Sect. 6.5, however, here with a
finite length to enforce that the crack stops to propagate at a critical point).
Once the crack tip reaches the end of the weak layer, it cannot propagate
any further and is forced to stop. The crack tip does not sense the existence
of the barrier before it actually reaches it because the material is elastically
homogeneous.
Fig. 6.57 Simulation geometry for the stopping crack simulation
The simulations are performed using a microcanonical N V E ensemble

(constant number of particles N , constant volume V , and constant energy
E), an appropriate choice for nonequilibrium phenomena such as dynamic
fracture. The slab is initialized with very low temperature, T ≈ 0, which
increases during the simulation to slightly higher temperatures. The loading
starts when the outermost rows are displaced according to a given strain rate.
To avoid wave emission from the boundaries, an initial velocity field according
to the prescribed strain rate is established prior to simulation. The simulations
are done with a slab size of 1512 × 3024 atom rows, and the system contains
about 4,500,000 particles.
Suddenly Stopping Mode I Crack: Simulation Setup
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.
Suddenly Stopping Mode II Crack: Simulation Setup
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
to a large class of real physical systems. We deliberately avoid the specific

complexities of a particular atomic force law. The simple interatomic force
laws can be regarded as providing model materials for computer experiments.
In this study, we use two model materials to study brittle fracture:
1. Linear elastic material with bond snapping across the weak layer
2. Hyperelastic stiffening material, also with bond snapping across the weak
layer
As in the studies reviewed above, here we use pair potentials (for instance,
harmonic potentials or modified Lennard-Jones potentials), rather than multi-
body EAM potentials [94], to model a generic brittle material.
A horizontal slit of 400 atoms distance is cut midway along the left-hand
vertical slab boundary. The crack is oriented orthogonal to the close-packed
direction of the the triangular lattice. For positions y < ystop , atomic bonds
are assumed to snap at rbreak = 1.1625 across the weak layer. The quantity
rbreak can be used to control the fracture surface energy distribution. This con-
fines the crack to propagate along prescribed weak layer without branching.
With this approach, we deliberately try to suppress branching and dislocation
emission in the current work by introducing a weak interface.
For the linear spring potential, the value for k is assumed to be k = 140/r02 .
Wave velocities are cs ≈ 7.10, cl ≈ 12.29, and cR ≈ 6.55 [258]. The nonlinear
tethered LJ potential is described in detail in Sect. 4.4.3, and we choose 0 =
1.9444 to match the small-strain elastic properties with the elastic properties
of the harmonic potential associated with the spring constant k = 140/r02 .
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)
The energy release rate G can be universally expressed as
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.
Crack Initiation Time in an Infinite Solid

with a Semi-Infinite Crack
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
σ̇12 = Gε̇12 . (6.120)
The stress intensity factor KI can be determined as [22]

√
4 2 − 2ν(1 + ν)cs t3
KI (t) = σ̇11 (6.121)
3 π
and KII is found to be [22]

4 2(1 + ν)cs t3
KII (t) = σ̇12 . (6.122)
3 π
Equations (6.121), (6.122), and (6.116) can be used to derive an expression
for the initiation time of crack motion:
2
3
tinit = 3
9πµγ
pred
43 ' . (6.123)
2 + 1−ν 2
8cs σ̇12 2 σ̇11
In the case of pure mode I cracks, the shear rate σ̇12 is set to zero in (6.123).
Crack Initiation Time Predictions
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
Theory prediction Simulation result
Fracture loading mode Linear tpred

init Nonlinear tpred
init Linear tinit Nonlinear tinit
I 41.51 45.19 42.12 46.26

II 28.26 30.76 32.04 35.82
Table 6.4 Griffith analysis of the atomistic models, for mode I and mode II cracks,
and different potentials. The predicted values based on continuum calculations agree
well with the molecular dynamics simulation results
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.
6.8.4 Atomistic Simulation Results of a Suddenly Stopping

Mode I Crack
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
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
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.
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
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)
These results show good agreement among atomistic simulations, contin-

uum theories and experiments. The atomistic simulation demonstrated that
the rapid thermalization of elastic waves near the crack tip did not change
the basic nature of crack tip stress fields predicted by continuum mechanics.
This result would not have been possible by continuum mechanics alone.
Figure 6.63 displays the experimental results that have been used for com-
parison with the MD simulations. The data obtained from gage 1 (closest to
the crack tip) have been used for comparison with MD results, as shown in
Fig. 6.62.
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
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
be solved in closed form. We present simulations to address the following

questions:
• Does the result agree qualitatively and quantitatively with the linear
solution?
• What is the “wave” speed in the nonlinear case, that is, how fast can the
static field be established?
We start with a simple Griffith analysis to calculate the time for the onset
of fracture due to the applied loading. Crack initiation time is found to be
tinit ≈ 46.26. The initiation time agrees well with the prediction tpred
init = 45.19.
The loading is kept up to tl = 144. The maximum strain we achieve is εxx =
0.0144. The limiting speed observed is v ≈ 7.5. As soon as loading is stopped,
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
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
Fig. 6.65 Evolution of principal maximum stress and potential energy along the
prospective crack line; for a mode I nonlinear crack
soon after crack arrest, and subsequent thermalization suppresses additional

wave emission.
The maximum principal stress field is shown in Fig. 6.66 for various
instants in time. The discontinuities are smeared out, and it becomes evident
that the definition of a unique wave front can be difficult. In the nonlinear
case the shape of the wave fronts is different (elliptical vs. circular shape).
The normalized maximum principal stress over time, recorded at a con-
stant distance ahead of the crack tip, is plotted in Fig. 6.67 for an anharmonic
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
Discussion – Mode I
One important observation in the simulation is that a large number of waves

are generated after crack stopping. This effect is less significant in the prospec-
tive propagation direction and becomes more pronounced in other directions.
The waves attenuate quickly after the crack is stopped, and in the long-time
limit become dissipated as heat (“thermalizing”). In the nonlinear case, we
summarize the following findings:
1. There is no unique wave velocity, and the static field does not spread out
behind the shear wave front. We find that there exists a train of “longitu-
dinal” and “shear” waves associated with the rapidly changing stress state
near the crack tip. The static field is not established until all waves have
passed.
2. There is an anisotropy effect. Ahead of the crack, the wave speed appro-
aches the linear limit, and orthogonal to the crack, the wave speed is
significantly larger. We observe slightly elliptical wave fronts instead of
circular wave fronts.
6.8.5 Atomistic Simulation Results of a Suddenly Stopping

Mode II Crack
We consider crack propagation under in-plane shear dominated loading, start-

ing with the linear case and moving subsequently to the nonlinear case. For
all mode II simulations, a mother–daughter mechanism to overcome the for-
bidden velocity zone is observed. This mechanisms is assumed to be governed
by a Burridge–Andrew mechanism [254, 259]. A peak shear stress ahead of
the crack continuously increases as the mother crack propagates through the
material. Once this peak of shear stress reaches the cohesive strength of the
interface, the daughter crack nucleates at some distance ahead of the mother
crack and starts to propagate at an intersonic speed. The observation regard-
ing the mother–daughter mechanism and the limiting speed of cl is consistent
with the discussion in [165, 237].
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.
Fig. 6.68 Schematic of waves emitted at a suddenly stopping mode II crack; (a)
stopping of daughter crack, (b) stopping of mother crack
The loading is stopped at tl ≈ 84 soon after the daughter crack is nucleated.

The mother crack hits the stopped daughter crack at t = 105. The mother
crack propagates at v ≈ 6.5, and the daughter crack quickly attains a velocity
v ≈ 12.3. The suddenly stopping intersonic crack shows the following sequence
of events. The daughter crack is stopped, and the mother crack continues until
it reaches the end of the weak layer. For each crack stopping event, two wave
fronts are emitted yielding a total number of four wave fronts. The mecha-
nism of the suddenly stopping intersonic crack is visualized schematically in
Fig. 6.68a, b [208].
The stresses continuously change after the daughter crack is stopped [208].
Once the mother crack hits the daughter crack, stresses begin to increase dra-
matically. The static field radiates out from the crack tip with a velocity
v ≈ 7.4. This velocity is the shear wave velocity and the observation provides
good agreement to the prediction by continuum theory. Other propagation
velocities measured from the data also agree with the continuum mechan-
ics predictions. In Fig. 6.69a, the maximum principal stress is shown, and in
Fig. 6.69b, the potential energy field is depicted some distance ahead of the
crack tip. The potential energy field is shown in Fig. 6.70 at several instants
in time.
Figure 6.71 shows the normalized maximum principal stress at a fixed
measurement location some distance ahead of the crack. When comparing
this quantity to the stress intensity factor, care must be taken because the
singularity changes continuously as the crack passes through the distinct inter-
sonic velocity phases. The arrival of the shear wave front is characterized by a
strong discontinuity induced by the Rayleigh wave [165, 208, 217]. As soon as
the shear wave reaches the measurement location, the static field is established
and the stress no longer changes afterwards.
The simulation results are consistent with the analysis in [208].
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.
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
The time to crack initiation is determined to be tinit = 35.82, which is

somewhat larger than the prediction tpred
init = 30.76. The loading is stopped
at tl = 129.6, after nucleation of the daughter crack at t ≈ 112. This leads
to a far-field strain of εxx = 0.0039 and εxy = 0.052. The crack propagation
speeds we measure are v ≈ 8 for the mother crack, and v ≈ 16.8 for the
daughter crack. Both velocities are higher than the corresponding velocities
of the linear case (22% higher, and 36% higher for mother and daughter crack,
respectively) due to material nonlinearities. The daughter crack propagates
supersonically through the material. Figure 6.72 plots the crack extension
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
history a(t). In this figure, the mother–daughter mechanism can be identified

straightforwardly.
The mother crack hits the daughter crack at t ≈ 197, δt ≈ 77 after the
nucleation of the daughter crack. This can also be estimated from Fig. 6.72.
Like in the simulations of a mode I crack in nonlinear material, the wave
fronts are smeared out, but we can still identify a localized group of nonlinear
waves which may be interpreted as a “nonlinear wave front.” The first waves
emitted after stopping of the crack can be regarded as the longitudinal wave
moving with a velocity v ≈ 16.4 through the solid in the prospective crack line.
The strong spike ahead of the crack corresponds to the shear wave induced
by the stopped daughter crack and is propagating with a velocity v ≈ 10.5.
Both values are higher than the corresponding linear shear wave velocity.
The propagation speed of the longitudinal wave is close to the velocity of the
daughter crack before stopping. The stopping mother crack leads to nucleation
of additional waves. A discontinuity propagating at v ≈ 10.5 is observed,
which is emitted when the mother crack is stopped. We associate this with
the shear wave front. It is difficult to observe the longitudinal wave front from
the molecular dynamics data in this case.
Plots of the maximum principal stress field and the potential energy field
ahead of the crack tip in the prospective crack line are shown in Figs. 6.73a, b.
The plots look quite different from the harmonic case. The shape we have
observed in the harmonic counterpart, in particular the strong discontinuity
before arrival of shear wave front seems to have disappeared. The discontinuity
associated with the shear wave front of the daughter crack is more pronounced
than in the linear case. The fact that the discontinuities are smeared out can
lead to stresses changing significantly even after the bulk of the shear waves
of the mother crack has passed. This effect is reminiscent of the phenomenon
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
appears in different simulations when the impact of nonlinearities is stronger

due to higher lateral strains (see further discussion below).
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
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
We start with the results for the linear case:

1. It was shown that the static field does not establish until the mother crack
reaches the stopped daughter crack for the linear reference system.
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
We have studied suddenly stopping cracks by atomistic simulations. We

considered a plane-stress elastic solid consisting of a two-dimensional, trian-
gular atomic lattice. For the interatomic interactions, we assumed a tethered
Lennard-Jones potential, as well as a harmonic potential. We presented four
simulations for mode I and mode II loading conditions, and linear and non-
linear simulation potentials. In addition to the atomistic simulations, we have
done continuum analyses to determine the crack initiation time and wave
velocities associated with the interatomic potentials.
The harmonic atomistic simulations have shown good agreement with
the continuum mechanics and experimental results. We summarize the main
results below.
1. In mode I, the static solution spreads out as soon as the crack is stopped.
This is in agreement to the continuum theory [234].
2. In mode I, experimental results and atomistic studies show similar results
on very different scales [23] (see Fig. 6.62).
3. In mode II, the static solution spreads out as soon as the mother crack has
reached the stopped daughter crack. The nature of the mode II intersonic
crack is very different from the sub-Rayleigh crack [208].
4. In both mode I and mode II, we observe emission of waves from the stopped
crack. These waves attenuate quickly, in which process energy is dissipated
as heat. This does not change the basic nature of stress fields near the crack
tip as predicted by continuum mechanics.
The anharmonic simulations give somewhat different results, which cannot be
explained by linear elastic fracture mechanics only. For the nonlinear case, we
would like to summarize the main findings as follows.
1. In both mode I and mode II, the wave fronts and discontinuities are not as
sharp as in the linear case. This nonlinear dispersion effect could cause the
field to change continuously even after the main wave discontinuities have
passed.
2. For supersonic mode II cracks, the mother crack could transform from a
stress singularity into a less localized stress distribution. The static field
is not instantaneously established but stresses increase continuously, until

they eventually reach steady state.
3. The nonlinearity can lead to an anisotropy effect of wave propagation.
Instead of circular wave fronts, we find elliptical wave fronts.
The simulations show that the linear model is a reasonable approximation even
when moderate nonlinearities are present. The simulations provide evidence
that the crack behaves like a massless particle only in the sub-Rayleigh regime.
The Griffith criterion works well in all simulations we presented.
6.9 Crack Propagation Along Interfaces

of Dissimilar Materials
In this section, we review studies of cracks propagating along interfaces
between two dissimilar materials, as schematically shown in Fig. 6.76. Cracks
at interfaces are technologically important, since the bonding between two dis-
similar materials as for instance between epoxy and aluminum is usually weak
and serves as a potential failure initiation point of a structure. The atomistic
model featuring the weak layer could be regarded as an idealization of such
cases. Another important field where interfaces between dissimilar materials
play an important role is the dynamics of earthquakes.
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)
Several theoretical studies were carried out on cracks in dissimilar media

[260–262]. Most of the early investigations focused on static cracks. One of
the interesting features of the elastic interfacial crack problem is the charac-
teristic oscillating stress singularity that was determined by Williams [263].
This theoretical finding is incompatible with real materials since the crack
faces would penetrate each other at the crack tip. The stress intensity factor
is complex-valued for interfacial cracks and it is generally difficult to define a
crack nucleation criterion based on the Griffith condition [260].
In recent years, the dynamics of cracks along dissimilar interfaces was
increasingly researched. For instance, the asymptotic stress field near dynamic
cracks at bimaterial interfaces was studied [264, 265]. The analysis discussed
in [264] assumed steady-state crack propagation and provided the spatial

structure of square-root singular stress field very close to the dynamic crack
tip. The analysis led to definition of a complex dynamic stress intensity fac-
tor. Later, this analysis was refined relaxing the steady-state assumption and
including higher order terms [265].
There are also experimental results available on interfacial cracks, as
for instance studies reported in [216, 266, 267] of cracks propagating along
interfaces of PMMA and metals. In [266], the researchers focused on the devel-
opment of a crack growth criterion along interfaces. They also compared the
experimental results with theoretical predictions of the stress field near the
crack tip. Crack speeds that exceeded that of the shear wave speed of the soft
PMMA material were observed.
Rather few molecular dynamics analyses of dynamic fracture along bimate-
rial interfaces have been reported. One example is recent molecular dynamics
simulations of mode II cracks along a weakly bonded interface of harmonic–
anharmonic materials (material defined by a harmonic potential neighboring
a material defined by a tethered LJ potential) [219].
In summary, for cracks at interfaces, existing theory and experiment
predicts that
• The limiting speed of mode I cracks at bimaterial interfaces can exceed the
Rayleigh wave speed of the soft material [265, 268]. However, intersonic or
supersonic crack propagation with respect to the soft material layer is not
predicted by theory and has not been observed in experiment.
• The limiting speed of mode II cracks is given by the longitudinal wave
speed of the stiff material and the crack speed can thus be truly supersonic
with respect to the soft material [267].
The most important research objective of the following studies is the lim-
iting speed of cracks: The fact that the wave speed changes discontinuously
across the interface makes it difficult to define a unique wave speed near the
interface, and thus difficult to predict the limiting speed of the crack. Using
molecular dynamics, can we determine what is the limiting speed of a crack
along dissimilar materials?
For mode II cracks in an earlier study a mother–daughter–granddaughter
mechanism was observed through which the crack finally approached a velocity
faster than the longitudinal wave speed of one of the layers [219]. In this setup,
however, one of the half spaces was modeled by harmonic interactions, and the
other was modeled by a tethered LJ potential. Although this setup constitutes
an interface of different materials, the wave speeds associated with each half
space could not be clearly defined since one of the material was hyperelastic.
To obtain a more clean model of cracks at interfaces, we propose to study two
half spaces with harmonic interatomic potentials, but with different spring
constants k0 < k1 . The ratio
k1
Ξ= (6.124)
k0
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
measures the elastic mismatch

√ of the two materials, and the wave speeds are
thus different by a factor Ξ.
The plan of this section is as follows. We start with simulations of mode I
cracks along interfaces and show that under sufficiently large loading, the crack
approaches the Rayleigh wave speed of the stiffer of the two materials via a
mother–daughter mechanism. We continue with a study of mode II cracks
along interfaces. It will be shown that a mother–daughter–granddaughter
mechanism, in agreement with previous analyses [219], exists and allows the
crack to approach the longitudinal wave speed of the stiffer of the two mate-
rials. We finally discuss the simulation results and compare the elastic fields
of the mode I crack with the solution of continuum mechanics results and the
results of atomistic modeling as reported earlier.
6.9.1 Mode I Dominated Cracks at Bimaterial Interfaces
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
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,
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
the crack propagates at a super-Rayleigh speed through the material. Since

crack motion is subsonic, no shock front is established.
Figure 6.80 shows the potential energy field for a crack after the sec-
ondary crack is nucleated and crack motion of the daughter crack is supersonic.
Figs. 6.81 and 6.82 show the stress field and the particle velocity field. The
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
The mother–daughter mechanism in mode I cracks at interfaces is also

observed for elastic mismatch Ξ = 2, 5, 7, and 10 (note that not in all cases
crack motion is supersonic with respect to the soft material since the Rayleigh
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))
The secondary crack is nucleated approximately at a distance 11 atomic

distances ahead of the mother crack and rapidly propagates at Mach 1.7 with
respect to the soft material. If the strain rate is too low, the crack moves at
a sub-Rayleigh speed (soft material) until the solid has separated, without
nucleation of secondary cracking. In Fig. 6.83, we plot the shear stress field
σxy for several instants in time during nucleation of the secondary crack (note
we chose quite small time intervals between the snapshots). Atoms with the
energy of a free surface are colored blue and highlighted by larger spheres.
The plot suggests that a peak of shear stress ahead of the crack tip may have
caused breaking of atomic bonds. After the secondary crack is nucleated,
it coalesces with the mother crack and moves supersonically with respect
to the soft material. The mechanism of nucleation of a secondary crack is
reminiscent of the mother–daughter mechanism predicted by the Burridge–
Andrew mechanism and observed in studies of intersonic mode II cracks in
homogeneous solids (see discussion in previous sections). It is noted that the
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.
6.9.2 Mode II Cracks at Bimaterial Interfaces
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
For the studies of mode II cracks along interfaces we choose Ξ = 3. The

crack tip history is depicted in Fig. 6.84. The loading rates are ε̇xx = 0.000 03
for slight mode I opening loading and for the shear loading ε̇xy = 0.000 125.
Fig. 6.85 shows the crack tip velocity history during the mother–daughter–
granddaughter mechanism. The plot is obtained by numerical differentiation
of the crack tip history shown in Fig. 6.84. The crack speed changes abruptly
at the nucleation of the daughter crack, and rather continuously as the
granddaughter crack is nucleated.
Initially, the (mother) crack propagates close to the Rayleigh velocity of the
soft slab part (v ≈ 4.8). After a secondary daughter crack is born that travels
at the longitudinal wave speed of the soft material, a granddaughter crack is
Fig. 6.85 Crack tip velocity history during the mother–daughter–granddaughter

mechanism, for elastic mismatch Ξ = 3. The plot is obtained by numerical dif-
ferentiation of the crack tip history shown in Fig. 6.84. The crack speed changes
abruptly at the nucleation of the daughter crack, and rather continuously as the
granddaughter crack is nucleated. Characteristic wave speeds for the stiff and soft
solid are indicated in the plot
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,
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
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
studies, here carried out for mode II dominated cracks at Al–Homalite

interface. The Mach cone are clearly visible in the soft material.
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
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
mechanism [219]. This agrees with results of earlier computer simula-

tions. The results also confirm theoretical [265] as well as experimental
results [267].
• The elastic fields in mode I and mode II cracks are very different from
homogeneous materials. If crack propagation is supersonic with respect
to one of the half spaces, multiple shock fronts are observed as shown in
Fig. 6.88. If the elastic mismatch is small or nucleation of daughter cracks
is suppressed in mode I cracks, the elastic fields in the left and right half are
asymmetric. The asymmetric shape of the asymptotic deformation fields
matches the predictions by continuum mechanics theories [265].
Atomistic simulations are a feasible approach to study the dynamics

of cracks at interfaces. Future investigations could focus on the compari-
son of the asymptotic stress field in simulation and theory, as well as on
a more detailed and theoretical analysis of the observed mother–daughter
and mother–daughter–granddaughter mechanisms, particularly focusing on
the nucleation process of secondary cracks.
6.10 Dynamic Fracture Under Mode III Loading

In this section, we study three-dimensional models of mode III cracks. A
schematic of the mode III antiplane shear crack loading is shown in Fig. 6.6.
The study of mode III cracks is motivated by the fact that in mode III,
there exists only one wave speed associated with crack dynamics, the shear
wave speed cs . This simplifies the theoretical continuum mechanics analysis
of the crack dynamics. Recently, a closed form solution for the crack speed of
a crack propagating in a stiff material layer embedded in a soft matrix was
derived [24]. The analysis revealed that the same concept of a characteristic
energy length scale χ also holds for mode III cracks. The most important
objective of this chapter is therefore to validate this finding using atomistic
simulations similar to those presented in Sect. 6.6.5.
Mode III cracks have rarely been studied with molecular dynamics meth-
ods before. Some simulations were reported in the literature, but these focused
on cracks in ductile materials under mode III loading [270, 271].
According to classical linear elastic theories [22], for mode III cracks all
velocities below the shear wave speed are admitted, thus
v ≤ cs . (6.125)
The allowed crack propagation speeds for mode III cracks in linear and non-
linear solids are shown in Fig. 6.90. Similar to the results of mode I and mode
II cracks where cracks can move faster than the wave speeds in the material,
mode III cracks can also move faster than the shear wave speed and thus
become supersonic once the material stiffens with strain. This phenomenon
was verified using a tethered LJ potential (results not shown here).
Fig. 6.90 Allowed velocities for mode III crack propagation, linear and nonlinear
case
The objectives of the studies in this section are summarized as follows.

First, we verify that the limiting speed associated with a crack propagating
in a harmonic lattice agrees with the theoretical prediction. We then dis-

cuss simulation results of crack motion in a thin stiff layer embedded in a
soft matrix, also yielding supersonic crack motion (similarly to the Broberg
problem discussed in Chap. 6.6). The recently derived analytical continuum
mechanics solution of the problem is quantitatively compared with the molec-
ular dynamics results [24]. We find that the energy length scale described in
Chap. 6.6 also applies to mode III cracks.
6.10.1 Atomistic Modeling of Mode III Cracks

Previous studies have provided evidence that 3D molecular dynamics is a good
framework to investigate the dynamics of fracture. For instance, Abraham and
coworkers [156] studied dynamic fracture in a three-dimensional solid with
LJ interactions. They showed that unlike in two dimensions where the LJ
potential yields a very brittle solid (see Fig. 6.48), in three dimensions the
LJ potentials leads to a very ductile solid [138]. The researchers studied the
dynamics of fracture in different crystal orientation and provided a Schmidt
factor analysis [272]. Later, a three-dimensional model using harmonic inter-
actions in the bulk, and using the concept of a weak fracture path was adopted
in simulations of dynamic crack propagation [219]. This model corresponds to
a perfectly brittle system which allows to study the dynamics of fracture in a
clean environment.
Here we adopt a similar approach and confine crack motion along a weak
layer, which is characterized by a fracture surface energy much smaller than
in the bulk. This confined fracture path helps to avoid crack branching and
allows to purely focus on the dynamics of cracks. In previous studies, a weak
LJ layer was used to model the weak fracture layer [219]. Here we assume
a homogeneous material with harmonic interactions. The interactions are
defined according to (4.4.3) in the bulk, and according to (4.43) across the
weak fracture layer.
The slab is initialized at zero temperature and loaded according to mode
III, and we also give a slight mode I loading. The loading rates are ε̇xx =
0.000 1 for mode I and the (engineering) shear rate γ̇xz = 0.000 2. The loading
is kept up during a loading time tl , and then the boundaries are held fixed.
6.10.2 Mode III Cracks in a Harmonic Lattice – The Reference

Systems
The results show that the limiting speed of mode III cracks in a harmonic
lattice is given by the shear wave speed. This was verified for two choices
of the spring constant k1 ≈ 57.32 and k1 = k0 /2. This observation is in
agreement with the predictions by continuum mechanics theories [22]. The
crack tip history for the soft and stiff reference system is shown in Fig. 6.91.
In both systems, the loading is stopped at tl = 135. Both soft and stiff systems
init ≈
approach the theoretical limiting speeds. Fracture initiation times are tsoft
47 for the soft system and tstiff
init ≈ 41 for stiff system.
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
6.10.3 Mode III Crack Propagation in a Thin Stiff Layer

Embedded in a Soft Matrix
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
v = f˜(c0 , c1 , h/χ) (6.127)
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.
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].
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
6.10.4 Suddenly Stopping Mode III Crack
In Chap. 6.8, we discussed suddenly stopping mode I and mode II cracks

in linear and nonlinear materials. We have conducted similar studies for a
suddenly stopping mode III crack. Theory predicts that the dynamics of the
suddenly stopping mode III crack is very similar to the mode I crack [22]. An
important difference of the suddenly stopping mode III crack to the mode I
case is that the static field spreads out in the whole area around the crack
tip, and not only in the line ahead of the crack tip as in mode I [22].
Figure 6.94 shows the potential energy field close to a suddenly stopping
mode III crack. The simulation technique is the same as described in Chap. 6.8
with the only difference that a three-dimensional model is used.
The result is very reminiscent of the mode I simulation results discussed
in Chap. 6.8. The static field spreads out with the shear wave speed as soon
as the crack is stopped. In snapshot “1” of Fig. 6.94, the crack propagates at
a velocity close to the shear wave speed prior to stopping. Behind the crack,
the static field is transported with the Rayleigh wave speed. The Rayleigh
surface wave can clearly be observed in Fig. 6.94, snapshots “3” and “4.” The
static stress field spreads out in the whole area around the crack tip, and not
only in the line ahead of the crack tip as in mode I [22].
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,
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 Brittle Fracture of Chemically Complex Materials

In the previous sections we have reviewed a series of studies, primarily
using a model potential approach. These studies provided a comparison
between atomistic approaches and continuum theory, and led insight into
the significance of the large-deformation interatomic bonding properties or
hyperelasticity for the dynamics of cracks in brittle materials. Here we extend
these studies to materials in which the atomic bonds forming the lattice are
more complex, focusing on a case study of a brittle material whose fracture

mechanics has received considerable attention, silicon.
We will review a study of dynamic fracture in a silicon single crystal in
which the ReaxFF reactive force field is used for several thousand atoms
near the crack tip, while more than 100,000 atoms of the model system are
described with a computationally less expensive Tersoff force field. The hybrid
simulation method introduced in Sect. 5.4.5 is used for the simulation studies
reviewed here [274, 275].
Since the ReaxFF force field is completely derived from quantum mechan-
ical calculations of simple silicon systems without any empirical parameters,
this model provides a fundamental, quantum mechanics based description of
fracture of silicon. It will be shown that the results reproduce experimental
observations of fracture in silicon including differences in crack dynamics for
loading in the [110] or [100] orientations, as well as dynamical instabilities with
increasing crack velocity. Further, a correlation of the atomistic simulation
results with single silicon crystal fracture experiments will be presented. This
study reveals that after the critical fracture load is reached, the crack speed
instantaneously jumps from zero to approximately 2 km s−1 . The simulation
results provide a mechanistic explanation for these observations, illustrating
that chemical rearrangement effects of the atomic lattice control these phe-
nomena. This shows that for many materials, the properties of an ensemble of
chemical bonds under large load can control the fracture behavior, and that
a simple Griffith condition of crack initiation is insufficient to capture these
effects. The experimental validation of simulation results is a crucial aspect
in the studies of complex fracture phenomena.
6.11.1 Introduction
The basic notion of brittle fracture is that continuous breaking of atomic

bonds lead to the formation of two new materials surfaces. Is this simple pic-
ture also true for chemically complex materials like silicon? The atomistic
models of fracture discussed earlier in this book assume an empirical relation-
ship between bond stretch and force, typically described using a simple pair
potential. However, breaking of bonds in real materials can be an extremely
complicated process. For instance, it can be captured with sufficient accu-
racy by using quantum mechanical (QM) methods. However, such methods
are limited to approximately 100 atoms, which limits the applicability of these
methods to model dynamic fracture, since here thousands of atoms participate
in the bond breaking mechanisms.
Fracture of silicon has received tremendous attention due to its com-
plexity of bond breaking and due to interesting failure dynamics observed
experimentally [25, 222–224, 276–278]. These experimental efforts led to criti-
cal insight into deformation modes, such as the mirror-mist-hackle transition
and orientation dependence of crack dynamics in silicon single crystals [25].
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.
Here we focus on modeling fracture in pure silicon and the interactions of

silicon with oxygen competing with crack extension.
6.11.2 Hybrid Atomistic Modeling of Cracking in Silicon:

Mixed Hamiltonian Gormulation
We employ the hybrid multiparadigm approach as described in Sect. 5.4.5. As

described in [131], the ReaxFF potential for silicon has been tested against
quantum mechanical data for a wide range of processes [283], including Si–
Si bond breaking in H3 Si–SiH3 and Si=Si bond breaking in H2 Si=SiH2 ,
equations of state for 4-coordinate silicon (diamond-configurations) and 6-
coordinate silicon phases (β-tin), and simple cubic crystal. This force field is
also capable of treating interactions of Si with O and H [131]. It is emphasized
that all parameters are completely derived from QM calculations.
The simple approach described in Sect. 5.4.5 is used to describe the tran-
sition region between two paradigms, using two parameters, Rtrans for the
size of the transition region, and Rbuf for the size of the ghost atom region.
A schematic of this approach is shown in Fig. 5.8. In the examples discussed
here, we model a reactive region Ωrx embedded into a nonreactive domain
Ωnr . The reactive region is updated every Nu = 10 steps during the inte-
gration. We choose Rtrans = 6 Å and Rbuf = 5 Å. These parameters have
been chosen by trial and error to make sure that the crack dynamics is not
affected by changing these parameters. The shape and size of the reactive
region surrounding the moving crack is based on the strain energy density of
each atom. All atoms with a strain energy larger than Ecrit = −3.5 eV are
embedded in a cylindrical reactive region of R = 10 Å. The critical strain fil-
ters atoms at the tip of cracks and atoms in the vicinity of cracks whose bonds
are stretched significantly. The union of all cylindrical regions yields the total
reactive region, allowing representation of arbitrary shapes. The final reactive
region is typically not circular. Initially, the systems contain several hundred
reactive atoms, which corresponds to a small reactive region at the crack tip.
This initial size of the reactive region may increase during the simulations
approaching several thousand atoms because of crack branching or due to
formation of microcracks.
When oxygen atoms are present in the system, a similar procedure is
applied and in addition to the criterion based on the strain energy density,
each oxygen atom is embedded in a cylindrical reactive region of R = 10 Å.
6.11.3 Atomistic Model
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).
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.
6.11.4 Simulation Results
First we compare the dynamics of crack propagation described using pure

Tersoff [10] to the results incorporating a region treated by ReaxFF close to
the crack tip to demonstrate the importance of QM level accuracy at the crack
tip to describe the bond breaking process. The crack direction is [110] with a
(110) crack surface. We strain the system by 20% in mode I loading and then
minimize the potential energy. We consider two cases, one in which a small
region around the crack tip is treated using ReaxFF embedded in a Tersoff
region, and the other case in which all atoms are described using Tersoff.
The results of these computational experiments are shown in Fig. 6.96. It is
observed that with pure Tersoff, the crack does not propagate. Instead the
crack becomes blunt (Fig. 6.96a) and eventually amorphizes as the loading is
increased sufficiently. This incorrect description of brittle fracture has been
observed in previous studies with Tersoff-type potentials [224, 279], and is
in sharp contrast to experimental results [25]. Experiment clearly suggest a
highly brittle behavior of silicon, in particular when initial cracks have (110)
faces. In agreement with experiment, the hybrid treatment leads directly to
a correct description of the fracture process, showing a dominance of brittle
fracture (Fig. 6.96b) [25,222,224,276]. For the [100] crack direction with (100)
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
Figure 6.97 shows various snapshots of a crack propagating in a sili-

con crystal strained by 10% with the temperature controlled to be around
T ≈ 300 K. The crack propagates through the material in a perfectly brittle
manner. The crack approaches a speed of 3.41 km s−1 , which is approxi-
mately 75% of the Rayleigh wave speed, the limiting speed predicted by
continuum theory [22] (cR ≈ 4.5 km s−1 [284]). The observations from the
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
Fig. 6.99 Crack speed as a function of load, for the (110) system (subplot (a), and
the (111) system (subplot (b))
features a minimum crack speed, in agreement with the molecular dynamics

simulations.
Figure 6.101 shows the dynamic fracture toughness as a function of the
energy release rate (Fig. 6.101a) and the average crack velocity (Fig. 6.101b).
This experimental result agrees nicely with the molecular dynamics simula-
tions of the 1D fracture model, as shown in Fig. 6.15.
6.11.5 Dynamical Fracture Mechanisms
To investigate the atomistic details of the fracture dynamics in silicon, we

analyze a computational experiment with a slowly increasing tensile mode I
load, for the (111) oriented crystal (at a strain rate of 0.0005% strain increment
per integration step). Figure 6.102 depicts a series of snapshots of fracture
mechanics in silicon, for a crack oriented in the (111) plane. The initial, static
regime is followed by a short period of crack growth during which a perfectly
flat, mirror-like surface is generated. Crack propagation becomes increasingly
erratic until the entire crystal is fractured.
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
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
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)
and experimental results. This behavior has been confirmed in simulations of

various pulling rates.
It is believed that the threshold crack speed is related to formation of the
5–7 double ring. Two observations support this hypothesis. First, there exists
a geometric effect due to crack blunting, effectively leading to a reduction of
the stress concentration at the tip of the crack (compare Fig. 6.104a, b). In this
spirit, the 5–7 defect corresponds to the fracture process zone; since it is only
a few atomic distances wide and much smaller than the specimen dimension,
the small-scale yielding condition is satisfied.
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 ,
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
cR (1 − G0 /G0,5−7 ) > 0, (6.130)
while
G0 /G0,5−7 < 1. (6.131)
Thus the admissible crack speeds v are
cR (1 − G0 /G0,5−7 ) ≤ v < cR . (6.132)
This is in contrast to the prediction by classical LEFM that predicts that
0 ≤ v < cR (6.133)
under identical boundary conditions. Figure 6.105 shows a quantitative com-

parison of the LEFM model, the modified LEFM model and molecular
dynamics simulation. The predictions of theoretical model and the compu-
tational results are in good agreement. The modified LEFM model predicts
larger crack speeds for larger crack driving forces. A possible reason for this
disagreement could be elastic softening at large strains that lead to a reduction
of cR .
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.
Finally, it is noted that the mode of loading can also strongly influence the
crack direction [26]. Figure 6.106 depicts a comparison of the crack dynamics
under mode I (Fig. 6.106a) and mode II (Fig. 6.106b) loading. Under mode I
loading, the crack begins to move straight after the critical load is reached.
Under mode II loading, the crack begins to move at an angle of approximately
30◦ to the left.
6.11.6 Reactive Chemical Processes and Fracture Initiation
The potential inherent in the hybrid scheme becomes particularly apparent

when reactive chemical processes such as oxidation are allowed to compete
with brittle crack extension. Figure 6.107 depicts a series of studies in which we
investigate crack dynamics in a small penny-shaped crack containing oxygen
molecules O2 . As previously, the slab is under mode I tensile loading. The
oxygen atoms react strongly with the silicon surface (see Fig. 6.107b, d).
Fig. 6.106 Comparison of the crack dynamics under mode I (subplot (a)) and
mode II loading (subplot (b)) [26]
It is noticed that 5% prestrain is sufficient to nucleate a crack for pure

silicon, but when oxidative processes are present such a strain is not sufficient
to nucleate a crack. The oxidation leads to crack blunting effectively reducing
the stress intensity factor at the crack tip, while leading to formation of strong
Si–O bonds, making this Si–O layer harder to break than pure silicon (see
Fig. 6.107a vs. Fig. 6.107b).
6.11.7 Summary
The hybrid ReaxFF–Tersoff model reproduces experimental results quantita-

tively, with modest computational cost, thus providing some advantage over
other more expensive computational models. The ReaxFF reactive force field
was developed solely from first-principles quantum mechanical calculations of
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.
In this section we reviewed the application of a hybrid numerical method

integrating the ReaxFF and Tersoff force fields to allow a physics-based
description of the fracture mechanics and fracture dynamics of silicon. The
scheme represents a new multiscale approach of coupling the QM scale of
chemistry and bond breaking and formation with the scale of mechanics of
materials. Unlike other attempts in which empirical potentials were modified
heuristically to yield brittle fracture of silicon, our method is completely based
on first-principles, with no empirical parameters used for fitting of ReaxFF
for silicon.
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]
The results (see Fig. 6.96a, b) underline the importance of large-strain

properties at bond breaking for the dynamics of fracture, as suggested earlier
by Gao [27] and in [28, 29]. Figure 6.108 shows the large deviation between
Tersoff and ReaxFF at large strains close to bond breaking by stretching the
crystal in the [110] direction. The Tersoff description leads to a sharp rise of
the force close to rupture of bonds, which deviates significantly from the more
accurate solution by ReaxFF.
The model is further capable of reproducing key experimental observations
including crack limiting speed, crack instabilities, and directional dependence
on crystal orientation (see Fig. 6.97a–c). The studies lead to insight into the
atomistic details of the fracture processes, suggesting continuous formation of
microcracks ahead of a propagating mother crack (see Fig. 6.97b). This has
been debated in the literature and the simulations seem to corroborate this
concept.
We find that ReaxFF reactive force field [93, 131] successfully mod-
els chemomechanical properties of materials as crack propagation. Since
ReaxFF is capable of describing a wide heterogeneous range of materials

[93,128–131,282], our approach may be a practical means to studying the cou-
pling of complex chemical reactions to mechanical properties. This is shown
in the studies depicted in Fig. 6.107. Here we have demonstrated the effect
of oxygen in changing the fracture behavior of silicon by effectively blunting
the crack tip and thus making silicon more brittle, in agreement with exper-
iment. Our hybrid method could enable studies of stress corrosion processes
and other degradation and aging mechanisms.
A few thousand atoms in the reactive region at the crack tip are suffi-
cient to describe the crack dynamics correctly. Although such calculations are
not practical with pure QM, ReaxFF provides QM accuracy for the reacting
part of the system, while retaining speeds comparable to that of simple force
fields. As demonstrated in the literature and by the simulations reviewed here,
the Tersoff potential cannot describe fracture accurately [114, 279], but it is
adequate for describing the elastic regions.
6.12 Summary: Brittle Fracture

Brittle fracture is an important material phenomenon with great scientific
and technological impact. Atomistic simulations have been particularly help-
ful in understanding the science of fracture, as numerous contributions by
many researchers over the past decades illustrated. In the previous sections
we have only reviewed a small part of the literature. Particular emphasis of
these studies was on the joint investigation of fracture phenomena from a
continuum theoretical perspective and by using atomistic simulation. This
side-by-side comparison provides an illustration of how a complex topic such
as fracture can be tackled by joint analysis from different perspectives. The
chapter included both studies with a simple model potential (harmonic poten-
tial, biharmonic potential) as well as studies with quantum mechanics based
ReaxFF reactive force fields.
The investigations and reviews in the area of dynamic fracture focused on
the following points.
• Comparison of atomistic results with continuum mechanics theory predic-
tions (in particular crack limiting speed, crack tip instability speed, and
deformation fields).
• Investigation of hyperelastic effects in dynamic fracture (in particular
regarding the crack limiting speed and the crack tip instability speed).
• Effect of geometric confinement (cracks in thin elastically inhomogeneous
layers) and crack propagation along interfaces of elastically dissimilar
materials.
• Fracture mechanisms in chemically complex materials, here illustrated for
the case of silicon.
The first study discussed in Sect. 6.4 centered on a one-dimensional model

of dynamic fracture. The appeal of the one-dimensional model is that many of
the physical phenomena of dynamic fracture, such as maximum crack speed
and a condition for crack initiation similar to the Griffith theory, also appear
in this simple model. An important aspect was that analytical expressions for
the nonlinear dynamics of fracture could be derived (see Sect. 6.4.3). The ana-
lytical model predicted crack motion faster than the speed of sound, if there
is an elastically stiff zone near the crack tip. This illustrates the significance of
the details of bond breaking on understanding dynamic fracture. The atom-
istic model with harmonic interactions is found to reproduce the predictions
of the linear elastic continuum theory well. The simulations carried out with
nonlinear interatomic potentials revealed that a small zone with stiff elas-
tic properties at the crack tip significantly changes the dynamics and allows
the crack to break through the sound barrier. This observation is in agree-
ment with the theoretical predictions of the model described in Sect. 6.4.3.
As shown in Fig. 6.20a, the crack propagation speed depends critically on the
onset strain of the hyperelastic effect and therefore, the crack speed is highly
sensitive to the size of the hyperelastic region. The deformation fields near
the one-dimensional crack in nonlinear materials agrees reasonably well with
the predictions by continuum theory, as shown in Fig. 6.20b.
After studying the one-dimensional crack, we moved on to two-dimensional
models. We utilized the elastic properties of two-dimensional solids, as
reviewed in Sect. 4.4.3. It was demonstrated that the choice of the potential
allows to construct model materials with the objective to probe the effect of
specific material properties on the dynamics of fracture. One of the examples
of such model materials is the biharmonic potential. This potential yields a
solid composed of two linear elastic materials, with one Young’s modulus asso-
ciated with small strains and one with large strains, representing a simplistic
model material for hyperelasticity. Further, the fracture energy was calculated
for different choices of the interatomic potential serving as input parameter
for the prediction of crack initiation time by the Griffith model [22].
A quantitative comparison of the deformation fields near a rapidly propa-
gating mode I crack in a harmonic lattice revealed that the continuum theory
predictions of angular variation of stress and strain agree well with the results
of atomistic simulations (Sect. 6.5). It was found that the prediction that the
hoop stress becomes bimodal [210] at a critical crack propagation speed is
reproduced in atomistic calculations. The occurrence of the bimodal hoop
stress is an important aspect in the theories of crack tip instabilities. In sum-
mary, the studies in Sect. 6.5 (together with the results of the one-dimensional
crack discussed in Sect. 6.4) reveal that atomistic simulations with harmonic
potentials are a good model for the linear elastic continuum theory. The results
in this section and the results of the one-dimensional model with harmonic
interactions both show reasonable agreement with the linear elastic continuum
theory [22].
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
such mechanisms for crack propagation along interfaces of aluminum (stiff)

and PMMA (soft) [267] corresponding to our atomistic model. The molecular
dynamics simulations reproduce some of the experimental findings.
The next section was devoted to a discussion of suddenly stopping cracks.
The main result was that mode I cracks in harmonic lattices carry no inertia,
and the static field spreads out behind the shear wave front immediately after
the crack is stopped. This result matches the prediction by continuum theory
[22]. A comparison of the suddenly stopping mode I cracks with experimental
results [23] also reveals good agreement. The results in discussed in this section
are in accordance with the results of a suddenly stopping one-dimensional
crack shown in Fig. 6.16. As soon as the crack stops, the strain field of the
solution corresponding to zero crack velocity is spread out. We have then
shown that mode II cracks behave differently than mode I cracks: In agreement
with the predictions by continuum mechanics theories of suddenly stopping
intersonic cracks [208], an intersonic mode II crack does carry inertia and the
static field does not spread out until the mother crack has reached the stopped
daughter crack. In the nonlinear cases of mode I and mode II cracks, the wave
fronts are smeared out and the static field is not instantaneously reached but
after all trails of waves have passed.
In the last two remaining sections we focused on the dynamics of mode
III cracks by using three-dimensional atomistic simulations. Firstly, we con-
sidered the mechanical and physical properties of three-dimensional solids.
We discussed the elastic properties and compared theoretical predictions with
numerical estimates for various crystal orientations in an FCC crystal. As
in the two-dimensional case, we also calculated the fracture surface energy.
The results of the section on the mechanical and physical properties of three-
dimensional solids provide important information for the studies discussed in
Sect. 6.10. After studying cracks in harmonic lattices and showing agreement
of the corresponding limiting speed (the shear wave speed), we focused on
the critical energy length scale χ. We find that the same concept as that dis-
covered in Sect. 6.6 also holds for mode III cracks. A quantitative comparison
with an analytical continuum theory solution [24] showed good agreement. For
realistic experimental conditions and cracks propagation within a thin steel
layer, the characteristic energy length scale was estimated to be on the order
of millimeters. Further studies of mode III cracks included suddenly stopping
cracks. It was shown that, as for sub-Rayleigh mode I and mode II cracks,
mode III cracks in harmonic lattices carry no inertia [22].
6.12.1 Hyperelasticity can Govern Dynamic Fracture
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
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.
Limiting Speed of Cracks
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
This characteristic length scale is found to be proportional to the fracture

surface energy and elastic modulus, and inversely proportional to the square
of the applied stress,
γE
χ∼ 2. (6.134)
σ
Contrary to the common belief, the crack does not need to transport energy
from regions far away from its tip, rather only from a small local region
described by the characteristic length scale. The assumption of linear elas-
ticity, and hence the classical theories, fails if the hyperelastic zone becomes
comparable to the local energy flux zone. This is because in soft materials
energy is transported slower, in stiff materials faster. Correspondingly, the

crack velocity becomes slower or larger once the hyperelastic region is suf-
ficiently large. If the region around the crack tip becomes stiff due to
hyperelasticity, more energy can flow to the crack tip in shorter time. In
the opposite, energy transport gets slower when there is a local softening zone
around the crack tip. Therefore, hyperelasticity is crucial for understanding
and predicting the dynamics of brittle fracture.
When hyperelasticity dominates, cracks can move faster than all elastic
waves as shown in Fig. 6.44. This is in clear contrast to the classical theories in
which it is believed that the longitudinal elastic wave speed is an impenetrable
upper limit of crack speed. Such phenomenon can only be understood from
the viewpoint of hyperelasticity.
Hyperelasticity dominates fracture energy transport when the size of the
hyperelastic zone approaches the energy characteristic length. Under normal
experimental conditions, the magnitude of stress may be one or two orders
of magnitude smaller than that under atomic simulations. In such cases, the
characteristic length is relatively large and the effect of hyperelasticity on
effective velocity of energy transport is relatively small. At about 1% of elas-
tic strain, the energy characteristic length is on the order of a few hundred
atomic spacing and significant hyperelastic effects are observed. It seems that
hyperelasticity can play the governing role especially in nanostructured mate-
rials such as thin films, or under high-impact conditions where huge stresses
occur, so that the region from which the crack needs to draw energy is small.
In the conventional picture of dynamic fracture, there exist three impor-
tant length scales near the crack tip, as shown in Fig. 6.109a. The fracture
process zone in which atomic bonds are broken is usually very small and
extends only a few angstroms in perfectly brittle systems. Another important
length scale is the K-dominance zone, which is relatively large. In between the
fracture process zone and the K-dominance zone is the region where material
response is hyperelastic. We proposed that there exists an additional length
scale near the dynamic crack, the characteristic energy length scale. This
new energy length scale is shown in Fig. 6.109b, and it is in between the
K-dominance zone and the hyperelastic region. If the size of the hyperelastic
region becomes comparable to the energy length scale, hyperelasticity governs
dynamic fracture. If it is much smaller, hyperelasticity can be neglected.
Crack Tip Instabilities
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
speed to propagate stable. On the basis of systematically varying the ratio

of large-strain elastic properties while keeping the small-strain elastic prop-
erties constant, it was shown that the instability speed depends on the local
wave speed. A generalized Yoffe criterion [210] and Gao’s analysis [27] of local
limiting speed helped to explain some of the simulation results.
With respect to the governing mechanism of the dynamic crack tip insta-
bility, the stiffening and softening case need to be distinguished. We illustrated
that in softening systems, the reduction in local energy flow governs the insta-
bility, and in stiffening systems, the change in deformation field near the crack
tip is responsible for the crack to leave its straightforward motion and branch.
The analysis of the stress and strain field support these assumptions. As dis-
cussed in Sect. 6.11, in chemically complex materials additional atomistic
mechanisms may play a role in initiating fracture instabilities. Here we have
discussed the formation of 5-7 defects during crack initiation in silicon, as
shown in Fig. 6.104b.
6.12.2 Interfaces and Geometric Confinement
Interfaces and geometric confinement play an important role in the dynamics

of cracks. Crack propagation constrained along interfaces can significantly
change the associated maximum speeds of crack motion. This is illustrated
for instance by the studies using the concept of a weak fracture layer where
the Rayleigh wave speed of cracks can be attained by cracks (see Sect. 6.5),
vs. the studies of cracks in homogeneous materials where the crack starts to
become unstable at 73% of the theoretical limiting speed.
If cracks propagate along interfaces of elastically dissimilar materials, the
maximum crack speed can significantly change and new mechanisms of crack
propagation such as daughter and granddaughter cracks appear. Geometric
confinement as cracks moving inside thin strips (the Broberg problem) has
proven to provide strong impact on the dynamics of cracks. If the crack prop-
agates in a small strip with different elastic properties, a significant effect on
the propagation speed of the crack is observed as illustrated in Fig. 6.46 for
mode I and in Fig. 6.93 for mode III cracks. An implication of crack motion
within a thin stiff layer is that mode I cracks can break through the shear wave
speed barrier and propagate at intersonic velocities as shown in Fig. 6.47, and
that mode III cracks become supersonic as shown in Fig. 6.92.
In summary, the atomistic studies suggest that geometric confinement has
strong impact on how cracks propagate. This is potentially important in com-
posite materials where understanding of crack dynamics may be critical in
designing robust and reliable devices.
The results reviewed here suggest that the definition of wave speeds accord-
ing to the small-strain elastic properties is questionable in many cases and
should be replaced by a local wave speed. Similar thoughts apply to the
definition of wave speeds across interfaces: When the elastic properties are
discontinuous, no unique definition of the wave speed and therefore the crack
limiting speed is possible.
7
Deformation and Fracture of Ductile Materials
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
macroscopic scales of material behavior that can be directly observed in exper-

iment. The goal of such atomistic models is to understand the macroscopic
response of materials, for example to mechanical stimulation, based on their
fundamental, atomistic ultrastructure.
Modern multiscale modeling techniques use a sequence of overlapping hier-
archies encompassing various simulation tools to bridge the scales from nano
to macro. These modeling techniques allow a rigorous linking of material
properties from quantum mechanics to mesoscale and macroscale. This fun-
damental viewpoint could revolutionize the engineering approach to use and
create materials, by incorporating the atomistic scale into materials analysis
and synthesis.
Experimental techniques such as TEM have been very fruitful in the anal-
ysis of dislocation structures and dislocation networks. HRTEM can even
resolve individual atoms, which has proven to be very useful in the analy-
sis of the structure of grain boundaries, (see, for instance, the plot shown in
Fig. 8.2 for an example of such an analysis). For mechanical testing, in addi-
tion to macroscopic tensile tests, the use of nanoindentation has led to a novel
tool for the investigation of ultra-small-scale plasticity.
7.2 Continuum Theoretical Concepts of Dislocations

and Their Interactions
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
This section is dedicated to a brief review of continuum mechanical

approaches in treating dislocations. Dislocations are the fundamental carrier
7 Deformation and Fracture of Ductile Materials 329
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.
7.2.1 Properties of Dislocations
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
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
In FCC crystals, slip is limited almost exclusively to {111} planes in 110

directions. This results in 12 independent slip systems of an FCC lattice.
Dislocations in FCC crystals are typically split up into partial dislocations. A
complete dislocation with Burgers vector 12 110 decomposes into two partial
dislocations (see Fig. 7.2)
1 1 1
110 → 211 + 121. (7.8)
2 6 6
√
The magnitude of the Burgers vector√of the complete dislocation is a0 / 2,
and that of a partial dislocation is a0 / 6. Frank’s rule can be used to illustrate
that this split into partial dislocations is energetically favorable. This can be
done by considering the square of the Burgers vector for both cases, showing
that it is a20 /2 for a full dislocation and a20 /3 for the two partial dislocations.
Thus, the energy of the partial dislocations is smaller than that of a full
dislocation. The nucleation of the first partial dislocation leads to a stacking
fault in the FCC lattice, which is repaired by the second partial dislocations.
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
7.2.2 Forces on Dislocations
Fig. 7.3 Geometry to explain the Schmid law, illustrating the angles φ and λ as
well as the uniaxially applied stress σy
The local stress tensor in a material leads to forces on dislocations. These

forces control the activation of specific slip systems. The particular activated
slip system is controlled by the resolved shear stress that acts on this particular
plane; among all possible planes the one with the largest resolved shear stress
is most likely to be activated.
The activation of particular slip systems is described based on Schmid’s
law:
τc = σy cos(φ) cos(λ) (7.10)
where σ0 is a uniaxially applied stress, and φ and λ are the angles of the
slip plane normal and the slip direction with respect to the axis in which the
tensile load is applied. A dislocation is nucleated when τc reaches a critical
value (please see Fig. 7.3). These considerations are used to develop criteria
that describe the critical conditions for nucleation of dislocations.
7.2.3 Rice–Thomson Model for Dislocation Nucleation
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 )
The Rice–Thomson model is a simple model to characterize the critical

conditions for nucleation of a dislocation from a crack tip. It is also a very
educational model to illustrate the physical process of nucleation of a disloca-
tion, which motivates us to review it here. In the following paragraph, we will
summarize the basic steps to obtain the solution for the critical condition for
dislocation nucleation from the tip of a crack, as shown in Fig. 7.4.
The basis for this model is to consider the balance of competing forces
acting on a dislocation, as shown in Fig. 7.5. In that sense, the dislocation is
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.
Fig. 7.6 Geometry of a dislocation close to a surface, at distance d
The image force of an edge dislocation at a surface is given by
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,
Fstress = τ (d)bL0 , (7.14)
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)
Then, the critical condition is given by

2d µ b
σ∞ (d) = (7.20)
a fτ (θ)4π(1 − ν) d
We note that the critical stress σ∞ is still a function of d, which is defined

as the critical distance at which the dislocation is considered to nucleate. It
remains unknown, but can be estimated to be one to three Burgers vectors.
For d = b (assuming that nucleation occurs at a distance of one Burgers
vector),
2b µ
σ∞ = (7.21)
a fτ (θ)4π(1 − ν)
and therefore
b
σ∞ ∼ µ. (7.22)
a
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.
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
7.2.4 Rice–Peierls Model
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
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.
7.2.5 Link with Atomistic Concepts
Dislocation mechanics concepts can be quite conveniently linked with atom-

istic concepts, and vice versa. Here we briefly discuss two aspects of how such
a link can be achieved.
7.2.6 Generalized Stacking Fault Curves
The Rice–Peierls model provides a direct link with atomistic concepts. As we

discussed, both the unstable stacking fault energy as well as the surface energy
can be calculated directly from atomistic calculation. Therefore, for a given
potential the parameters are defined uniquely.
In [113], the generalized stacking fault energy curves are calculated for
different EAM potentials of FCC metals. The authors show that the resulting
curves show similar characteristics but vary with respect to their agreement
with the experimental estimates of the intrinsic stacking fault energy. The
curves have been used to obtain estimates of the unstable stacking fault
energy γus [30]. Figure 7.10 shows such curves calculated from atomistic sim-
ulation, and indicates also how γus and γsf are calculated from such curves.
Figure 7.10a shows the calculation method of the GSF curve by sliding two
parts of the crystal along the [112] direction, and Fig. 7.10b shows the GSF
curves for four different potentials [31–35]. Note that the best agreement of γsf
to first principle calculation results are obtained from the potential proposed
by Mishin et al. [33] and Angelo et al. [32]. This finding is in agreement with
the results reported in [113].
Because of the nature of the stacking fault, the generalized stacking fault
energy curve depends on nonnearest neighbor interactions. This is impor-
tant to consider when models based on pair potentials are developed. In
[113], it is further shown that the EAM models produce the same value for
the dimensionless constant γus /(γslip bp ) as the simpler Lennard-Jones inter-
atomic potential. This suggests that the LJ pair potential should produce
the qualitatively same behavior as EAM models for a given type of plastic
deformation.
It is noted that the link between atomistic models and the Rice–Thomson
model is much more difficult to make. In that case, a possible strategy is
to calculate the stress intensity factor from molecular dynamics and use this
information to provide a nucleation criterion.
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
7.2.7 Linking Atomistic Simulation Results to Continuum

Mechanics Theories of Plasticity
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
Fig. 7.11 The multiplicative decomposition F = Fe Fp in continuum theory of

plasticity
mapping is illustrated in Fig. 7.11. In the continuum theory of plasticity, the

geometrically necessary dislocation density tensor A is defined as [287, 290,
291]
1
A = det(Fe )F−1 −1 T
e (curlFe ) = Fp (CurlF−1 T
p ) . (7.33)
det(Fe )
Note that Curl is the curl differential operator with respect to the material
point in the reference configuration, while curl is the curl operator with respect
to a material point in the current configuration.
From an atomistic point of view, the dislocation density can be expressed
as [287, 290, 291]
∆l
A=l⊗b (7.34)
∆v
where l denotes the unit tangent vector along the dislocation line segment,
∆l is the element of the dislocation line, and ∆v is the elementary volume.
The operator ⊗ denotes a dyad product. Under the assumption of infinitesi-
mal deformation and negligible elastic strain, the dislocation density tensor is
directly linked to the plastic distortion [287]. In the case of multiple disloca-
tion segments within a representative volume element, the dislocation density
tensor is defined by a linear combination of dislocations

A= ηk lk ⊗ bk (7.35)
k
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.
Equation (7.35) could be used to calculate the dislocation density tensor

from atomistic data. The slip vector approach, as described in Sect. 2.10.3, is
a possible candidate for this purpose. As discussed earlier, Fig. 2.41 shows the
result of a slip vector analysis of a single dislocation in copper. The quantita-
tive information obtained from atomistic results described in Sect. 2.10.3 can
be used to calculate the dislocation density tensor.
7.3 Modeling Plasticity Using Large-Scale Atomistic

Simulations
Unlike continuum mechanics approaches, atomistic techniques require no a
priori assumptions and no formulation of constitutive laws to model the
behavior of dislocations and thus describe mechanical properties of materi-
als. “Everything,” that is the complete material behavior, is determined once
the atomic interactions are chosen. Atomic interactions can be defined for a
specific material such as copper based on quantum mechanics calculations.
Alternatively, they can also be chosen such that generic properties common
to a large class of materials are incorporated. This allows to develop “model
materials” to study specific materials phenomena. Models for ductile materi-
als, for example, thus allow studying the generic features of ductile material
behavior.
The length and time range accessible to molecular dynamics is suitable for
studying dislocation nucleation from defects such as cracks, as well as compli-
cated dislocation reactions. The method also intrinsically captures dynamics
of other topological defects, such as vacancies or grain boundaries and its
interaction with dislocations. This is an advantage over mesoscopic methods
that require picking parameters and rules for defect interaction. Also, using
multibody EAM potentials (e.g., [94]), reasonably good models for some met-
als can be obtained. With sufficient computer resources it is possible to study
the collective behavior of a large number of dislocations in systems with high
dislocation density. Systems under large strain rates can be readily simu-
lated. In discrete dislocation dynamics methods, such conditions are difficult
to achieve.
Two distinct length scales are involved in the mechanics of networks of
crystal defects. The micrometer length scale is characteristic of the mutual
elastic interaction among dislocations, but dislocation cores and formation of
junctions and other reaction products is characterized by the length scale of
several Burgers vectors and occur at the atomic length scale [38]. The two
length scales span over several orders of magnitude, indicating the computa-
tional challenge associated with modeling. The rapidly advancing computing
capabilities of supercomputers approaching TFLOPs and beyond now allow
simulations ranging from nanoscale to microscale within a single simulation.
The state-of-the-art of of ultra-large-scale simulations can model billion atom
systems [138, 140, 219, 292].
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
et al. [299], using a QM-Sutton-Chen many body potential. The researchers

studied the core geometry of partial dislocations, as well as the motion and
annihilation of oppositely signed dislocations, and discussed cross-slip and
associated energy barriers. Atomistic simulations have also been applied to
study the interaction of dislocations with other defects.
Further studies focused on the ductility of quasicrystals (see also dis-
cussion in Sect. 6.3) [226, 227, 229]. Atomistic simulation particularly helped
to explain the mechanism of dislocation motion in quasicrystals. An impor-
tant contribution was the observation of phason-walls that are attached to
each moving dislocation. These phason-walls helped to clarify some of the
perplexing properties of quasicrystals found in experiments, such as a brittle-
to-ductile transition at about 80% of the melting temperature. Most recently,
three-dimensional atomistic simulations at elevated temperatures were car-
ried out [300] where it was found that dislocation climb processes play an
increasingly important role.
In summary, molecular dynamics simulations of plasticity have advanced
to a quite sophisticated level. Atomistic simulations with multibody EAM
potentials can also be applied to describe mechanical properties of thin metal
films, thereby providing a path to investigate size effects of materials behavior.
7.4 Case Study: Deformation Mechanics of Model FCC

Copper – LJ Potential
To illustrate the analysis of an atomistic simulation and the interpretation of
the deformation mechanisms in light of theoretical predictions, we review the
analysis of a large-scale molecular dynamics simulation of work hardening in
a model system of a ductile solid of a simulation originally reported in [138].
Two systems will be considered, first a simple model with an LJ potential (this
section). Then, a model simulation with an EAM potential (next section). The
comparison between the two results illustrates some of the differences between
both atomistic models of metals.
7.4.1 Model Setup
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
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
and cross-slip mechanisms, and (3) formation of sessile dislocations such as

Lomer–Cottrell locks. On the typical time scale of molecular dynamics simu-
lations, the dislocations self-organize into a complex sessile defect topology.
The analysis illustrates numerous mechanisms formerly only conjectured
in textbooks and observed indirectly in experiments.
The plastic or nonreversible deformation of materials occurs immediately
after a regime of recoverable elastic deformations and is governed by the nucle-
ation and motion of defects in the crystal lattice [1, 38, 60]. In experiment,
researchers often rely on indirect techniques to investigate the creation and
interaction of defects. In theory, predictions are primarily based on continuum
theory with phenomenological assumptions. While the continuum description
has been very successful in the past, some of the key features of plasticity
can only be understood when the atomistic viewpoint is taken into account,
and this may be achieved using atomistic simulation methods, like molec-
ular dynamics [28, 84, 138, 149, 152–154, 302–305]. However, most molecular
dynamics studies consider a small number of dislocations and address specific
dislocation mechanisms; for instance, dislocation nucleation from cracks or
dislocation reactions [173, 298].
In the present simulations, a simple interatomic force law is adopted. This
choice is motivated by the goal to investigate the generic features of a partic-
ular many-body problem common to a large class of real physical systems and
not governed by the particular complexities of a unique molecular interaction.
The model potential for the present study is the Lennard-Jones potential [84].
With its well-known shortcomings, it nevertheless provides a fundamental
description of the generic features of interatomic interaction: Atomic repulsion
at close distances, and attraction at large radii of separation.
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.
7.4.2 Visualization Procedure
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.
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
Dislocation Cutting Processes
The process of work-hardening or strengthening of materials is often due to

an additional constraint on the mobility of dislocations. For instance, once a
moving dislocation line hits an obstacle, it bends backwards as it senses the
retracting force of the obstacle. This general process is depicted in Fig. 7.13.
Figure 7.14 shows the dislocation–particle interaction in ordered matrix
materials [306]. A dissociated complete dislocation interacts individually with
the particle (Fig. 7.14a). A TEM weak-beam micrograph of the dislocation–
particle interaction is shown in Fig. 7.14b.
During work-hardening, dislocation cutting processes lead to similar mech-
anisms that hinder the free motion of dislocations.
Fig. 7.14 Dislocation–particle interaction in ordered matrix materials. A dissoci-

ated complete dislocation interacts individually with the particle (subplot (a) for a
schematic). A TEM weak-beam micrograph of the dislocation–particle interaction
in Fe–30 at.%Al is shown in subplot (b). Reprinted from Acta Metallurgica, Vol.
46(16), pp. 5611–5626, E. Arzt, Size effects in materials due to microstructural and
dimensional constrains: A comparative review, copyright c 1998, with permission
from Elsevier
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.
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
Generation of interstitials is energetically expensive and thus not observed

very frequently. The mechanism of dislocation cutting is shown in Fig. 7.16a.
Another possible mechanism of dislocation cutting is when a partial disloca-
tion moves through the stacking fault generated by another partial dislocation
as shown in Fig. 7.16b.
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
Intersection of a partial dislocation with the stacking fault of another dis-

location is observed in early stages when dislocations within the same cloud
intersect the stacking fault generated by other dislocations. The dislocation
line forms a jog that features a sessile edge component. Creation of a trail of
point defect causes a resulting drag force on the dislocation which results in
a bowing out of the segment. Because of the specific Burgers vectors of the
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
intersecting dislocations and the line orientation,

√ the cutting dislocations only
have a small screw component equal to 6/12a0 (half the length of a partial
dislocation). Therefore, not a complete point defect is generated but rather a
trail of local lattice distortion, a phenomenon that will be referred to as a trail
of partial point defects. The defect has a dipole structure, and is depicted in
Fig. 7.17a.
The dragging force of the trail of partial point defects was estimated by
calculating the energy per length for a trail of partial point defects and that
of a vacancy tube, and it was calculated to be about 20% of that of a com-
plete vacancy tube. This defect thus causes a significant dragging force on
the dislocations, in particular, in a situation where the dislocation density is
extremely high as in the present simulation. Two representative dislocation-
cutting processes are schematically shown in Fig. 7.18. The forces on the jog
segment are a combination of tensile and compressive forces in the two 112
directions. This immediately explains the dipole structure seen in Fig. 7.19c.
The bowing effect on the dislocations is shown in Fig. 7.19a and b, based on a
centrosymmetry analysis in Fig. 7.19a, and on an energy analysis in Fig. 7.19b.
So far we have only discussed the dislocation reactions that take place
when dislocations of the same clouds react. Numerous dislocation reactions
occur when dislocations of the two different clouds start to interact, as shown
in Fig. 7.20. Such reactions primarily involve the mechanism is dislocation-
cutting processes as depicted schematically in Fig. 7.16(a). Due to the Burgers
vectors and the dislocation line orientation, when the dislocation clouds meet
straight ahead of the crack, the reactions are very similar to those observed in
the previous stages when dislocations of the same clouds cut each other’s
stacking fault and thus, trails of partial point defects are generated. In
Fig. 7.19c, the significant effect of jog dragging on the motion of the dislo-
cations is clearly observed. The elastic interaction of dislocations repelling
each other causes a decrease in the dislocation velocity.
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.
Fig. 7.21 Activation of secondary slip systems and generation of Lomer–Cottrell

locks. Subplot (a) shows a schematic of the cross-slip mechanism. Subplots (b) and
(c) show details of activation of secondary slip systems (subplot (c) represents a
magnified view of subplot (b), with the region of interest highlighted by a circle).
This mechanism of cross-slip of partial dislocations, here first observed in molecular
dynamics simulation, was originally proposed theoretically by Fleischer [37], and
contrasts the well-known Friedel–Escaig mechanism [38]
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.
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)
Activation of Secondary Slip Planes
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
In the simulation, it is found that dislocations on secondary slip sys-

tems are generated by cross slip and Frank–Read mechanisms. This is an
unexpected observation because cross slip is only possible, according to the
classical dislocation mechanics, along at least locally constricted screw seg-
ments (Friedel–Escaig’s mechanism) [38]. Due to the low stacking fault energy
of the model material used here, only partial dislocations exist in the simula-
tion. Each occurring cross-slip event leaves behind a straight, sessile stair-rod
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.
Formation of Sessile Locks
Another mechanism of dislocation interactions is the formation of sessile locks.

An analysis of the Burgers vectors of the primary dislocations reveals that
formation of sessile locks is not observed until dislocations on secondary glide
planes are activated.
Sessile dislocation locks can be formed depending on the Burgers vector
whenever two partial dislocations on different glide planes get close together.
Some combinations of partial dislocations are attractive and form a disloca-
tion line with Burgers vector of the type 16 [110]. These dislocations are not
glissile on any glide system of the FCC lattice, and therefore it is sessile
and cannot move. Such defects provide a serious burden for further dislo-
cation glide through the material, since other dislocations approaching the
sessile dislocations cannot easily glide through this defect agglomerate, the
so-called Lomer–Cottrell lock. One of the possibilities how such locks may be
circumvented is by Orowan–mechanisms [1, 38].
Formation of sessile dislocations is also observed in the simulation. It is
found that the sessile dislocations severely hinder further dislocation motion,
as it is assumed in the classical theories of work hardening [38]. In Fig. 7.23 we
show some snapshots when such defects are generated. The typical structure
of the Lomer–Cottrell lock is characterized by a straight sessile arm connected
with two partials on different glide planes. This can be clearly seen in the fig-
ure. Formation of Lomer–Cottrell locks has been studied in several molecular
dynamics simulations, as for instance in [315].
The Work-Hardening Regime
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.
7.4.4 Summary
It is observed that dislocation-cutting processes generate a large number of

trails of point defects. An interesting aspect is the observation of generation
of trails of partial point defects by dislocation cutting processes with an orien-
tation such that the sessile edge component of the jog is relatively small. This
does not cause generation of a complete row of vacancies or interstitials, but
rather a local lattice distortion (see Fig. 7.17c). As it was shown in the simu-
lations reviewed here, it has an important effect on hindering free dislocation
motion. Such deformation mode could become important in nanostructured
materials, where it is now understood that partial dislocations can dominate
plastic deformation [153, 154, 303, 304]. Indeed, in some molecular dynam-
ics simulations of nanocrystalline materials, similar cutting mechanisms as
described in the present work have been observed [43, 153, 154, 304, 311].
Once the dislocation density in the center of the solid grain exceeds a
critical value, dislocations on secondary slip systems are activated. Since
only partial dislocations are present in the current study, the dislocations
must leave a sessile dislocation segment to conserve the Burgers vector (see
Fig. 7.21b). Such mechanism of partial dislocation cross-slip was actually pro-
posed theoretically almost 50 years ago [37], and directly simulated by the
molecular dynamics simulation.
Another important dislocation reaction mechanism is the formation of
sessile Lomer–Cottrell locks (see Fig. 7.21). Formation of sessile locks is not
observed until dislocations on secondary glide planes are activated.
The formation of complete vacancy tubes is noticed once secondary slip
systems are activated (see Fig. 7.16). The observation of generation of vacan-
cies is in agreement with the understanding that heavy plastic deformation
causes generation of vacancies. Mott (1960) [316] was the first to predict the
vacancy formation by motion of jogged screw dislocations. The production of
vacancies [307] and their influence on plasticity has been studied in experi-
ment and theory, [149,305,310,317,318] and has also been included in discrete
dislocation modeling recently [317].
Figure 7.25 plots the development of the density of different defects dur-
ing the simulation using a method of separating defects of different energies.
The potential energy of atoms allows discrimination between different types
of defects. This figure summarizes the hardening process in the simulation. At
the first stage, partial dislocations are emitted from cracks with a high rate. As
a consequence of cutting processes trails of point defects are produced. Finally,
the activation of secondary glide systems results in additional dislocation cut-
ting processes including the generation of complete vacancy tubes. Another
important hardening mechanism is formation of sessile locks. It is interesting
to note that the rapid production of partial dislocation ceases, with the cre-
ation of cutting debris (and the plastic relaxation), and the density of partial
dislocation finally goes into an equilibrium.
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
lead to activation of all of the possible dislocation mechanisms concurrently,

whereas under more realistic conditions, the system discriminates between
the different mechanisms, then favoring the lowest-energy path. Therefore,
although MD simulation can elucidate the possible mechanisms, experimen-
tal verification of the rate limiting processes remains a critical issue. In this
spirit, the molecular dynamics experiments could thus be viewed as informa-
tive, but not representative of the hardening mechanisms under more normal
conditions.
To investigate the hardening mechanisms with more realistic potentials
for metals, we review additional simulations of work-hardening using poten-
tials developed within the EAM framework in the next section. As expected,
similar hardening mechanisms are discussed in this chapter are observed. In
particular, dislocation cutting processes, cross-slip, and formation of sessile
locks play a dominating role.
We emphasize that such detailed atomistic views on fundamental aspects
of plasticity, as shown in the study reviewed here, can neither be obtained
from experimental techniques, nor can they be calculated with continuum
mechanics methods, especially in view of extremely high dislocation densities
and strain rates. In this respect, with its limitations understood, molecular
dynamics simulations represent a convenient, unique and straightforward way
to obtain such information.
7.5 Case Study: Deformation Mechanics

of a Nickel Nanocrystal – EAM Potential
The objective of this section is to review a computational experiment as dis-
cussed in the previous section, this time using a different interatomic potential.
Here an EAM potential, parameterized for nickel, is used to carry out a simu-
lation of large-deformation plastic behavior. Figure 7.26 depicts the geometry
and the setup of this experiment.
Figure 7.27 shows a sequence of snapshots that illustrate the dislocation
nucleation process from the crack tip. This snapshot depicts the early stage
of dislocation nucleation.
Figure 7.28 depicts a closer view on the dislocation structure at a later
stage in the simulation. At later stages, secondary slip systems are activated,
as shown in a centrosymmetry analysis depicted in Fig. 7.29. A few notable
differences to the LJ case are:
• Full dislocations are nucleated from the crack tip, due to the finite stack-
ing fault energy (in contrast to the zero stacking fault energy in the LJ
case). Full dislocations are visible due to the leading and trailing par-
tial, enclosing the stacking fault region (see also Fig. 7.30 for a direct
visualization).
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
• Different dislocation interactions take place, in particular, the sessile seg-

ments that are generated due to dislocation cutting mechanisms are of
finite length (see also Fig. 7.31). This is because the stacking fault regions
are also of finite length, and because the sessile defect only appears in
regions where two stacking fault regions intersect.
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)
7.6 Case Study: Multi-Paradigm Modeling of Chemical

Complexity in Mechanical Deformation of Metals
The prediction of the deformation behavior of metals in the presence of envi-
ronmentally embrittling species like water or hydrogen, or under presence of
organic oxidative chemicals presents a critical challenge in materials modeling.
This exceeds the capability of EAM-type potentials, since chemical reactiv-
ity and a variety of chemical bonds are the key to describe the behavior of
these materials appropriately. This task can be accomplished by a combina-
tion of the first principles based reactive force field ReaxFF and the embedded
atom method (EAM) in a generic multiscale modeling framework, as reported
in [41]. This hybrid method enables one to treat large reactive metallic systems
within a classical molecular dynamics framework. The hybrid method is based
on coupling multiple Hamiltonians by weighting functions that allows accurate
modeling of chemically active sites with the reactive force field, while other
parts of the system are described with the computationally less expensive
EAM potential.
In this section, a brief review of a case study is provided that illustrates
the significance and the potential of this approach.
Here we review the application of a hybrid modeling scheme in the study
of fracture of a nickel single crystal under the presence of oxygen molecules,
where thousands of reactive atoms are used to model the chemical reactions
that occur under the presence of oxygen. This hybrid method constitutes an
alternative to existing methods that are based on coupling quantum mechan-
ical methods such as DFT to empirical potentials. The goal of the studies
Fig. 7.29 Centrosymmetry analysis that shows the activation of secondary slip
systems, at later stages in the simulation with the EAM potential
reviewed here is to simulate the effect of organic molecules interacting with a

crack opening. It is observed that the oxide formed on the crack surface pro-
duces numerous defects surrounding the crack, including dislocations, grain
boundaries, and point defects. It is shown that the mode of crack propaga-
tion changes from brittle crack opening at crack tip to void formation ahead
of crack and void coalescence for {111}112 orientation of the crack. The
results illustrate the significance of considering oxidative processes in studying
deformation of metals.
The solution reviewed here is to take advantage of the best qualities of the
ReaxFF force field (that is, its ability to model metal–organic interactions
and charge transfer) together with the best qualities of EAM potentials (that
is, in its computational speed and accuracy of descriptions of metals and
defects in metals) combined in a hybrid scheme, the ReaxFF-EAM coupling
as discussed in Sect. 5.4.5 (for a schematic, see Fig. 5.8). This will enable us,
for the first time, to carry out finite temperature calculations of deformation
of large reactive metal systems that include metal atoms, oxygen molecules,
hydrogen molecules, or water molecules, covering a wide range of the periodic
table.
7.6.1 Atomistic Model and Validation
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
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
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.
Shear Loading (Mode II) of an Initial Crack in a Single Crystal

of Nickel
A single nickel crystal with an elliptical crack in the 110112 orientation is

shear loaded in its x-direction faces, so that a pair of partial dislocations is
emitted from the crack tip. As the partials move in the EAM region, the
separation distance between them is found to be approximately 17 Å (see
Fig. 7.32 for the geometry and Fig. 7.33 for simulation results).
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
Fig. 7.33 Centrosymmetry analysis plot of a complete dislocation emitted as two

partials from a crack under shear (mode II) loading moving through EAM and
ReaxFF regions (atoms with perfect FCC coordination have been removed). The
dashed black circles indicate the region of 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–e) are taken at
time interval increments of 500 fs each. The distance between the partials appears
unchanged throughout the nucleation and propagation process, and does not change
as it passes through the handshaking and ReaxFF regions and back into the EAM
region
region. However, as a conservative estimate the transition width is left at 7 Å

for all simulations reviewed in the remainder of this section.
This test calculation proves that the hybrid scheme does not influence
dislocation propagation dynamics. In particular, it shows that a dislocation
can easily enter and leave a reactive region inside a crystal. Figure 7.33 depicts
snapshots of this simulation.
Uniaxial Shock Loading of a Perfect Nickel Crystal
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.
7.6.2 Example Application: Modeling Hybrid Metal–Organic

Systems
The study of effect of environmental effects of O2 , H2 , and H2 O on crack

propagation in metal crystals provides important insight into the embrittling
mechanisms that are at play in changing the fracture strength of the material.
The effect of formation of a layer of oxide at the surface of cracks in the
metal can play an important role in modifying its fracture properties. Here
we particularly pay attention to the effect of atmospheric oxygen on single
crack propagation in a single crystal nickel under a tensile load in mode I
loading.
We are interested in early fracture events and in particular, in the first
deformation event near the crack tip. This problem has been intractable
to empirical potentials atomistic modeling because of the unavailability of
a proper empirical potential that describes the nickel oxide correctly. The use
of ab initio methods to model the entire problem is prohibitively large. This
constitutes a perfect problem for multiscale modeling in CMDF with the oxi-
dation regions, where nickel–oxygen chemistry can be handled by the ReaxFF
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
potential, and the unoxidised bulk Ni can be modeled by the computationally

more efficient EAM potentials [40, 41].
Deformation Dynamics of a Single Nickel Crystal

with an Elliptical Crack Under Presence of Oxygen Molecules
A perfect crystal with an initial elliptical crack to serve as failure initiation

point is considered here. Different orientations of a single Ni crystal with dif-
ferent crack orientations were considered in this study to investigate the effect
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].
Crack in 110112 Direction
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
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
ensemble is used at 400 K to equilibrate the system to allow faster reaction of

the nickel and oxygen.
After 10,000 molecular dynamics integration steps, the temperature is
slowly quenched down close to 0 K, allowing the system to settle into a local
energy minimum. Over equilibration, the oxygen molecule immediately binds
to the nickel surface, breaking apart into two oxygen atoms in the process.
The oxygen atoms at the surface pull nickel atoms from the bulk to create
a cluster around them, leading to formation of geometric misfit dislocations
inside the bulk nickel, away from the initial oxidized site. This is seen for
all cases where oxygen is added, showing the presence of additional defects
around the crack tip owing to the presence of oxygen. The shape of the crack
changes and volume of empty space inside the crack is decreased.
Upon straining the crystal, it is observed that the crack tip emits the first
partial dislocation at about the same applied tensile strain as in the case
without oxygen.
Since the critical nucleation strain does not change in these two cases, the
modification in shape and reduction of crack volume thus does not seem to
affect the strain at which the first defect is nucleated. The dislocation is emit-
ted from the crack tip end away from the oxide layer and it moves into bulk
nickel without encountering the oxide. This hints that a regularly distributed,
thick oxide layer all around the entire crack surface may be required for a
proper study of crack propagation through the oxide layer. However, a much
larger number of O2 molecules are required to successfully cover the entire
crack surface with oxide, with a much large reactive region that encompass
more than 10,000 reactive atoms. This large number of reactive atoms is
beyond the capabilities of the current numerical implementation of ReaxFF.
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
Crack in 111112 Direction
The simulation cell is set up identically as described in the previous section,

but with 111 being the straining direction and the elliptical crack aligned
with major and minor axes along 110 and 111 directions, respectively.
As before, we first investigate the reference case with no oxygen present
in crack region and all atoms modeled by EAM potentials. The crack tip
starts to fracture in a brittle fashion at a strain xx = 0.05. We also observe
emission of partial dislocations at the other crack tip end at a strain of xx =
0.051.
The placement of a random oxygen molecule inside the crack region leads
to rapid oxide formation at the surface of the crack. The system is equili-
brated at a temperature of 400 K, using an NVT ensemble at 400 K for 10,000
molecular dynamics steps at a time-step of 0.5 fs, as for the case described
above.
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]
The equilibration creates a geometric dislocation close to the crack tip

as seen in Figs. 7.37 and 7.38. The differently shaded region to the left of
the crack represents a grain boundary that has formed during the oxidation
process (dashed line indicates grain boundary). The tensile straining of the
cell, however, shows a different mechanism of fracture propagation (Fig. 7.37):
Upon straining to the critical strain of 0.028 – 0.03, void formation takes place
near the crack tip close to the location of the dislocation, and the void enlarges
until it joins the main crack at a strain of 0.034. Subsequent propagation
is through the emission of partial dislocation at the crack tip at strains of
0.05.
A similar behavior is found for a different initial placement of the oxygen
molecule near the crack tip, and thus different areas of oxide formation. The
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.
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
This chapter is dedicated to a discussion of atomistic simulation of size effects

in the deformation and fracture mechanics of geometrically confined materials,
including nanocrystalline materials, thin films, as well as biological nanoma-
terials. The discussion is focused on both brittle and ductile nanomaterials.
Particular emphasis is given to studies of thin metal films and their plastic-
ity behavior. A direct comparison of simulation results is provided for several
cases, illustrating how a direct link can be made between both approaches.
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
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
Fig. 8.2 High resolution TEM images of grain boundaries in electrodeposited

nanocrystalline Ni (a) and nanocrystalline Cu (b). The samples were produced by
gas-phase condensation. The grain boundaries constitutes itself as a very narrow
region between crystals of different orientation. Reprinted from Acta Materialia,
Vol. 51, K.S. Kumar, H. Van Swygenhoven and S. Suresh, Mechanical behavior of
c 2003, with permission
from Elsevier
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
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
critical condition and understand the underlying principles, for a variety of

materials and a variety of microstructures.
This concept has been confirmed in several studies and in different geome-
tries. Figure 8.3 depicts an analysis of the normalized flow stress obtained
from various sources, including bubble raft experiments, experiments with
electrodeposited nanocrystalline metals (see [323] for references), as well as
molecular dynamics simulations [324]. Figure 8.4 depicts the results of a sys-
tematic investigation of size effects in nanocrystalline copper, carried out using
molecular dynamics simulations [43]. These plots illustrate the existence of a
peak at a critical, strongest grain size. A similar behavior is found in other
geometries. Figure 8.5 depicts results of molecular dynamics simulations that
illustrate the size dependence of a bioinspired metallic nanocompsite, showing
the existence of “a strongest size” [44]. The increase in strength scales very
well according to the Hall–Petch relationship given in (8.1), with parameters
kY = 0.11 MPa m1/2 . and σ0 = 1.2 GPa. The physical reason for the departure
from the Hall–Petch behavior is the onset of interfacial sliding mechanism (at
the interfaces between the Ni and Al particles) and the associated breakdown
of dislocation-based plasticity.
Applying classical molecular dynamics to investigations of nanostructured
materials is particularly attractive because of the fact that the lengthscale
of several tens of nanometers fall well within the range accessible to molecu-
lar dynamics simulation. Indeed, classical molecular dynamics methods have
proven to be a very powerful tool for these materials. Studying deformation
Fig. 8.4 Illustration of the maximum in the strength of nanocrystalline copper, as

shown by molecular dynamics simulation of up to 100 million atoms [43]. Panel (a)
shows the stress–strain curves for ten simulations with varying grain sizes. Panel (b)
depicts the flow stress, defined as the average stress in the strain interval from 7 to
10% deformation. The error bars indicate the fluctuations in this strain interval (1
standard deviation). A maximum in the flow stress is seen for grain sizes of 10–15 nm,
caused by a shift from grain boundary mediated to dislocation mediated plasticity.
Reprinted from Science, Vol. 301, J. Schiotz and K.W. Jacobsen, A Maximum in
the Strength of Nanocrystalline Copper, copyright c 2003, with permission from
AAAS
Fig. 8.5 Size dependence of a bioinspired metallic nanocomposite (schematic of

structure shown in left part) [44, 45], illustrating the existence of “a strongest size”
[322]. The increase in strength for building block dimensions larger than approximately
50 nm scales rather well according to the Hall–Petch relationship given in (8.1)
of nanocrystalline materials with molecular dynamics still requires signif-

icant computer power. Atomistic studies of nanostructured materials and
associated size effects were reported by several groups (see, for example
[44, 45, 68–70, 153, 154, 305, 325]). In most of the molecular dynamics studies,
polycrystalline samples at nanoscale were created (for example by a Voronoi

construction), annealed, relaxed, and then exposed to tensile loading.
In the following paragraphs, we summarize the main results in this field
obtained for different geometries, materials, and simulation conditions (e.g.,
variation of temperature and loading conditions). Coble creep is a well-known
mechanism for creep of polycrystalline materials [320]. The characteristic time
for exponential stress relaxation scales as
τ ∼ d3 , (8.4)
where d is the grain diameter. Appreciating that the grain size in nanos-
tructured materials is on the order of tens of nanometers (in contrast to
micrometer grain sizes in coarse-grained materials), this scaling suggests that
at very small grain size, diffusive mechanisms at grain boundaries may play a
dominating role in nanomaterials even at moderate temperatures.
In recent publications, this was investigated using molecular dynam-
ics simulations at elevated temperature [68, 303, 326]. The temperature was
increased to render the process of diffusion accessible to the molecular dynam-
ics timescale. The authors [68, 303] used a fully three-dimensional model of
palladium with 16 grains having a truncated-octahedral shape arranged on
a three-dimensionally periodic BCC lattice. Grain sizes range from d ≈ 3.8
to 15 nm, and the grain boundary misorientations are chosen such that only
high-energy grain boundaries are present in the model. A multibody EAM
potential was used to model the atomic interactions. They find that grain
boundary processes indeed play a dominating role and conclude that grain
boundary diffusion fully accounts for plasticity. Under lower strain rates than
in molecular dynamics simulation, this result could be valid even at room
temperature, once microcracking and dislocation nucleation are suppressed.
Dislocation mechanisms are shut down due to the small grain size and mod-
erate loading of the sample! The authors derive a generalized Coble-creep
equation and show that the grain-size dependence of the strain rate decreases
from the 1/d3 scaling law appropriate for large grain size toward a 1/d2 scal-
ing law as expected in the limit of a very small grain size (critical grain size
d ≈ 7 nm in palladium). The grain size scaling observed in molecular dynamics
simulations indeed agrees with this prediction [68]. It is also concluded that
grain boundary diffusion creep must be accommodated by grain boundary
sliding (also referred to as Lifshitz sliding) to avoid microcracking.
Experimental reports of the inverse Hall–Petch behavior inspired numer-
ous simulation studies by Schiotz and coworkers [324, 327] and Swygenhoven
and coworkers [70, 152, 328–330]. In contrast to the above research of Coble
creep, these simulations are all performed at low temperatures making it
basically impossible to observe any Coble creep at the molecular dynamics
timescale. In these studies, very large stresses in the range of 1–3 GPa were
applied. Schiotz et al. [324, 327] determined the yield stress σY as a function
of the grain size d. In contrast, the group around van Swygenhoven focused
attention on the strain rate. Both groups concluded that the deformation
mechanism is controlled by grain boundary processes and that the material

softens with decreasing grain size (inverse Hall–Petch effect). Nucleation of
numerous partial dislocations was observed in their simulations.
Schiotz and coworkers [324, 327] considered nanocrystalline copper with
grain sizes from 3.3 to 6.6 nm and showed that grain boundary sliding occurs
together with grain rotation. When the grain size was larger than about
5 nm, nucleation of partial dislocation was identified under very large stresses.
Similar observations were also reported by van Swygenhoven and cowork-
ers [329, 330] in simulations of nickel at average grain sizes of about 5 nm at
a temperature of 70 K. The results were confirmed with simulations at higher
temperature and for larger grain sizes [70]. The authors suggested that grain
boundary sliding occurs through atom shuffling and stress-induced athermal
grain boundary diffusion. In a later paper by Wolf et al. [303], the missing
issue of the rate-limiting deformation mechanism was addressed. The authors
suggest that the accommodation mechanism in the simulations described by
van Swygenhoven’s and Schiotz’s group is the same as that in Coble creep,
with the difference that there is no activation energy for this athermal process.
Therefore, the Coble creep equation should apply. They verified this proposal
by an analysis of the data in [330], proving that the data points for the three
smallest grain sizes fall on a straight line with a slope 2.73 in a log–log plot of
the /σ
˙ vs. the grain size d (Fig. 4 in [303]). It was concluded that the ather-
mal mode of Coble creep is due to the fact that the simulations are carried
out in a regime where molecular dynamics can not be used. The fact that
Coble creep still dominates may indicate that grain boundary diffusion is a
very robust mechanism for stress relaxation [303].
Recent work by Hasnaoui et al. [331] discussed the influence of the grain
boundary misorientation on the ductility of nanocrystalline materials. It
was shown that at specific low-energy grain boundaries (e.g., twins), several
neighboring grains can be effectively immobilized, creating structures that
offer significant resistance to plastic deformation. The authors finally discuss
the possibility to design more ductile nanostructured materials that feature
less low-energy grain boundaries and therefore lead to a more homogeneous
deformation.
Other studies were carried out on dislocation processes of nanocrys-
talline aluminum [304]. The authors demonstrate that deformation twinning
may play a very important role in the deformation of nanocrystalline alu-
minum. The simulations demonstrate that molecular dynamics simulations
have advanced to predict deformation mechanisms of materials at a level
of detail not yet accessible to experimental techniques. Observation of twin-
ning is quite surprising because of the small grain size and the high stacking
fault energy of aluminum [112]. The predictions by these simulations have
recently been verified experimentally [332]. Experimentalists conclude that
twinning in aluminum only occurs in nanocrystalline materials, while it is not
observed in coarse-grained aluminum. The findings support the hypothesis
that in the nanograin-regime, a transition occurs from normal slip of complete
dislocations to activities dominated by partial dislocations. The critical stress

for nucleation of dislocations in nanocrystalline aluminum was estimated to
be 2.3 GPa.
Research has focused not only on the polycrystalline nanoscale mate-
rials, but also on the mechanics of single crystals with nanometer exten-
sion, so-called nanowires [333–338]. Such structures may become increasingly
important for example as interconnects in complex integrated circuits or
bioelectrical devices. Studies of defect-free single nanocrystals under tension
(the crystals had dimensions of several nanometers) have been carried out
by Komanduri et al. [106]. Due to the small structural size of the nanocrys-
tals, the dislocations glide quickly through the specimen leaving surface steps,
and repeated glide admits plastic deformation. Similar research of mechanical
properties of copper were carried out by Heino and coworkers [339].
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
The combination of new micromachining techniques and studies of size-

dependent plasticity is a particularly interesting approach. In a recent study,
researchers developed a test methodology that allows the exploration of
size-scale effects in inorganic materials by using a focused ion beam (FIB)
microscope for sample preparation. This enables one to create small-scale

structures in a controlled way, whose mechanical properties can be tested
using a modified nanoindentation technique. In a study reported in [340], the
researchers created pillars made out of nickel, consisting of different sizes,
and then measured the associated size-dependent plasticity. Figure 8.6 shows
results of this experimental study. The measurements of plastic yielding for
single crystals of micrometer-sized dimensions for three different types of met-
als showed that the characteristic size of the sample can limit the lengthscales
available for plastic deformation mechanisms. The results reported in [340]
show significant size effects even at sample sizes beyond a few micrometers.
At micrometer lengthscales, it is crucial to consider both the overall sample
geometry and the internal structure to determine the strength of the sample.
8.2 Thin Metal Films and Nanocrystalline Metals

The study of the mechanical properties of materials at nanoscales and sub-
micrometer scales is motivated by increasing need for such materials due
to miniaturization of engineering and electronic components, development of
nanostructured materials, thin film technology, and surface science. When the
material volume is lowered, characteristic dimensions are reduced that control
the material properties and this often results in deviation from the behavior of
bulk materials. Small-scale materials are often referred to as materials in small
dimensions, and they are defined as materials where at least one dimension is
reduced. For instance, thin films bond to substrates are a relevant example of
materials in small dimensions since the film thickness hf is small compared to
the planar extension of the film and the thickness of the substrate. Thin films
bond to substrates have become an increasingly active area of research in the
last decades. This can partly be attributed to the fact that these materials
are becoming critically important in today’s technologies, whereas changes in
material behavior due to the effects of surfaces, interfaces, and constraints are
not completely understood.
The focus of this part is on mechanical properties of ultra-thin submicron
copper films on substrates. We will show that in such materials, important
effects of the film surface and grain boundaries are observed and that the
constraint by the film–substrate interface governs the mechanical behavior
[46, 51, 96, 341–344].
Polycrystalline thin copper films as shown schematically in Fig. 8.7 are
frequently deposited on substrate materials to build complex microelectronic
devices. In many applications and during the manufacturing process, thin
films are subjected to stresses arising from thermal mismatch between the film
material and the substrate. This can have a significant effect on the produc-
tion yield as well as on the performance and reliability of devices in service.
In past years, an ever increasing trend of miniaturization in semiconductor
and integrated circuit technologies has been observed, stimulating a growing
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
interest to investigate the deformation behavior of such ultra-thin films with

film thicknesses well below 1 µm.
Different inelastic deformation mechanisms are known to operate relax-
ing the internal and external stresses in a thin film. Experiment shows
that for films of thicknesses between approximately 2 and 0.5 µm, the flow
stress increases in inverse proportion to the film thickness (see for example
[341–343]), that is,
1
σY ∼ . (8.5)
hf
Figure 8.8 depicts experimental results of the thin film strength dependence,
here for a copper thin film [51].
This has been attributed to dislocation channelling through the film [71,
72, 345], where a moving threading dislocation leaves behind an interfacial
segment. The relative energetic effort to generate these interfacial dislocations
increases with decreasing film thickness, which explains the higher strength
of thinner films. This model, however, could not completely explain the high
strength of thin films found in experiments [343]. More recent theoretical and
experimental work [51, 96, 346–348] indicates that the strength of thin metal
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
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. 8.8. This transition of the deformation mechanism will be discussed in

more detail in the forthcoming sections.
In-situ transmission electron microscopy observations of the deformation
of such ultra-thin films reveal dislocation motion parallel to the film–substrate
interface [51, 346]. This glide mechanism is unexpected, because in the global
biaxial stress field there is no resolved shear stress on parallel glide planes.
This indicates that there must be a mechanism involving long-range internal
stresses that decay only slowly on the lengthscale of the film thickness. For
sufficiently thin films these internal stresses have a pronounced effect on the
mechanical behavior. It has been proposed that constrained diffusional creep
may be the origin of this novel deformation mechanism [46]. This deformation
mechanism by parallel glide dislocations is not well understood as of today. In
this book, we therefore propose atomistic and continuum studies to investigate
the behavior of such thin films below 400 nm. The final objective is to draw
Fig. 8.10 Distribution of dislocations, result from a discrete dislocation dynamics

simulation analysis. The plot depicts the dislocation distribution and the stress field
σxx at T = 400 K for films with a grain size of 1 µm and various film thicknesses. The
plus and minus symbols denote positive and negative dislocations according to the
sign convention defined in Fig. 8.9a (all dimensions are in µm). Reprinted from Thin
Solid Films, Vol. 479(1–2), L. Nicola, E. Van der Giessen, A. Needleman, Size effects
in polycrystalline thin films analyzed by discrete dislocation plasticity, copyright
c
2005, with permission from Elsevier
a deformation map that summarizes all relevant deformation mechanisms in

submicron thin films.
8.2.1 Constrained Diffusional Creep in Ultra-Thin Metal Films
Similarly as in bulk nanostructured materials, it has also been hypothesized

that grain boundary processes dominate the mechanical properties in ultra-
thin films [46, 51, 346, 350, 351].
In recent theoretical studies of diffusional creep in polycrystalline thin
films deposited on substrates, a new class of defects called the grain boundary
diffusion wedges was predicted [46]. These diffusion wedges are formed by
stress driven mass transport between the free surface of the film and the
grain boundaries during the process of substrate-constrained grain boundary
diffusion. The diffusion wedges feature a crack-like opening displacement, and
due to the strong bonding between film and substrate, a stress concentration
at the root of the grain boundary builds up. This leads to a singular, crack-like
stress field in the film as the grain boundary tractions are relaxed. Because
the material inserted into the grain boundary by diffusion takes the shape of
a wedge, this new class of defects has been referred to as a diffusion wedge
[46,48]. An important implication of the crack-like stress field at the diffusion

wedges is that dislocations with Burgers vector parallel to the interface may
be nucleated at the root of the grain boundary, at the location with highest
shear stress. This is a new dislocation mechanism in thin films that contrasts
to the well-known Mathews–Freund–Nix mechanism of threading dislocation
propagation [71, 72, 345].
For films without relaxation due to diffusional creep, the maximum shear
stress is found on inclined planes, leading to strong driving forces for Mathews–
Freund–Nix threading dislocations. If diffusional creep is active, the largest
shear stresses occur on planes parallel to the film surface. The two modes of
deformation are illustrated in Fig. 8.11.
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
Indeed, results of TEM experiments show that, while threading disloca-

tions dominate in passivated metal films, parallel glide dislocations begin
to dominate in unpassivated copper films with thickness below 400 nm.
Figure 8.12 depicts TEM micrographs of an unpassivated copper film show-
ing parallel glide dislocations and a passivated copper film showing threading
dislocations [51]. The two deformation modes can be distinguished clearly
from the shape of the dislocation networks. Parallel glide dislocations are
curved lines, since the glide in a plane parallel to the film surface. Threading
dislocations are visible as straight lines, as they leave surface steps in these
orientations.
The discovery of parallel glide dislocations [350] provided experimental
support for the constrained diffusional creep model [46]. In turn, constrained
diffusional creep provided the basis for interpretation of certain experimental
results, especially in regard to the mechanisms for the creation and emission
of parallel glide dislocations.
Figure 8.13 summarizes this model including the occurrence of parallel
glide dislocations in three stages: In stage one, material is transported from
the surface into the grain boundary. In stage two, mass transport leads to
the formation of a diffusion wedge, as more and more material flows into and
accumulates in the grain boundary. The continuum model predicts that the
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
Fig. 8.13 Mechanism of constrained diffusional creep in thin films as proposed by

Gao et al. [46]
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.
Fig. 8.14 Development of grain boundary opening ux normalized by a Burgers

vector over time, for the case of a copper film on a rigid substrate. The loading σ0
is chosen such that the opening displacement at the film surface (ζ = 0) at t → ∞
is one Burgers vector [46]
The dislocations “stored” in the grain boundary are a measure of addi-

tional material in the grain boundary. With respect to the lattice distortion
around the diffusion wedge, the dislocations in the grain boundary exemplify a
type of geometrically necessary dislocations [289] that cause nonuniform plas-
tic deformation in the thin film. The eigenvalues measure the rate of decay
of each eigenmode. The results show that the higher eigenmodes decay much
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
λ0 = 8.10 + 30.65hf /d. (8.7)
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.
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.
νfilm νsubs µfilm /µsubs

Cu/rigid 0.32 – 0
Al/epoxy 0.3 0.35 23.08
isotropic – – 1
Table 8.1 Material parameters for calculation of stress intensity factor over the
reduced time
8.2.2 Single Edge Dislocations in Nanoscale Thin Films
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
Fig. 8.17 Geometry and coordinate system of the continuum mechanics model of
constrained diffusional creep
dislocation climb in the grain boundary is a discrete process becomes more

evident. Grain boundary diffusion requires insertion of climb dislocations into
the grain boundary one by one.
Fig. 8.18 Image stress on a single edge dislocation in nanoscale thin film constrained
by a rigid substrate
To investigate the effect, we consider a single edge dislocation climbing in

a grain boundary in an elastic film of thickness hf on a rigid substrate. The
elastic solution of edge dislocations in such a film can be obtained using the
methods described in [47,352]. The geometry, as well as the coordinate system
is shown in Fig. 8.17. In such a geometry, a dislocation placed inside the film
is subject to image forces due to the surface and the film–substrate interface.
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
Another consequence of the geometric confinement of dislocations is that

discrete dislocation effects can lead to quantization of stresses in nano-
structured devices. Figure 8.19 shows the 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 is taken from the analysis
shown in Fig. 8.18. The curves for two and more dislocations in the grain
boundary are estimates. For a given film thickness, for at least one disloca-
tion to be stable inside the film, the critical stress needs to exceed a critical
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].
8.2.3 Rice–Thompson Model for Nucleation of Parallel Glide

Dislocations
To characterize the nucleation condition of parallel glide dislocations, a crite-

rion based on a critical stress intensity factor K PG is used. The motivation is
that the concept of stress intensity factor is commonly used in the mechanics
of materials community and provides a possible link between atomistic and
mesoscopic simulation methods.
The critical value for nucleation of parallel glide dislocations from a dif-
fusion wedge could be thought of as a new material parameter. The stress
intensity factor is defined as
K = lim {[2π(ζ − hf )]s σxx (0, z)} , (8.8)
ζ→hf
where s refers to the stress singularity exponent determined by [354]

α−β α − β2
cos(sπ) − 2 (1 − s)2 + = 0. (8.9)
1−β 1 − β2
It is assumed that the diffusion wedge is located close to a rigid substrate
and the corresponding Dundurs parameters for this case are α = −1 and β =
−0.2647. The Dundurs parameter measures the elastic mismatch of film and
substrate material [47]. The singularity exponent is found to be s ≈ 0.31 for
the material combination considered in the simulations (comparing to s = 0.5
in the case of a homogeneous material). Close to the bimaterial interface, the
stress intensity factor is

∂ux (ζ)
K = A × lim (1 − (ζ/hf ) ) (πhf )s ,
2 s
(8.10)
ζ→hf ∂ζ
where
E (1 − α) 3 − 2s 1 − 2s
A= − . (8.11)
1 − ν 4 sin(πs)
2 1+β 1−β
The stress intensity factor provides an important link between the atom-
istic results and continuum mechanics. To calculate the stress intensity factor
from atomistic data, the atomic displacements of the lattice close to the dif-
fusion wedge are calculated and the stress intensity factor is then determined
using (8.10).
The Peach–Koehler force on a dislocation can be written as Fd = (σ · b) ×

dl, where dl is a dislocation element and σ is the local stress [38]. The variable
bx stands for the√magnitude of the Burgers vector in the x = [110] direction
and bx ≈ 3.615/ 2 × 10−10 m for copper at 0 K. A dislocation is assumed to be
in an equilibrium position when Fd = 0. Following the approach of the Rice–
Thomson model [66], we consider the force balance on a probing dislocation
in the vicinity of a dislocation source to define the nucleation criterion. The
probing dislocation is usually subject to an image force attracting it toward
the source and a force due to applied stress driving it away from the source.
The image force dominates at small distances and the driving force due to
applied stress dominates at large distances. There is thus a critical distance
between the dislocation and the source at which the dislocation is at unstable
equilibrium. Spontaneous nucleation of a dislocation can be assumed to occur
when the unstable equilibrium position is within one Burgers vector of the
source.
Nucleation Mechanism of Parallel Glide Dislocations
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
Nucleation of parallel glide dislocations from a crack in comparison to that

from a diffusion wedge is shown in Fig. 8.20. The crack is treated as by [66],
and forces involved are Fc due to the crack tip stress field, Fimage because
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.
Critical Stress Intensity Factor for Dislocation Nucleation

in Homogeneous Material
Fig. 8.21 Dislocation model for critical stress intensity factor for nucleation of
parallel glide dislocations
It is now assumed that dislocation nucleation at a diffusion wedge is stress

controlled (rather than kinetics controlled) and adopt the first scenario of dis-
location nucleation as described above. This assumption will later be verified
by molecular dynamics simulation results. With this assumption, it is possi-
ble to define a nucleation criterion in terms of a critical stress intensity factor
for both cracks and diffusion wedges. We illustrate the critical condition for
dislocation nucleation in Fig. 8.21. A force balance on a dislocation near a

crack tip leads to the critical stress intensity factor for dislocation nucleation
from a crack
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.
8.2.4 Discussion and Summary
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.
8.3 Atomistic Modeling of Constrained Grain Boundary

Diffusion in a Bicrystal Model
The continuum model [46, 48] of constrained diffusional creep has been very
successful in explaining the origin of the internal stresses in thin films. How-
ever, a continuum viewpoint alone can neither yield a description of the
nucleation process of parallel glide dislocations from the diffusion wedge,
nor incorporate the parallel glide mechanism into the prediction of residual
stresses in a film. Under the guidance of the continuum model [46–48], atom-
istic simulations are an appropriate tool to provide a detailed description of
how parallel glide dislocations are nucleated near a diffusion wedge.
In this chapter, we review large-scale atomistic simulations to study plastic
deformation in submicron thin films on substrates. The simulations reveal that
Fig. 8.22 Disordered intergranular layer at high-energy grain boundary in copper

at elevated temperature (85% of melting temperature)
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.
8.3.1 Introduction and Modeling Procedure
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
Fig. 8.23 Sample geometry of the atomistic simulations of constrained diffusional

creep in a bicrystal model
elevated temperatures, the grain boundary structure of metals may transform

into a liquid like structure with a width of 1–2 nm referred to as “glassy phase.”
Glassy, disordered phases in grain boundaries were found in copper at
homologous temperatures as low as Th ≈ 0.4 [356, 357]. Experimental evi-
dence for glassy intergranular phases was discussed by [358]. Such local phase
transformation at the grain boundary may play a significant role in the plas-
tic properties at elevated temperatures because the different grain boundary
structure has significant influence on the diffusivities [355, 356]. Further dis-
cussion on the topic is found in a recent review article [303]. Figure 8.22 plots
such a disordered intergranular phase obtained in a thin film of copper.
Computational Method and Interatomic Potential
We use a massively parallel classical molecular dynamics code to model the

problem of constrained grain boundary diffusion. A multibody potential [107]
derived using the embedded atom scheme is used, and we integrate the equa-
tions of motion using a velocity verlet algorithm [84, 359]. The studies are
carried out using a microcanonical N V E ensemble.
The potential [107] shows good agreement with ab initio calculations in

terms of elastic properties, stacking fault energies, and excited phases of cop-
per [107]. We determine the melting point to be close to the experimental
value of Tm = 1,360 K with this potential.
Modeling of a Thin Film on Substrate

The simulation sample is built by creating a < 111 > FCC lattice, rotating
one-half counterclockwise and cutting out two rectangular pieces. This proce-
dure leads to an asymmetric < 111 > tilt grain boundary with approximately
α ≈ 7◦ mismatch. Note that the choice of this geometry and the tilt angle is
motivated by recent experimental results [51].
The simulation geometry is depicted in Fig. 8.23. In some simulations, we
further rotate the tilted grain counterclockwise around its [112] axis, creating
a higher energy grain boundary.
The structure is periodic in the y-direction. We impose a homogeneous
strain throughout the sample to account for thermal expansion [107]. The
boundary and substrate atoms are chosen such that atoms inside the film do
not sense the existence of the surface. The atoms of the boundary and the
substrate are held pinned at a prescribed location by introducing an additional
term in the potential energy of the form
1
φp = k0 rx2 d −xc , (8.14)
2
where xd is the prescribed location of an atom and xc is its current position.
The expression rxd −xc =| xd − xc | stands for the radius of separation of the
desired and present location, while k0 is a harmonic spring constant which
is chosen k0 = 20 in reduced atomic units. The locations of atoms in the
substrate are prescribed only in the x- and z-direction, and the y-direction
is left unconstrained. The atoms in the boundary are only constrained in the
x-direction, so that the boundary is allowed to relax in the y- and z-direction.
In terms of a continuum mechanics interpretation of the simulation cell,
this resembles a plane strain case with a thin compliant (copper) film on a
rigid substrate, with no sliding and no diffusion at the film–substrate inter-
face. The choice of the rigid substrate is partly motivated by the finding that
the stress intensity factor of a crack is reached faster in the case of a rigid
substrate than in the isotropic case or when the film is attached to a compliant
substrate (see Fig. 8.16 and associated discussion). After the “raw” sample is
created, a global energy minimization scheme is applied to relax the struc-
ture. Subsequently, the sample is heated up to an elevated temperature and
annealed for a longer time so that the grain boundary structure relaxes and
takes its equilibrium configuration. The virial stresses [239] is relaxed to zero
in this initial configuration before loading is applied.
To make the diffusive processes accessible to the molecular dynamics
timescale, the simulations are carried out at elevated temperatures. The sim-
ulations to investigate dislocation nucleation in conjunction with diffusional
creep are performed at a homologous temperature of Th ≈ 0.8 and Th ≈ 0.9.

At elevated temperature, the grain boundary exhibits a highly confined glassy
intergranular phase of less than 1-nm width, in accordance with [356].
After annealing we proceed with applying a lateral strain. The prescribed
positions xd are calculated according to a homogeneous strain throughout the
simulation sample. We use a time step of ∆t ≈ 3 × 10−15 s for integration.
The strain rate is on the order of 107 s−1 corresponding to approximately
1% strain per nanosecond. The strain rate is adjusted during the simulation
such that the stress in the film remains low to avoid nucleation of dislocations
on inclined slip planes, similar to the procedure adopted by [68]. The only
deformation mechanism allowed is diffusional creep in the grain boundary.
Whenever activation of a different mechanism such as threading dislocations
is observed, the simulation is restarted at a lower stress and the strain rate is
lowered. This procedure has proven to allow more time for diffusive processes
and effectively shut down competing mechanisms. In the simulation with Th ≈
0.8, the grain boundary width remains less than 1 nm and increases slightly
at higher temperatures when the sample is loaded. In the simulation with
Th ≈ 0.9 the grain boundary width is further increased and is found to be
around 1–2 nm.
The systems contain more than 1,000,000 particles, which is a signifi-
cant number since simulations are carried out over several nanoseconds. The
film thickness ranges from 4.5 to 35 nm, the latter value becoming compa-
rable to experimental investigations where films between 35 and 50 nm were
investigated [346, 350].
8.3.2 Formation of the Diffusion Wedge
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.
Crack-Like Displacement Near 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
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
[x, y, z]new = [10 · x, y, z]orig . (8.15)
This visualization technique is applied throughout this section. We highlight

the additional half planes of atoms close to the grain boundary. The continuous
dark line corresponds to the continuum mechanics solution in the long-time
limit t → ∞ discussed earlier. The results suggest that the displacement field
near the diffusion wedge approaches the continuum mechanics solution.
Diffusive Displacement of Atoms in the Grain Boundary
Fig. 8.25 Diffusional flow of material into the grain boundary. Atoms that diffused
into the grain boundary are highlighted
To illustrate diffusional motion of atoms in the grain boundary, each atom

with diffusive displacement δz larger than a few Burgers vectors is colored
according to its displacement. Figure 8.25 plots these fields for several instants
in time. Diffusion leads to significant surface grooving, with groove depths
up to several nm. One can clearly identify the wedge-shape of the diffused
atoms. The atomistic simulations show that atoms inserted into the grain
boundary instantaneously crystallize, rendering the structure of the grain
boundary invariant (this was observed for temperatures below 1,150 K; at
higher temperatures, the width of the grain boundary increases slightly).
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.
8.3.3 Development of the Crack-Like Stress Field and Nucleation

of Parallel Glide Dislocations
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.
Temperature T (in K) hf (nm) Stress intensity factor K PG (in MPa × ms )

Crack
300 27.2 4.95
Diffusion wedge
1150 27.2 11.91
1250 27.2 11.35
1250 34.2 11.23
Table 8.2 Critical stress intensity factor K PG for nucleation of parallel glide
dislocations under various conditions, for both a diffusion wedge and a crack
It is observed that nucleation of parallel glide dislocations depends on the film

thickness. In the present quasi-two-dimensional setup with rigid boundaries, it
is found that dislocations from the boundaries are nucleated when very large
strains are applied, thus imposing a condition on the minimum thickness for
nucleation of parallel glide dislocations.
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
Dislocation nucleation at a diffusion wedge can be divided into different

stages shown in Fig. 8.26. After the critical stress intensity factor is achieved,
a dislocation dipole is formed. One end of the dipole is pinned in the grain
boundary, while the dislocation at the other end of the dipole slides away
from the grain boundary. Subsequently, the pinned dislocation is annihilated
or “dissolves into” the grain boundary, while the dislocation at the right end
of the dipole begins to move away from the nucleation site. As usually found
in FCC metals, the dislocation is decomposed into two Shockley partials. The
parallel glide dislocation glides on a slip plane parallel to the plane of the film
at a distance of a few Burgers vectors above the film–substrate interface (and
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
When classical mechanisms of plastic deformation based on the creation and

motion of dislocations are severely hindered in thin films on substrates, con-
strained diffusional creep becomes a major mechanism for stress relaxation,
leading to the formation of a new class of defects called the grain boundary
diffusion wedge. Large-scale atomistic simulations are performed to investigate
the properties of such diffusion wedges. It was shown by atomistic simulations
that material is indeed transported from the surface into grain boundaries
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.
Theoretical, Experimental, and Simulation Results
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,
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.
Diffusion Wedge vs. Crack
We discuss some of the common and distinct properties observed in atomistic

simulations for the two kinds of defects (crack vs. diffusion wedge), both are
assumed to lie along the grain boundary under elevated temperatures. For a
crack, it is observed:
• As the applied stress σ0 is increased, the normal stress σxx along the grain
boundary remains zero throughout the film thickness, in consistency with
the traction-free crack condition.
• The loading rate for dislocation nucleation can be much higher than in the
case of diffusional creep; there is no rate limiting factor.
• Dislocation nucleation occurs at relatively small stress intensity factor.
• Dislocation nucleation starts with an incipient dislocation close to the
crack tip.
In contrast, for a diffusion wedge, it is observed:
• The loading rate must be slow enough to allow for diffusion as a dominant
relaxation mechanism. Otherwise dislocation activities on inclined planes
are observed instead of grain boundary diffusion.
• The stresses in the film are determined by the competition of processes
causing stresses to be generated and inelastic deformation mechanisms
that allow the stresses to be relaxed.
• The nucleation process proceeds much slower, because in order to nucleate
a parallel glide dislocation, dislocation climb in the grain boundary has to
take place to annihilate part of the newly created dipole. On atomistic
timescales, nucleation is an extremely slow process (for a crack, nucleation
is very fast).
• Dislocation nucleation starts when the stress intensity factor is sufficiently
large to create a dislocation dipole near the diffusion wedge.
• There exists a minimal thickness for parallel glide dislocation nucleation.
If the film is very thin, the applied stress reaches the cohesive strength of
the material before the critical stress intensity factor K PG for dislocation
nucleation is reached.
The two defects have major differences in the timescale associated with
creation of dislocations. A crack is a ready source for dislocations, while a dif-
fusion wedge has an intrinsic characteristic time associated with dislocation
climb. We finally note that no difference in the mechanism of parallel glide dis-
location nucleation is observed at different temperatures. We propose further
investigations on discrete dislocation effects at the nanoscale. In particular, it
is important to develop continuum level solutions for dislocation nucleation in
the spirit of Rice’s analysis based on Peierls concept [30, 360]. It should also
be interesting to study thin films creep using mesoscopic methods such as
discrete dislocation dynamics. Results of the molecular dynamics simulations
can be used as input parameter in a multiscale modeling procedure.
8.3.5 Summary
This review illustrates how large-scale atomistic simulations can be used to

study constrained diffusional creep in thin films deposited on substrates. The
formation of diffusion wedges has been confirmed using atomistic simulations.
In agreement with theory and experiment, this illustrates that the flow of
matter from the film surface into grain boundaries represents an important
mechanism of plasticity in submicron thin films. It has been verified that the
diffusion wedge exhibits crack-like stress field at the atomistic level, and this
mechanism occurs even if the background stress in the film is insufficient to
create dislocations.
The results of molecular dynamics simulations enabled us to calculate a
critical stress intensity factor for nucleation of parallel glide dislocations from
the diffusion wedge. The critical stress intensity is found to be independent of
the film thickness and does not significantly change in the temperature range
of the investigation (from Th = 0.9 to Th = 0.8).
The most important result of these simulations is that when grain bound-
ary diffusion is active, the grain boundary can be treated as a crack in a first
approximation.
In the present section, we have only studied a two-dimensional geometry. In
the following chapters, we will show that the condition of traction relaxation
along grain boundaries also has dramatic consequences on the dislocation
mechanisms in polycrystalline thin films.
8.4 Dislocation Nucleation from Grain Triple Junction

In this chapter, we focus on the details of dislocation nucleation close to a
triple junction between three grains misoriented with respect to one another.
The simulation geometry is shown in Fig. 8.28.
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
In contrast to the simple bicrystal geometry used in numerical studies

discussed in Sect. 8.3, experiments are carried out in polycrystalline thin films
[51, 346]. The first goal of this section and Sect. 8.5 is hence to extend the
quasi-two-dimensional studies of constrained diffusional creep to more realistic
polycrystalline microstructures.
We will investigate plasticity of thin films with traction relaxation along
the grain boundaries due to constrained diffusional creep. The continuum
model and the quasi-two-dimensional geometry of previous atomistic simu-
lation of plasticity in thin films could not provide a clear understanding of
dislocation nucleation processes from different types of grain boundaries. Yet,
experiments show a high selectivity of dislocation nucleation from different
grain boundaries [51, 350]. Understanding the features of the grain bound-
aries as sources for dislocation nucleation is critically important to form a
clear picture of thin film plasticity. Therefore, an important objective of this
section is to study the details of the dislocation nucleation process near the
grain boundary–substrate interface. We show that the grain boundary struc-
ture indeed has a significant influence on the dislocation nucleation process.
Another interesting result is that the role of partial dislocations is important
in very thin films with very small grain diameters.
8.4.1 Atomistic Modeling of the Grain Triple Junction
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
the high-energy grain boundaries is more homogeneously distributed. After

creation of the sample, the structure is annealed for a few picoseconds and
then relaxed for a few thousand integration steps using an energy minimization
scheme.
Boundary Conditions and Integration Scheme
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.
8.4.2 Atomistic Simulation Results
We will investigate the details of parallel glide dislocation nucleation process

near the grain boundary–substrate interface. Dislocation mechanisms associ-
ated with grain boundary cracks will be compared and related to experimental
results.
Nucleation of Parallel Glide Dislocations from a Grain Triple

Junction
In this section, we focus on the details of dislocation nucleation close to a

triple junction between three grains misoriented with respect to one another.
As soon as a threshold stress is overcome during loading, the generation
of parallel glide dislocations from the grain boundaries is initiated. Snapshots
of this processes are shown in Fig. 8.29.
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
We begin with a description of the nucleation process from the low-energy

grain boundary between grains 1 and 2. This boundary is composed of an array
of misfit grain boundary dislocations which serve simultaneously as multiple
nucleation sites for new dislocations. The nucleation sites are therefore not
necessarily located close to the triple junction, the region of largest shear
stresses. This observation could be reproduced in different geometries. A small
number of incipient dislocations grow along the low-energy grain boundary
and coalesce to form dislocation half-loops. This mechanism is also visualized
schematically in Fig. 8.30. In Fig. 8.31, we show that deformation twinning
occurs due to repeated nucleation of partial dislocations with the same Burgers
vector.
Fig. 8.31 Deformation twinning by repeated nucleation of partial dislocations.

Repeated slip of partial dislocations leads to generation of a twin grain boundary
Jog Dragging
An interesting observation in the simulations is that some dislocations are

strongly bowed at defect junctions. Dislocation junctions obstructing further
glide motion are highlighted in Fig. 8.32. The reason for this effect is that the
glide planes of the incipient half-loops of partial dislocations are different, but
have the same Burgers vector. Using the slip vector approach proposed by
Zimmerman and coworkers [147], we have verified that the Burgers vector of
the dislocations nucleated in each grain are indeed identical.
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
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
The mechanism observed in the simulations is similar. The difference is

that no dislocation cutting process occurs, but instead the jog in the dislo-
cation line develops due to nucleation of incipient dislocations on different
glide planes. It is observed that the sessile component of the jog is rather
small and is only a fraction of the partial Burgers vector. Therefore, not a
complete point defect is generated but only a trail of “partial point defects.”
Trails of “partial point defects” have recently also been observed in large-scale
computer simulations of work-hardening in ductile materials [14]. As shown
by calculations in [14], the dragging force of the partial point defects is esti-
mated to be approximately 20% of the dragging force exerted by generation
of complete vacancy tubes. If these defects appear in large numbers, the effect
on dislocation motion can be significant.
High-Energy vs. Low-Energy Grain Boundaries

In the case of high-energy grain boundaries (as between grains 2 and 3) with a
more homogeneous structure, nucleation of parallel glide dislocations is found
to occur preferably at the triple junction. The process proceeds with an incip-
ient dislocation growing until the second partial is emitted. The parallel glide
dislocations often have semi-circular shapes as observed in early stages of
dislocation nucleation in experiment [346, 350].
In contrast to the low-energy grain boundary where misfit dislocations
serve as nucleation sites for new dislocations, the triple junction acts as the
main nucleation source at high-energy grain boundaries.
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].
8.5 Atomistic Modeling of Plasticity of Polycrystalline

Thin Films
Atomistic modeling of plasticity in polycrystalline thin films is just at its
beginning. Few studies of such systems have been reported in the literature.
Recently, atomistic simulations of two-dimensional systems were reported by

Shen [361]. In this study, the increase in yield strength was investigated and
nucleation and motion of threading dislocations was in the focus. However,
the model did not contain grain boundaries despite the fact that grain bound-
aries can serve as fertile sources for dislocations. In contrast to this simplistic
model, we propose a three-dimensional model of thin films with a more realis-
tic microstructure. The model studied in this section is a polycrystalline thin
film consisting of hexagonal shaped grains as shown in Fig. 8.34. The choice
of this geometry is motivated by the grain microstructure found in experi-
ments [346, 350]. An advantage of this model over the geometry studies in
Sect. 8.4 is that fully periodic boundary conditions in the x- and y-direction
can be assumed.
Here we will focus on dislocation nucleation and motion from grain bound-
aries and a crack–grain boundary interface. One of the important objectives
will be to study the effect of grain boundary traction relaxation by diffu-
sional creep on the dislocation mechanism that operate in the film. Further
studies will be focused on the details of dislocation nucleation from different
type of grain boundaries. As known from Sect. 8.4, the structure of the grain
boundaries has strong influence on the nucleation of dislocations.
The plan of this section is summarized as follows. After presenting details
about the atomistic modeling procedure, we will continue with a discussion
of the results of several large-scale atomistic studies comprising of up to 35
million particles. Even for today’s supercomputers, this represents a signifi-
cant system size. We will show that grain boundary relaxation by diffusional
creep gives rise to dominance of parallel glide dislocations, in accordance with
experiment. In contrast, if grain boundary tractions are not relaxed, threading
dislocations dominate plasticity. We show that low-energy grain boundaries
are more fertile sources for plasticity than more homogeneous high-energy
grain boundaries. This hypothesis is further supported by a set of atomistic
simulations of bulk nanocrystalline copper.
8.5.1 Atomistic Modeling of Polycrystalline Thin Films
As indicated in Fig. 8.34, cracked grain boundaries with traction-free surfaces

(along zc < z < hf , where zc ≈ 1.5 nm) are used to mimic the existence of
diffusion wedges in some of the grain boundaries. Grain 1 is in the reference
configuration ([110] in the x-direction, [112] in the y-direction). Grain 2 is
rotated counterclockwise by 7.4◦ , grain 3 is rotated by 35◦ , and grain 4 is
rotated by 21.8◦ with respect to grain 1. The model contains up to 35 million
particles. With this procedure, a low-energy grain boundary is constructed
between grains 3 and 4. After creation of the sample, the structure is relaxed
for a few thousand integration steps using an energy minimization scheme.
Figure 8.35 shows the atomistic model of the polycrystalline thin film. Only
surfaces and grain boundaries are shown (explanation of visualization in figure
caption).
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.
8.5.2 Atomistic Simulation Results
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
change should significantly alters the dislocation microstructure that devel-

ops in the film under mechanical deformation. While threading dislocations
dominate plasticity in films where grain boundary diffusion is shut down, in
films where grain boundary diffusion is active parallel glide dislocations are
expected to dominate. Indeed, since the continuum model was proposed [46],
an experimental group has reported the observation of parallel glide dislo-
cations in copper films with thicknesses below 400 nm [51]. The researchers
concluded that grain boundary traction relaxation by diffusional creep leads
to change in the deformation field near the crack tip. Experimental results sug-
gest that threading and parallel glide dislocations are competing mechanisms
[51, 346, 350] In submicron, uncapped thin films on substrates.
In this section, we want to probe this hypothesis by large-scale atomistic
simulations. Atomistic modeling of thin film plasticity at the nanoscale pro-
vides an ideal tool to study such competing mechanisms and to determine
conditions under which they are active.
It is known that grain boundaries are important sources for dislocations
in nanostructured materials. We illustrate that the structure of the grain
boundaries has significant influence on the motion of dislocations into the
grain interior. Furthermore, we find that the role of partial dislocations seems
to be increasingly important as the grain size approaches nanoscale.
Threading Dislocations
We start with the polycrystalline sample without relaxation of tractions along

the grain boundaries, corresponding to the case when grain boundary diffusion
is not active. The simulation results are depicted in Figs. 8.36 and 8.37. In
this case, the dominating inelastic deformation mechanism is clearly glide
of threading dislocations on inclined glide planes. No dislocations on glide
planes parallel to the film surface are observed as expected, because there is
no resolved shear stress and thus no driving force for dislocation nucleation
on parallel glide planes.
Figure 8.36a reveals a complex dislocation structure in the interior of the
film. The dislocation structure is analyzed with the centrosymmetry tech-
nique. Figure 8.36b shows a more detailed magnified view of a section of the
film. Threading dislocations are observed to leave behind interfacial disloca-
tion segments at the film–substrate interface and atomic steps at the film
surface. Figure 8.37 shows snapshots of a top view of the film surface at differ-
ent times, including a magnified view of the surface at snapshot 4 in Fig. 8.38.
The surface steps emanate from the grain boundaries, suggesting that disloca-
tions are nucleated at the grain boundary–surface interface. From the direction
of the surface steps it is evident that different glide planes are activated.
From the number of surface steps created during plastic deformation, it
seems that dislocation motion concentrates in grains adjacent to low-energy
grain boundaries which apparently provide more fertile sources for dislocation
nucleation (within grains 1 and 2). This can also be verified in Fig. 8.36b. The
Fig. 8.36 Nucleation of threading dislocations in a polycrystalline thin film. Sub-

plot (a) shows a view into the interior, illustrating how threading dislocations glide
by leaving an interfacial segment. Subplot (b) shows a top view into the grain where
the surface is not shown. The plot reveals that the dislocation density is much higher
in grains 3 and 4
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
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. 8.39 Sequence of a nucleation of a threading dislocation, view at an inclined

angle from the film surface. Threading dislocations are preferably nucleated at the
grain boundary–surface interface and half-loops grow into the film until they reach
the substrate. Due to the constraint by the substrate, threading dislocations leave
an interfacial segment
dislocation nucleation and with experimental results [341–343]. The threading

dislocations intersect the surface at an angle of 90◦ [38].
Parallel Glide Dislocations
In the following, some of the grain boundaries are treated as traction-free

cracks as shown in Fig. 8.34. In Fig. 8.40, we show several snapshots of the
dislocation structure. Parallel glide dislocations are generated close to the
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
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
develops on a timescale of several picoseconds after the first dislocation

nucleation.
Figure 8.41 shows the simulation results for a larger grain size. The section
shown has dimensions of approximately 300 nm × 400 nm; the film thickness
is hf ≈ 15 nm. The grain diameter in the x-direction is approximately dx ≈
180 nm, about four times larger than in Fig. 8.40 while the film thickness is
kept constant at hf ≈ 15 nm.
More dislocations are observed to nucleate than in Fig. 8.40, indicating
that more dislocations “fit” into the larger grain, and consequently, a more
complex dislocation microstructure develops. As the laterally applied strain is
continuously increased, the first dislocations to be nucleated are occasionally
complete dislocations, while the following dislocations are often pure partial
dislocations. Figure 8.42 shows a view of the surface of the results shown in
snapshot 2 of Fig. 8.41, revealing surface steps generated from the motion of
threading dislocations. Even when the traction of some of the grain boundaries
are relaxed, threading dislocations occur. The figure shows that threading
dislocations are predominantly nucleated at the junction between traction-free
grain boundaries and normal grain boundaries where traction is not relaxed.
It is observed that dislocations cannot glide as easily along the low-energy
grain boundaries as along the more homogeneous high-energy grain bound-
aries between grains 1 and 2. This can be verified in Fig. 8.40a. While an
extended dislocation (marked by “PG”) in grain 1 is almost a straight line,
all dislocations in grains 3 and 4 are strongly curved.
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.
8.5.3 Plasticity of Nanocrystalline Bulk Materials with Twin

Lamella
In the preceding sections, the role of low-energy vs. high-energy grain bound-
ary was discussed. The importance of this concept is further underlined by
the studies reported in this section. Here we focus on polycrystalline bulk
copper, where the grain size is on the order of several nanometers to tens of
nanometers.
We consider a polycrystalline microstructure with hexagonal grains, but
with different grain orientations as in the previous case (see Fig. 8.44 for
details). To further study the effect of geometric confinement on plasticity, we
introduce a subnanostructure in the grains. This subnanostructure is estab-
lished by assuming twin grain boundaries forming very thin twin lamella.
Such microstructure can be produced experimentally in copper [362]. With
this model, two main objectives are pursued:
1. We show that in bulk nanostructured materials, the type of the grain
boundary plays a very important role for dislocation nucleation, as it was
found for thin films.
2. We show that the subnanostructure composed of twin grain boundaries
provides a very effective barrier for dislocation motion and therefore leads
to a very “strong” nanostructured material.
Modeling
To underline the first point regarding dislocation nucleation, we consider two

samples in our simulation study. The first sample (i) has the same grain misori-
entations as in Sect. 8.5.2 (and therefore features a low-energy grain boundary
between grains 3 and 4) and we construct a second sample (ii) where all grain
boundaries are of the same type. If the proposed concept is correct that dis-
location nucleation occurs predominantly from low-energy grain boundaries,
the dislocation density in sample (i) should be higher in grains 3 and 4, and
should be comparable in all grains in sample (ii).
Fig. 8.44 Nanostructured material with twin grain boundary nanosubstructure.

The light gray lines inside the grains refer to the intragrain twin grain boundaries.
The thickness of the twin lamella is denoted by dT
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
misorientation as in the study described above. We perform the simulation for

two different lamella sizes dT . The results are shown in Fig. 8.45.
It is observed that dislocations are generated exclusively from the low-
energy grain boundaries between grains 3 and 4. This is in agreement with
the results of the polycrystalline thin films. The fact that we use a different
grain orientation in this study with different boundary conditions suggests
that the nucleation conditions discussed previously is a more general concept.
The results indicate that the twin grain boundaries are an effective barrier for
further dislocation motion, since it is found that dislocation pileups at the twin
grain boundaries. An important consequence is that the thinner the lamella
structure (that is, a small dT ), the less plasticity can be transmitted via the
motion of dislocations. This indicates that grains with a nanosubstructure of
twin grain boundaries could serve as an effective hardening mechanism for
materials.
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
Burgers vector annihilate. Further, dislocations cross-slip (see highlighted

region in the center of Fig. 8.46b, region i.) in regions with high dislocation
densities. The activated primary and secondary glide planes are highlighted
in the plot. The primary glide planes are parallel to the twin grain boundaries
so that dislocation glide is not restricted. In contrast, once dislocations have
cross-slipped to the secondary glide plane their motion is restricted due to the
twin grain boundaries (see Fig. 8.46b, ii.). Figure 8.46b, iii. shows intersection
of dislocations. A defect is left at the intersection of the stacking fault planes.
As in the previous studies of nanostructured materials [303], we also
observe that partial dislocations dominate plasticity. Dominance of partial
dislocations is verified by the fact that dislocations leave behind a stacking
fault.
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
8.5.4 Modeling of Constrained Diffusional Creep in Polycrystalline

Films
We have also modeled constrained grain boundary diffusion in polycrystalline
thin films, thus extending the two-dimensional studies discussed in Chap. 8.3
to the three-dimensional case. We apply biaxial loading rate on the order
of 1% strain per nanosecond. The temperature is, as in the two-dimensional
studies, chosen around 90% of the melting temperature.
High-energy grain boundaries transform into liquid-like intergranular layers,
while low-energy grain boundaries establish as arrays of misfit dislocations.
In the following sections we will show that the basic mechanism of parallel
glide dislocation nucleation is identical to the results observed in the two-
dimensional case. The simulations provide direct evidence that the diffusivities
depend on the grain boundary structure.
Constrained Grain Boundary Diffusion and Dependence on Grain

Boundary Structure
We model a film of thickness hf ≈ 11 nm with a grain diameter of about 22 nm
in the x-direction. The simulation sample is constructed such that we have
high-energy as well as low-energy grain boundaries. This is motivated by the
goal to investigate the effect of grain boundary structure on the deformation
mechanisms. Grain 1 is completely surrounded by high-energy grain bound-
aries and the other grains feature low-energy grain boundaries (grain 1 is in
its reference configuration, grain 2 is rotated by 35.4◦ , grain 3 by 44.7◦ , and
grain 4 by 53.4◦ ). It is observed that, in agreement with the predictions in
the literature [303], high-energy grain boundaries provide rather fast diffusion
paths in contrast to low-energy grain boundaries. This strongly underlines the
notion that the grain boundary structure plays a major role in determining
the resulting deformation mechanism, and that it therefore needs to be taken
into account when diffusivities are determined.
Formation of grain boundary diffusion wedges is accompanied by surface
grooving at the grain boundary interface. Therefore, the surface height profile
provides a reliable indication of diffusive activities in the grain. Fig. 8.47a plots
the surface profile of a polycrystalline sample in early stages of the simulation.
The observation of surface grooves is in agreement with recent experimental
reports [351].
Compared with all other diffusion paths, grain triple junctions provide
the fastest paths for diffusion. This is verified in the simulation results since
at grain triple junctions, the surface grooves are deepest. Similar as in the
two-dimensional case, it is observed that the grain boundaries become curved
along the z-direction.
Nucleation of Parallel Glide Dislocations

According to the hypothesis by continuum theory [46], parallel glide dislo-
cations should only be nucleated along grain boundaries whose tractions are
Fig. 8.47 Modeling of constrained diffusional creep in polycrystalline samples;

nucleation of threading vs. parallel glide dislocations [49]. The blowup in panel
(c) shows an energy analysis of the dislocation structure and visualizes a parallel
glide dislocation nucleated from a grain boundary diffusion wedge. The surface steps
indicate that threading dislocations have moved through the grain and no threading
dislocations exist in grain 1. The dark lines show the network of parallel glide dislo-
cations in grain 1 (in other grains we also find parallel glide dislocations in snapshot
(d) but they are not shown)
relaxed by diffusional creep. Since high-energy grain boundaries are predom-

inant paths for diffusion, in grains neighboring high-angle grain boundaries
parallel glide dislocations should dominate plasticity. In other grains, where
tractions along the grain boundaries are not relaxed threading dislocations
should dominate.
This prediction is verified by the atomistic simulations. Nucleation of par-
allel glide dislocation only occurs in grains that are completely surrounded by
high-energy grain boundaries. In the computational sample, there is no other
dislocation activity than parallel glide dislocations after diffusion has pro-
ceeded sufficiently long. In other grains where little grain boundary relaxation
is possible by diffusional creep, threading dislocations are easily nucleated.
Predominant nucleation site are, in agreement with previous results, misfit
dislocations at the grain boundary. The observation of threading dislocations

in other grains is consistent with the studies where some grain boundaries
were assumed traction free (see Fig. 8.41).
This result of the study of constrained grain boundary diffusion in poly-
crystalline films is documented in Fig. 8.47, where the first two snapshots in
Fig. 8.47a, b show the initial stages of diffusive deformation. The nucleated
parallel glide dislocation is shown in Fig. 8.47c as a black line, and additional
parallel glide dislocations are nucleated subsequently as shown in Fig. 8.47d.
Its shape was determined using the energy filterning method shown in the
blow-up of Fig. 8.47c. Additional analysis was performed based on geomet-
rical methods identical to those applied in Sect. 8.3. After the first parallel
glide dislocation is nucleated, additional dislocations appear as the loading is
increased and the dislocations form a network that is similar to the results
shown in Fig. 8.40. It is noted that even at higher applied loads, there are
exclusively parallel glide dislocations in grain 1 as shown in Fig. 8.40d. In
other grains, we observe parallel glide dislocations at later stages in addition
to threading dislocations.
The most important result of this section is that constrained diffusional
creep and nucleation of parallel glide dislocations can also be modeled in a
polycrystalline model. Therefore, this result enables one to perform a direct
comparison with experimental results. Figure 8.12a (discussed earlier) shows
the structure of parallel glide dislocations. Figure 8.48 depicts a time sequence
of experimental snapshots that illustrates the subsequent emission of parallel
glide dislocations.
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).
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
The simulations show that relaxation of grain boundary tractions changes

the dislocation microstructure and triggers completely different stress relax-
ation mechanisms in thin films. We have also investigated nucleation of parallel
glide dislocations from diffusion wedges using the quasicontinuum method.
The results of this simulation are discussed in Sect. 5.1 and results were shown
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.
8.5.6 Summary: Results of Modeling of Thin Films
We have reviewed deformation mechanisms in submicron thin copper films,

focusing on the competition between constrained diffusional creep with subse-
quent parallel glide dislocation nucleation and the nucleation of threading dis-
locations. Together with experimental observations [51], the results of atom-
istic modeling of constrained grain boundary diffusion provide evidence that
it is an important deformation mechanism in very thin uncapped copper films.
The observation of parallel glide dislocations nucleated at grain boundary dif-

fusion wedges (shown in Fig. 8.26) agrees with experimental investigations of
deformation of ultra-thin copper films [51,346,350]. The results also agree well
with the predictions by recently developed continuum mechanics theory [46,
48], thus closing the experiment–theory–simulation linkage. The most impor-
tant result is that once other stress relaxation mechanisms (e.g., by dislocation
motion) are shut down, diffusional flow of matter along grain boundaries
dominate the mechanical properties of thin metal films on substrates.
An important contribution of this work is the atomistic study of con-
strained diffusional creep and subsequent parallel glide dislocation nucleation.
We find that grain boundary diffusion could be modeled by classical molecular
dynamics, in agreement with reports in previous studies [68, 303]. However,
modeling of grain boundary diffusion is difficult with classical molecular
dynamics and is only possible under large stress and high temperatures. In the
simulations, stress of magnitude of several GPa is applied and diffusion is stud-
ied at elevated temperatures on the order of 90% of the melting temperature
of copper.
Modeling at the atomic scale helped to identify the key mechanism of dislo-
cation nucleation from a diffusion wedge. The simulations allowed to establish
a detailed understanding of the dislocation nucleation process close to a dif-
fusion wedge. This led to the definition of a critical stress intensity factor for
parallel glide dislocation nucleation. The critical value for dislocation nucle-
ation from a diffusion wedge is twice as large compared to a crack. This was
explained by the difference in force balance on the incipient dislocation: For
dislocation nucleation near a diffusion wedge, a dislocation dipole needs to
be generated where the dislocations are one Burgers vectors apart. In the
case of a crack, the incipient dislocation senses the image force of the sur-
face corresponding to a “virtual” dislocation dipole where the dislocations are
two Burgers vectors apart (see Fig. 8.21 for a schematic visualization of these
considerations). The atomic simulation results support this theoretical model
since the ratio of critical stress intensity factor of a diffusion wedge to a crack
is found to be around two.
As discussed in Sect. 8.3.4, a crack and a diffusion wedge have major dif-
ferences in the timescale associated with creation of dislocations. A crack is
a ready source for dislocations, while a diffusion wedge has an intrinsic char-
acteristic time associated with dislocation climb. In the long-time limit on
the order of a characteristic time τ , the diffusion wedge behaves as a crack
in agreement with theoretical considerations [46]. The change of maximum
resolved shear stress due to climb of edge dislocations into the grain boundary
is schematically visualized in Fig. 8.11.
Atomistic studies of polycrystalline thin films helped to clarify the nucle-
ation mechanisms of dislocations from different types of grain boundaries. We
find that low-energy grain boundaries provide more fertile sources for dis-
locations than high-energy grain boundaries. This concept helps to explain
why the dislocation density is several times higher in grains neighboring to

low-energy grain boundaries in the simulations.
We observe that mostly partial dislocations are nucleated from the grain
boundaries in nanostructured thin films. This contradicts the classical theories
of deformation [38] where it is predicted that complete dislocations dominate,
but it is in agreement with other studies of deformation of nanostructured
bulk materials [153, 154, 304, 328, 332]. A detailed investigation of the dislo-
cation nucleation process from low-energy and high-energy grain boundaries
provided insight into the atomic mechanisms of this process. One of the impor-
tant observations is that dislocation half-loops are generated on different glide
planes causing formation of sessile jogs in the dislocation line that generate
point defects as they move.
Additional simulations of constrained grain boundary diffusion in polycrys-
talline films suggested that there is a strong dependence of the diffusivities
on the grain boundary structure. We could also show that parallel glide
dislocations are nucleated along grain boundaries with highest diffusivities.
This result shows that diffusion and nucleation of parallel glide dislocations
are highly coupled, thus supporting the theoretical understanding of the
process [46].
8.6 Use of Atomistic Simulation Results in Hierarchical

Multiscale Modeling
Classical molecular dynamics is rather constrained with respect to the acces-
sible timescale. One of the drawbacks is that parameter studies with varying
film thickness are difficult, if not impossible to carry out with today’s comput-
ers (maximum film thicknesses that can be reached are on the order of 50–100
nm). Therefore, hierarchical multiscale modeling may serve as a useful tool
to reach higher lengthscales and timescales. Such studies are now being fre-
quently applied in materials modeling [179, 180, 348]. Here we briefly describe
how a hierarchical coupling of molecular dynamics with mesoscopic simula-
tions was achieved and review the results of a mesoscopic study that was
carried out based on the molecular dynamics results discussed in this book.
Mesoscopic methods must rely on phenomenological input parameters or
rules. The most important contribution by the molecular dynamics simu-
lations was the concept of a critical stress intensity factor which could be
translated into a discrete dislocation formulation of diffusional creep [50]. The
mesoscopic model reported in [50] follows the well-known discrete dislocation
models in two and three dimension described in the literature (see for exam-
ple [96, 181, 182, 348] for thin film plasticity). In such models, dislocations are
considered sources of stress and strain in a linear elastic continuum.
The proposed discrete dislocation model for diffusional creep in ultra-thin
films proved to be capable of predicting experimentally measurable quantities
like the flow stress [50]. The two-dimensional model suggests the existence of
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
to other materials phenomena for which fully atomistic simulations are not yet
feasible.
8.6.1 Mechanisms of Plastic Deformation of Ultra-thin Uncapped

Copper Films
The results of atomistic simulations, experiments, continuum modeling as well

as mesoscopic modeling have advanced to a level that they allow drawing gen-
eral conclusions about the deformation mechanism in ultra-thin films. Here we
summarize the main results in a deformation map of thin films. The objective
of this is to provide a clear overview over the different deformation mechanisms
in ultra-thin films.
8.6.2 Deformation Map of Thin Films
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
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.
8.6.3 Yield Stress in Ultra-Thin Copper Films
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)
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].
8.6.4 The Role of Interfaces and Geometric Confinement
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.
Grain Boundary Structure
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 geometrical constraint of no sliding at the interface of film and sur-

face is the reason for the singular stress field to develop around the diffu-
sion wedge and is therefore responsible for the occurrence of parallel glide
dislocations [46].
The geometrical constraint imposed by the grain size strongly influences
the dislocation network that develops inside the grain. In very small grains of
a few tens of nanometers, only one or two dislocations fit into a grain. In larger
grains of several hundred nanometers, a much larger number of dislocations
fit into each grain and may form a more complicated network (see Fig. 8.43).
Similar considerations apply to the Hall–Petch hardening [1].
Deformation Mechanisms of Small-Scale Materials
The dominance of grain boundary processes during deformation of ultra-thin

films is in qualitative agreement with recent investigations of other small-scale
materials, such as nanostructured materials [303].
The preliminary study on nanostructured materials discussed in Sect. 8.5.3
showed that an intergranular nanosubstructure constituted by twin lamellas
could play an important role in effectively strengthening materials. Since twin
grain boundaries are relatively poor diffusion paths (since they are low-energy
grain boundaries), such materials could potentially be successfully employed
at elevated temperatures where “usual” materials with ultra-fine grains can-
not be utilized since creep becomes the dominant deformation mechanism.
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.
8.7 Deformation and Fracture Mechanics of Carbon

Nanotubes
Carbon nanotubes (CNTs) constitute a prominent example of nanomaterials,
with many potential applications that could take advantage of their unique
mechanical, electrical, and optical properties [363–368]. A fundamental under-
standing of the properties of individual CNTs, or assemblies of CNTs in
bundles or nanopillars, or in conjunction with biological molecules such as
DNA may be critical to enable new technologies and to engineer CNT-based
devices. A series of different views of a single wall carbon nanotube (SWNT)
is shown in Fig. 8.52.
Fig. 8.52 Geometry of a (15,15) single wall carbon nanotube (SWNT) shown in
different views
The mechanical properties of CNTs are particularly important in many

technological applications. This includes cases in which the primary role of
CNTs is not related to their mechanical properties. Nevertheless, a thorough
understanding of the mechanical properties is essential to design manufac-
turing processes or to ensure reliability during operation of devices. The
interactions of individual CNTs can play a critical role in application and dur-
ing fabrication processes, and may pose significant challenges compared with
macroscopic classical engineering applications. This is because at nanoscale,
Fig. 8.53 Compressive deformation mechanics of a CNT with length L = 6 nm and

a diameter of d = 1 nm ((7,7) armchair CNT). The plot illustrates the increase of
strain energy as a function of compressive strain (subplot (a)), and shows associated
deformation mechanisms (subplot (b)). The analysis revealed that CNTs begin to
buckle according to a shell-like behavior. Subplot (c) depicts a similar analysis,
showing bending of a (13,0) CNT (length L = 8 nm and a diameter of d = 1
nm). The strain energy density increases according to a harmonic behavior until
the buckling point is reached. Reprinted from: B.I. Yakobson, C.J. Brabec, and
J. Bernholc, Physical Review Letters, Vol. 76(14), 1996, pp. 2511–2514. Copyright
c 1996 by the American Physical Society
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
modulus (approximately 1 TPa) and a very high strength (exceeding 40–

50 GPa).
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
illustrates that the mechanical properties of CNTs are strongly lengthscale

dependent.
Fig. 8.55 Illustration of the process of coarse graining a CNT, leading to a bead-type
representation [52]
8.7.1 Mesoscale Modeling of CNT Bundles
Atomistic modeling is typically limited to one or few CNTs with moder-

ate lengths. In order to reach longer timescale and lengthscale in simulation,
mesoscale methods have been developed that describe the behavior of CNTs
at a larger, coarse-grained scale [52]. Figure 8.55 reviews the process of
coarse-graining a CNT, leading to a bead-type representation. Many different
approaches in coarse-graining CNTs have been described, and the bead-type
model is only one of many possible approaches.
We briefly review the details of this molecular mesocopic formulation. The
total energy of the system is expressed as
U = UT + UB + Uweak , (8.16)
where UT is the energy stored in chemical bonds due to stretching along

the axial direction, UB is the energy due to bending of the CNT, and Uweak
constitute weak (vdW) interactions. The total energy contribution of each
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
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
Variable Numerical value

Equilibrium bead distance r0 (Å) 10
Bead particle mass m0 (amu) 1,953
Tensile stiffness parameter kT (kcal mol−1 Å−2 )
(0)
1,000
Tensile stiffness parameter kT (kcal mol−1 Å−2 )
(1)
700
Hyperelastic parameter (Å) 10.5
Fracture parameter (bond rupture distance) (Å) 13.2
Equilibrium angle (degrees) 180
Bending stiffness parameter kbend (kcal mol−1 Å−2 ) 14,300
Dispersive parameter (kcal mol−1 ) 15.1
Dispersive parameter (Å) 9.35
Table 8.3 Summary of mesoscopic parameters derived from atomistic modeling [52]
an analysis of tensile deformation). For instance, the dispersive interactions

between CNTs are determined to reproduce the adhesion energy of CNTs
as obtained in a full atomistic representation. The tensile spring constant is
determined from atomistic calculations of stretch vs. deformation, consider-
(0) (1)
ing the small- and large-deformation regime to determine kT and kT . The
bending spring parameter is determined based on the results of the bending
stiffness of a single CNT. Table 8.3 summarizes all parameters used in the
mesoscopic model.
Fig. 8.57 Bundle of several individual SWNTs as obtained from a molecular

simulation model [52]
8.7.2 Mesoscale Simulation Results
CNTs usually assembly in bundles as shown in Fig. 8.57. Large-aspect ratio

CNTs are extremely flexible, and can be deformed into almost arbitrary shapes
with relatively small energetic effort. Figure 8.58 depicts the response of a
CNT bundle to compressive mechanical load. The mesoscale model enables
the simulation of such large assemblies of CNTs, which would otherwise not
be accessible to atomistic simulation.
Fig. 8.58 Response of a CNT bundle to mechanical compressive loading. Even

for relatively small strains, the structure starts to buckle, eventually leading to
significantly deformed and buckled shapes [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
“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
The critical length scale was estimated to approximately 79 nm for a (5,5)

CNT in vacuum.
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
The presentation of atomistic simulation CNTs was extremely brief and by no

means comprehensive. However, it provides a small glimpse into this exciting
new field. Further, the development of the mesoscale model has illustrated how
atomistic models (in particular the “molecular dynamics” algorithm) can be
immediately used to define coarse grained models, as shown here for a bead
model [52]. This simple mesoscopic model is particularly suited to describe the
mechanical properties and self-assembly mechanisms of CNTs with ultra-large
aspect ratio. The mesoscopic model can reach timescales and lengthscales not
accessible to the full atomistic model, but still includes information about
the fracture mechanics of individual CNTs via the inclusion of appropriate
terms in the mesoscale model formulation. The mesoscopic model discussed
here can be straightforwardly implemented for other CNTs than those studied
here, including multiwall CNTs.
8.8 Flaw-Tolerant Nanomaterials: Bulk Fracture

and Deformation
The material discussed in the previous sections have clearly shown that once
the characteristic size of materials reaches nanoscale, the mechanical prop-
erties may change drastically and classical mechanisms of materials failure
cease to hold. In the following two sections, we focus on joint atomistic-
continuum studies of failure and deformation of nanoscale materials, focusing
on a particular phenomenon referred to as “flaw-tolerance” [54, 373, 374].
In particular, we discuss the size dependence of brittle fracture and the
effect of characteristic dimensions on adhesion. It will be illustrated that if the
characteristic dimension of a material is below a critical lengthscale that can
be on the order of several nanometers, the classical Griffith theory of fracture
no longer holds. An important consequence of this finding is that materials
with nanosubstructures may become flaw-tolerant, as the stress concentration
at crack tips disappears and failure always occurs at the theoretical strength
of materials, regardless of defects. The atomistic simulations reviewed here
complement recent continuum analyses [54]. Both approaches reveal a smooth
transition between Griffith modes of failure via crack propagation to uniform
bond rupture at theoretical strength below a nanometer critical length. The
results may be important for understanding failure of many brittle nanoscale
materials, in particular biological structures. Biological materials typically
feature nanoscopic features as a most fundamental building block. It has been
suggested that in light of the flaw-tolerance concept, scale reduction may be
a common design principle found in biological materials to create structural
links leading to robust architectures.
The analysis reviewed here is also interesting as another illustration of how
continuum mechanics theory (here linear elastic fracture mechanics) can be
coupled with atomistic simulation approaches to study fracture and adhesion
of nanostructures [375].
8.8.1 Strength of Brittle Nanoparticles
In this section, we focus on fracture properties of ultra-small brittle particles

and the impact of size variation on fracture properties.
The goal of these studies is to understand the limiting cases for the validity
of Griffith’s theory of fracture. In addition to general interest to understand
if Griffith’s theory can be applied to brittle fracture at ultra-small scales,
this study is motivated by the fact that in bone, mineral platelets appearing
at ultra-small nanoscale dimensions seem to play an integral role in the load
transfer process. Thus, their properties may have implications on the strength
Fig. 8.60 Bone-like materials consist of a hierarchical microstructure made of

nanoscale constituents [53, 54]. Left: The plot depicts the microstructure of such
bone-like biological materials at the smallest scale. Such materials typically consist
of fragile, brittle mineral platelets embedded in protein matrix materials as soft as
human skin. The combination of these two phases in a nanocomposite results in
superior materials properties. In the studies, we focus on the fracture properties of
mineral platelets since they play a critical role in determining the strength of these
materials. Right: The tension–shear chain model showing the path of load transfer
in the mineral–protein composites. The mineral platelets carry tensile load and the
protein transfers loads between the platelets via shear. In this section we focus on
the strength of the mineral platelets
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
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.
Theoretical Considerations: Breakdown of Griffith Continuum

Theory
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
Similar expressions for critical lengthscales can be derived for a variety of

different geometries, and also for the case of ductile material behavior [45].
These considerations have led to the question if the continuum theory based
on the Griffith concept is still applicable at very small scales. Now we use
atomistic simulation as a tool to gain further insight.
Atomistic Modeling
Atomistic modeling of the strip crack problem is conducted by classical molec-

ular dynamics simulations, utilizing a global energy minimization scheme. A
three-dimensional model is used to study crack nucleation and propagation.
Consider the geometry depicted in Fig. 8.61. The initial crack extends over
half of the slab in the x-direction. The slab size in the x-direction is several
times larger than that in the y-direction. The model is built using a FCC crys-
tal oriented in cubic orientations, with x = [100], y = [010], and z = [001].
The crystal is periodic in the z-direction with crack faces along (010)-planes.
We use the concept of virtual atom types to distinguish various atomic
interactions and to allow application of boundary conditions. Atoms in the
boundary region (as shown in Fig. 8.61) are assigned a specific virtual atom
type and are displaced according to a prescribed displacement field.
We use an energy minimization scheme to relax the crystal after each
increment of loading. An increment of strain of magnitude ∆εyy = 0.001
is applied every 3,000 integration steps. Different loading rates are chosen
to assure that the results have reached equilibrium before the next loading
increment is applied. The loading is constant along the x-direction.
As reviewed in Sect. 2.6, interatomic potentials for a variety of differ-
ent brittle materials exist, many of which are derived from first principles.
However, it is difficult to identify generic relationships between potential
parameters and macroscopic observables when using such complicated poten-
tials. Furthermore, in many cases the potential parameters do not have
immediate physical meaning for the bonding between atoms. Here we use
an alternative approach based on model potentials describing the behavior
of model materials. To investigate universal scaling behavior between micro-
scopic and macroscopic variables, this has been shown to be very fruitful and
allows fundamental insight, in particular into the fracture mechanisms, as
illustrated in Chap. 6.
By using simple model potentials, the complexities with sophisticated
potential formulations can be deliberately avoided. Based on this concept,
a simple pair potential based on a harmonic interatomic potential with spring
constant k0 in combination with a Lennard-Jones potential is used to describe
smooth bond breaking. The harmonic potential is chosen to model linear elas-
tic material behavior as assumed in Griffith’s fracture theory. This allows one
to define a well-described reference system.
It is important to note that although simple pair potentials do not allow
drawing conclusions for unique phenomena pertaining to specific materials,
they enable one to understand universal, generic relationships between poten-
tial shape and brittle fracture mechanics and adhesion properties of materials,
and help elucidate universal scaling laws of fracture mechanics.
To model a perfectly brittle solid, harmonic interactions are assumed in the
bulk of the strip. Atoms in the bulk only interact with their nearest neighbor,
and bonds never break. Crack propagation is constrained along a weak fracture
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.
The study begins with carrying out computational experiments of studies of

the fracture strength of thin layers depending on their size ξ. In a series of
studies, we calculate the critical fracture stress σ as a function of the material
size (layer width ξ).
Figure 8.62 shows the results of large-scale atomistic simulations of fracture
strength of small perfectly brittle platelet as a function of the inverse of the
square root of the size of the material characterized by ξcr /ξ. In the plot,
the predictions of Griffith’s theory and the theoretical strength of the material
are both included. The fracture stress σ is normalized with respect to the
maximum strength at the onset of failure, σth .
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
Clearly, the brittle fracture mechanism shows a strong size dependence.

Whereas the strength of the materials
is predicted well based on Griffith’s
theories for large dimensions ( ξcr /ξ < 1), reduction of dimension results in
deviation from this prediction, and eventually
failure of the material at σcr
regardless of the presence of flaws (for ξcr /ξ > 1). This observation suggests
a change in behavior once the dimensions of the solid are below the critical
length scale: The Griffith theory is not valid for extremely thin layer widths.
Now the focus is on the stress distribution ahead of the crack slightly
before failure occurs. In macroscopic systems and according to the classical
understanding, it is expected that the crack tip always constitutes a location
of high stress concentration, according to the inverse square root scaling dis-
cussed in Sect. 6.2.2. Is this also true when the characteristic dimensions of
materials reach nanoscale?
Calculation of the stress distribution ahead of the crack reveals that the
stress becomes increasingly homogeneous as the lengthscale of the structure
reduces. In fact, it is observed that the distribution becomes completely uni-
form for structural dimensions well below the critical lengthscale. This is in
strong contrast to what we would expect based on a completely classical
picture.
Further, we observe that the stress distribution does not change with struc-
tural size any more
once the size of the material is below a critical value (this
is observed for ξcr /ξ > 2.67 in the simulations). The results are shown in
Fig. 8.63. This size-independence of the fracture strength of materials is also
in clear contrast to the conventional knowledge. The results suggest that at
nanoscale, materials may behave dramatically differently.
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
It is noted that a similar behavior as observed here for tensile loading

is also expected for shear loading, as the underlying equations take a quite
similar form.
8.9 Nanoscale Adhesion Systems

Similar concepts as discussed in the last section also apply to adhesion
systems, as reported first in [376]. This application provides another opportu-
nities to link atomistic methods with theories developed within the continuum
assumption of engineering mechanics [375, 377].
The focus of the discussion presented here is on the size dependence of
cylindrical adhesion systems. It will be demonstrated that optimal adhesion
can be achieved by either lengthscale reduction or by optimization of the
shape of the surface of the adhesion element. This illustrates two paths to
increase the efficacy of adhesion systems. It is found that change in shape can
lead to optimal adhesion strength; those systems are not robust against small
deviations from the optimal shape. In contrast, reducing the dimensions of the
adhesion system results in robust adhesion devices that fail at their theoretical
strength, regardless of the presence of flaws. An important consequence of this
finding is that even under presence of surface roughness, optimal adhesion
is possible provided the size of contact elements is sufficiently small. The
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
atomistic results corroborate earlier theoretical modeling at the continuum

scale [376]. The section concludes with a discussion of the relevance of the
studies with respect to the biological architecture of bone nanostructures and
nanoscale adhesion elements in Geckos.
8.9.1 Strength of Fibrillar Adhesion Systems
In this section, the focus is on fibrillar adhesion structures as they appear

in many biological systems such as geckos. Figure 8.64 depicts a microscopic
view of flies and gecko attachment systems, illustrating that they are made
up of a large number of ultra-small fibrillar adhesion elements [55, 378].
To understand adhesion properties at small scales, we have modeled an
elastic flat-ended cylindrical hair in adhesive contact with a rigid substrate.
The radius of the cylinder is R. To test the ability of the flat cylinder to adhere
in the presence of adhesive flaws, imperfect contact between the spatula and
substrate is assumed such that the radius of the actual contact area is a = αR,
where 0 < α < 1, as shown in Fig. 8.65(left), and the outer rim αR < r < R
represents flaws or regions of poor adhesion. In the present model, which is
similar to continuum models of this situation [376, 378], there is no adhesion
in the region αR < r < R, and so the model resembles a cylinder attached to
the substrate with a circumferential crack.
The adhesive strength of such an adhesive joint can be calculated by treat-
ing the contact problem as a circumferentially cracked cylinder, in which case
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
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.
8.9.2 Theoretical Considerations of Shape Optimization

of Adhesion Elements
The results in the previous sections indicate that optimal adhesion and opti-
mal fracture strength can be achieved by reducing the dimension of the
structure.
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
In this section, we focus on the question: Can optimal adhesion be achieved

at any size of the punch? The system of interest is a periodic array of rigid
punches attached to an elastic substrate as schematically shown in Fig. 8.66
in a quasi-two-dimensional geometry with periodic boundary conditions.
The reader is referred to the primary literature [376] for discussions on the
concepts of optimal and singular shapes in adhesive contact mechanics. For a
punch array, the optimal shape is given by a series expression as described in
detail in [376]. The theoretical prediction of the optimal shape as a function
of a shape parameter Ψ is shown in Fig. 8.67 (where Ψ = 1 is the optimal
shape and Ψ = 0 corresponds to the singular, straight surface shape).
The critical lengthscale for a single fiber on a substrate (in analogy the
earlier derivations) is given by
8 E ∗ ∆γ
Rcr = 2 . (8.34)
π σth
8.9.3 Atomistic Modeling
The simulation geometry of the atomistic studies discussed in this section is

shown in Fig. 8.65 (studies of size reduction only) and Fig. 8.66 (studies of
shape variation). In the first case, the punch is elastic and the substrate is
rigid, and in the second case, the punch is rigid and the substrate is elastic.
This is in accordance with the continuum mechanics models described above.
In both cases, we model the elastic part using a harmonic potential, and
we treat the interface between the two parts using a Lennard-Jones potential
to model the vdW adhesion interactions. The expressions for the interatomic
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)
potentials, as well as the overall simulation method, including the expressions

for elastic properties and fracture surface energy are identical to the procedure
described above. It is noted that this Griffith model of adhesion corresponds
to the JKR model [379].
Studies of variations of surface shape require a method to achieve small
and smooth variations of surface shape in atomistic models. Such a change in
surface shape is achieved by displacing the rows of atoms as shown in Fig. 8.68.
This method allows one to achieve a smoothly varying surface in the molecular
dynamics simulations. We note that the alternative approach in cutting the
optimal shape out of an atomic lattice does not work well because of the
discreteness of the lattice and the resulting steps on the order of a Burgers
vector.
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.
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)
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)
Finally we discuss the change in the adhesion strength due to variations

of the surface shape. The atomistic studies are based on the continuum con-
siderations reviewed above where it was stated that the choice of a specific,
optimal surface shape that would always lead to optimal adhesion, at any
lengthscale.
Figure 8.71 shows the 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. It is observed 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.
Figure 8.72 shows the maximum adhesion strength as a function of the
shape parameter Ψ for different sizes of the punch. These observations allow
drawing conclusions about the robustness: 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.
Robustness thus seems to be closely coupled to nanodimensioned adhesion
systems. We believe that only reduction of lengthscale results in (1) robustness
and (2) disappearance of the stress concentration. This may be an explanation
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
We have reviewed a series of continuum and atomistic studies to investigate

how the fracture and failure behavior of materials changes as a function of
material size.
The atomistic simulations reveal a smooth transition between Griffith
mode of failure via crack propagation to uniform bond rupture at theoret-
ical strength below a nanometer critical length as shown in Fig. 8.62 for both
fracture of nanoparticle and surface adhesion of an elastic punch on a rigid
substrate. It was found that below a critical length, the stress distribution
becomes increasingly homogeneous and eventually uniform near the crack tip.
The atomistic simulations fully support the hypothesis that materials become
insensitive to flaws below a critical nanometer lengthscale.
The results corroborate earlier suggestions made at the continuum level
that the concept of nanoscale flaw tolerance may play a critical role in devel-
oping structural links in biological materials, as many biological materials
feature small nanoscale dimensions. Small nanosubstructures lead to robust,
flaw-tolerant materials.
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
Surprisingly, this principle is found in different geometries and different

situations, including bulk and surface materials. The studies on adhesion
systems indicate that optimal adhesion can be achieved at any scale if the
adhesion surface shape is adapted to eliminate locations of stress concentra-
tion (see Fig. 8.71). However, this design approach does not lead to robust
adhesion elements as the smallest deviations from the optimal shape lead to
catastrophic failure (see Fig. 8.72).
It is found that the reduction of the diameter of cylindrical adhesion sys-
tems leads to optimal adhesion including robustness; therefore, “nano is also
robust.” We note that the actual displacements of atoms necessary to realize
the optimal shape are on the order of a few Angstroms. Due to the discrete
nature of atoms with typical interatomic bond distances between 1 and 3 Å,
the realization of the optimal shape at such small scale may become diffi-
cult. This may be another, alternative reason why shape optimization is not
a preferred strategy to achieve optimal adhesion. We refer the reader to other
articles for a more detailed discussion of linking the flaw-tolerance concept to
biological materials [54, 376, 378].
Atomistic modeling represents a powerful tool in designing virtual exper-
iments to demonstrate the concept and effect of scale reduction and shape
changes. Unlike purely continuum mechanics theories, molecular dynam-
ics simulations can intrinsically handle stress concentrations (singularities)
well and provide accurate descriptions of bond breaking. Also, it can be
rather straightforwardly used to study ultra-small scale systems, thus prob-
ing the limiting cases where continuum theories start to break down. The
approach of using simplistic model potentials rather than utilizing highly
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
1. T.H. Courtney. Mechanical behavior of materials. McGraw-Hill, New York,

1990.
2. H. Gao. In: European White Book on Fundamental Research in Materials Sci-
ence, chapter: Modelling Strategies for Nano- and Biomaterials, pages 144–148
Max-Planck Institut, Germany 2001.
3. B. Wu, A. Heidelberg, and J.J. Boland. Mechanical properties of ultrahigh-
strength gold nanowires. Nat. Mater., 4:525–529, 2005.
4. S.E. Cross, Y.-S. Jin, J. Rao, and J.K. Gimzewski. Nanomechanical analysis
of cells from cancer patients. Nat. Nanotechnol., 2:780–783, 2007.
5. R.K. Nalla, J.H. Kinney, and R.O. Ritchie. Mechanistic fracture criteria for
the failure of human cortical bone. Nat. Mater., 2:168–168, 2003.
6. P.E. Marszalek, H. Lu, H. Li, M. Carrion-Vazquez, A.F. Oberhauser, K. Schul-
ten, and J.M. Fernandez. Mechanical unfolding intermediates in titin modules.
Nature, 402:100–103, 1999.
7. M.J. Buehler. Hierarchical chemo-nanomechanics of stretching protein
molecules: Entropic elasticity, protein unfolding and molecular fracture.
J. Mech. Mater. Struct., 2(6):1019–1057, 2007.
8. M.J. Buehler. Atomistic modeling of elasticity, plasticity and fracture of protein
crystals. J. Comput. Theor. Nanosci., 3(5):670–683, 2006.
9. F. Cleri, S. Yip, D. Wolf, and S. Philpot. Atomic-scale mechanism of crack-
tip plasticity: Dislocation nucleation and crack-tip shielding. Phys. Rev. Lett,
79:1309–1312, 1997.
10. J. Tersoff. Empirical interatomic potentials for carbon, with applications to
amorphous carbon. Phys. Rev. Lett., 61(25):2879–2883, 1988.
11. M.T. Yin and Marvin L. Cohen. Will diamond transform under megabar
pressures? Phys. Rev. Lett., 50(25):2006–2009, 1983.
12. M.J. Buehler, A. van Duin, T. Jacob, Y. Jang, B. Merinov, and W.A. Goddard.
Formation of water at a pt(111) catalyst surface: A study using the reaxff
reactive force field. Mater. Res. Soc. Symp. Proc. Vol., 900E:0900–O03–09.1–
0900–O03–09.6, 2006.
13. M.J. Buehler and H. Gao. In: Handbook of Theoretical and Computational
Nanotechnology, chapter Ultra-large scale simulations of dynamic materials
failure. American Scientific Publishers (ASP), Stevenson Ranch, CA 2006.
464 References
14. M.J. Buehler. Multi-million atom simulation of work-hardening in ductile

metals with multi-body potentials. Unpublished results.
15. F. Ercolessi and J.B. Adams. Interatomic potentials from 1st principle-
calculations – the force matching method. Europhys. Lett., 28(8):583–588,
1994.
16. Y. L. Sun, Z.P. Luo, A. Fertala, and K.N. An. Stretching type II collagen with
optical tweezers. J. Biomech., 37(11):1665–1669, 2004.
17. M.J. Buehler and S.Y. Wong. Entropic elasticity controls nanomechanics of
single tropocollagen molecules. Biophys. J., 93(1):37–43, 2007.
18. W.K. Liu, E.G. Karpov, S. Zhang, and H.S. Park. An introduction to com-
putational nanomechanics and materials. Comput. Methods Appl. Mech. Eng.,
193(17–20):1529–1578, 2004.
19. H.S. Park, E.G. Karpov, W.K. Liu, and P.A. Klein. The bridging scale for two-
dimensional atomistic/continumn coupling. Phil. Mag., 85(1):79–113, 2005.
20. E.B. Tadmor, M. Ortiz, and R. Phillips. Quasicontinuum analysis of defects in
solids. Phil. Mag. A, 73:1529, 1996.
21. D.H. Warner, W.A. Curtin and S. Qu. Rate dependence of crack-tip processes
predicts twinning trends in f.c.c. metals. Nat. Mater., 6:876–881, 2007.
22. L.B. Freund. Dynamic Fracture Mechanics. Cambridge University Press,
Cambridge ISBN 0-521-30330-3, 1990.
23. B.Q. Vu and V.K. Kinra. Britle fracture of plates in tension – static field
radiated by a suddenly stopping crack. Eng. Fract. Mech., 15(1–2):107–114,
1981.
24. S. Chen, M.J. Buehler, and H. Gao. Supersonic mode III cracks propagating
in a stiff strip embedded in a soft matrix. Unpublished results.
25. T. Cramer, A. Wanner, and P. Gumbsch. Energy dissipation and path instabil-
ities in dynamic fracture of silicon single crystals. Phys. Rev. Lett., 85:788–791,
2000.
26. M.J. Buehler, A. Cohen, and D. Sen. Multi-paradigm modeling of fracture
of a silicon single crystal under mode ii shear loading. J. Algorithm Comput.
Technol., 2(2):203–221, 2008.
27. H. Gao. A theory of local limiting speed in dynamic fracture. J. Mech. Phys.
Solids, 44(9):1453–1474, 1996.
28. M.J. Buehler, F.F. Abraham, and H. Gao. Hyperelasticity governs dynamic
fracture at a critical length scale. Nature, 426:141–146, 2003.
29. M.J. Buehler and H. Gao. Dynamical fracture instabilities due to local
hyperelasticity at crack tips. Nature, 439:307–310, 2006.
30. J.R. Rice. Dislocation nucleation from a crack: an analysis based on the Peierls
concept. J. Mech. Phys. Solids, 40:239–271, 1992.
31. D.J. Oh and R.A Johnson. Simple embedded atom method model for fcc and
hcp metals. J. Mater. Res., 3:471–478, 1988.
32. J.E. Angelo, N.R. Moody, and M.I. Baskes. Trapping of hydrogen to lattice
defects in nickel. Model. Simul. Mater. Sci. Eng., 3:289–307, 1995.
33. Y. Mishin, D. Farkas, M.J. Mehl, and D.A. Papaconstantopoulos. Interatomic
potentials for monoatomic metals from experimental and ab-initio calculations.
Phys. Rev. B, 59:3393407, 1998.
34. A.F. Voter. In: Intermetallic Compounds: Principles and Applications, chapter
The embedded-atom model, page 7790. Wiley, New York, 1995.
References 465
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
175. V. Shenoy, R. Miller, E.B. Tadmor, D. Rodney, R. Phillips, and M. Ortiz.

An adaptive methodology for atomic scale mechanics - the quasicontinuum
method. J. Mech. Phys. Sol., 73:611–642, 1999.
176. V. Bulatov and T.D. de-la Rubia. Materials research by means of multiscale
computer simulation. MRS Bull., 26(3):169–170, 2001.
177. W.A. Curtin and R.E. Miller. Atomistic/continuum coupling in computational
materials science. Model. Sim. Mater. Sci. Eng., 11(3):R33–R68, 2003.
178. M.F. Horstemeyer, M.I. Baskes, V.C. Prandtl, J. Philliber, and S. Vonderheid.
A multiscale analysis of fixed-end simple shear using molecular-dynamics, crys-
tal plasticity and a macroscopic internal state variable theory. Model. Sim. Mat.
Sci. Eng., 11:265–386, 2003.
179. A.J. Haslam, D. Moldovan, S.R. Phillpot, D. Wolf, and H. Gleiter. Combined
atomistic and mesoscale simulation of grain growth in nanocrystalline thin
films. Comput. Mat. Sci., 23:15–32, 2003.
180. D. Moldovan, D. Wolf, S.R. Phillpot, and A.J. Haslam. Mesoscopic simulation
of two-dimensional grain growth with anisotropic grain-boundary properties.
Phil. Mag. A, 82(7):1271–1297, 2002.
181. K.W. Schwarz. Simulation of dislocations on the mesoscopic scale. I. Meth-
ods and examples, II. Application to strained-layer relaxation. J. Appl. Phys.,
85:108–129, 1999.
182. L. Nicola, E. Van der Giessen, and A. Needleman. Discrete dislocation analysis
of size effects in thin films. J. Appl. Phys., 93:5920–5928, 2003.
183. P. Gumbsch and G.E. Beltz. On the continuum versus atomsitic descriptions
of dislocation nucleation and cleavage in nickel. Model. Sim. Mat. Sci. Eng.,
3(5):597–613, 1995.
184. F.F. Abraham, N. Bernstein, J.Q. Broughton, and D. Hess. Dynamic fracture
of silicon: Concurrent simulation of quantum electrons, classical atoms, and
the continuum solid. MRS Bull., 25(5):27–32, 2000.
185. G. Lu, E.B. Tadmor, and E. Kaxiras. From electrons to finite elements: A
concurrent multiscale approach for metals. Phys. Rev. B, 73(2), 2006. 024108.
186. H. Lu, N.P. Daphalapurkar, B. Wang, S. Roy, and R. Komanduri. Multiscale
simulation from atomistic to continuum-coupling molecular dynamics (md)
with the material point method (mpm). Phil. Mag., 86(20):2971–2994, 2006.
187. G.J. Wagner and W.K. Liu. Coupling of atomistic and continuum simulations
using a bridging scale decomposition. J. Comput. Phys., 190(1):249–274, 2003.
188. D.E. Farrell, H.S. Park, and W.K. Liu. Implementation aspects of the bridging
scale method and application to intersonic crack propagation. Int. J. Numer.
Meth. Eng., 71(5):583–605, 2007.
189. V.B. Shenoy, R. Miller, E.B. Tadmor, R. Phillips, and M. Ortiz. Quasicontin-
uum models of interfacial structure and deformation. Phys. Rev. Lett., 80:742,
1998.
190. Z. Tang, H. Zhao, G. Li, and N.R. Aluru. Finite-temperature quasicontin-
uum method for multiscale analysis of silicon nanostructures. Phys. Rev. B,
74(6):064110, 2006.
191. J. Knap and M. Ortiz. An analysis of the quasicontinuum method. J. Mech.
Phys. Solids, 49:1899–1923, 2001.
References 473
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
236. A. Boresi and K.P. Chong. Elasticity in Engineering Mechanics. Wiley-

Interscience, New York, 2nd edition, 2000.
237. F.F. Abraham and H. Gao. How fast can cracks propagate? Phys. Rev. Lett.,
84(14):3113–3116, 2000.
238. M. Zhou. Equivalent continuum for dynamically deforming atomistic particle
systems. Phil. Mag. A, 82(13), 2002.
239. D.H. Tsai. Virial theorem and stress calculation in molecular-dynamics. J.
Chem. Phys., 70(3):1375–1382, 1979.
240. L.I. Slepyan. Models and Phenomena in Fracture Mechanics. Springer, Berlin,
2002.
241. J. Fineberg, S.P. Gross, M. Marder, and H.L. Swinney. Instability in the
propagation of fast cracks. Phys. Rev. B, 45(10):5146–5154, 1992.
242. K. Ravi-Chandar. Dynamic fracture of nominally brittle materials. Int. J.
Fract., 90:83–102, 1998.
243. P.D. Washabaugh and W.G. Knauss. A reconcilation of dynamic crack velocity
and rayleigh-wave speed in isotropic brittle solids. Int. J. Fract., 65:97–114,
1994.
244. E. Sharon and J. Fineberg. Confirming the continuum theory of dynamic brittle
fracture for fast cracks. Nature, 397:333, 1999.
245. J.E. Field. Brittle fracture–its study and application. Contemp. Phys., 12:1–31,
1971.
246. J.A. Hauch and M. Marder. Energy balance in dynamic fracture, investigated
by a potential drop technique. Int. J. Fract., 1–2:133–151, 1998.
247. D. Hull and P. Beardmore. Velocity of propagation of cleavage cracks in
Tungsten. In. J. Fract., 2:468–488, 1966.
248. H. Gao. Elastic waves in a hyperelastic solid near its plane-strain equibiaxial
cohesive limit. Phil. Mag. Lett., 76(5):307–314, 1997.
249. K.B. Broberg. Dynamic crack propagation in a layer. Int. J. Solids Struct.,
32(6–7):883–896, 1995.
250. P.J. Petersan, D. Robert, M. Deegan, M. Marder, and H.L. Swinney. Cracks in
rubber under tension exceed the shear wave speed. Phys. Rev. Lett., 93:015504,
2004.
251. H. Gao. Surface roughening and branching instabilities in dynamic fracture. J.
Mech. Phys. Solids, 41(3):457–486, 1993.
252. J.D. Eshelby. Elastic field of a crack extending non-uniformly under general
anti-plane loading. J. Mech. Phys. Solids, 17(3):177–199, 1969.
253. H.H. Yu and Z. Suo. Intersonic crack growth on an interface. P. R. Soc. Lond.
A Mat., 456(1993):223–246, 2000.
254. D.J. Andrews. Rupture velocity of plane strain shear cracks. J. Geophys. Res.,
81:5679–5687, 1976.
255. R.J. Archuleta. Analysis of near-source static and dynamic measurements from
the 1979 imperial valley earthquake. Bull. Seismol. Soc. Am., 72:1927–1956,
1982.
256. H. Kuppers. The initial course of crack velocity in glass plates. Int. J. Fract.
Mech., 3:13–17, 1969.
257. A.F. Fossum and L.B. Freund. Nonuniformly moving shear crack model of a
shallow focus earthquake mechanism. J. Geophys. Res., 80:3343–3347, 1975.
258. B.A. Auld. Accoustic Fields and Waves in Solids. R.E. Krieger Publishing
Company, Malabar, FL, 2nd edition, ISBN 0-89874-783-X edition, 1990.
476 References
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
280. A. Strachan, T. Cagin, and W.A. Goddard. Crack propagation in a tantalum

nano-slab. J. Comput.-Aided Mater. Des., 8:151–159, 2002.
281. Martin Z. Bazant and Efthimios Kaxiras. Modeling of covalent bonding in
solids by inversion of cohesive energy curves. Phys. Rev. Lett., 77(21):4370–
4373, 1996.
282. A. Strachan, A.C.T. van Duin, D. Chakraborty, S. Dasgupta, and W.A.
Goddard. Shock waves in high-energy materials: The initial chemical events
in nitramine RDX. Phys. Rev. Lett., 91:098301, 2003.
283. A.D. Becke. Density-function thermochemistry. 3. The role of exact exchange.
J. Chem. Phys., 98(7):5648–5652, 1993.
284. T. Cramer, A. Wanner, and P. Gumbsch. Dynamic fracture of glass and single-
crystalline silicon. Z. Metallkd., 90:675–686, 1999.
285. V.P. Smyshlyaev and N.A. Fleck. The role of strain gradients in the grain size
effect for polycrystals. J. Mech. Phys. Solids, 44:465–495, 1996.
286. J.Y. Shu and N.A. Fleck. The prediction of a size effect in micro-indentation.
Int. J. Solids Struct., 35:1363–1383, 1998.
287. T. Mura. Micromechanics of Defects in Solids. Martinus Nijhoff, Dordrecht,
1987.
288. N.A. Fleck and J.W. Hutchinson. A phenomenological theory for strain
gradient effects in plasticity. J. Mech. Phys. Solids, 41:1825, 1993.
289. H. Gao, Y. Huang, W.D. Nix, and J.W. Hutchinson. Mechanism-based strain
gradient plasticity–I. Theory. J. Mech. Phys. Solids, 47:1239–1263, 1999.
290. E.H. Lee. Elastic-plastic deformation at finite strains. J. Appl. Mech., 1:1, 1969.
291. C.S. Han, H. Gao, Y. Huang, and W.D. Nix. Mechanism-based strain gradient
crystal plasticity - I. Theory. J. Mech. Phys. Solids, 53(5):1188–1203, 2005.
292. J. Roth, F. Gähler, and H.-R. Trebin. A molecular dynamics run with
5.180.116.000 particles. Int. J. Mod. Phys. C, 11:317–322, 2000.
293. R.G. Hoagland, J.P. Hirth, and P.C. Gehlen. Atomic simulation of dislocation
core structure and Peierls stress in alkali-halide. Phil. Mag., 34(3):413–439,
1976.
294. B. deCelis, A.S. Argon, and S. Yip. Molecular-dynamics simulation of crack
tip processes in alpha-iron and copper. J. Appl. Phys., 54(9):4864–4878, 1983.
295. S.J. Zhou, D.M. Beazly, P.S. Lomdahl, and B.L. Holian. Large-scale molecular-
dynamics simulations of three-dimensional ductile failure. Phys. Rev. Lett.,
78:479–482, 1997.
296. A. Girshick, D.G. Pettifor, and V. Vitek. Atomistic simulation of titanium -
II. Structure of 1/3 < 1210 > screw dislocations and slip systems in titanium.
Phil. Mag. A, 77:999–1012, 1998.
297. J. Cserti, M. Khantha, V. Vitek, and D.P. Pope. An atomistic study of the
dislocation core structures and mechanical-behavior of a model alloy. Mat.
Sci. Eng. A, 152(1–2):95–102, 1998.
298. S.J. Zhou, D.L. Preston, P.S. Lomdahl, and D.M. Beazley. Large-scale
molecular-dynamics simulations of dislocation interactions in copper. Science,
279:1525–1527, 1998.
299. Y. Qi, A. Strachan, T. Cagin, and W.A. Goddard III. Large-scale atomistic
simulations of screw dislocation structure, annihilation and cross-slip in FCC
Ni. Mat. Sci. Eng. A, 145:309–310, 2001.
478 References
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
Concurrent multiscale simulation tools, Diffusion, 15, 378

157 Diffusion wedges, 385, 395, 397, 409
Confinement, 244, 301, 436 crack like, 401
Constrained grain boundary diffusion, dislocation glide, 409
385 formation, 400
atomistic simulations, 396 versus crack, 406, 408
bicrystal model, 396 Diffusive displacement, 401
continuum model, 388 Diffusivity
experimental evidence, 386 copper, 178
Continuum mechanics, 15, 95 dependence, 436
Continuum theory, 16, 121 surface, 178
Copper Discretization, 76
nanocrystalline, 379 Dislocation
nanostructured, 422 cross slip, 425
Coulomb potential, 64 pileup, 404
Coupling pilups, 425
atomistic-continuum, 393 Dislocation bowing, 412
atomistic-continuum theories of Dislocation channelling, 382
plasticity, 339 Dislocation climb, 388
atomistic-experiment, 249, 271 Dislocation cutting, 346
atomistic-mesoscopic scale, 432 Dislocation density, 418
strain, 124 tensor, 341
stress, 123 Dislocation dipole, 396, 403
Coupling constant Dislocation dragging force, 349
heat bath, 41 Dislocation motion
Crack, 401, 419 grain boundaries, 421
diffusion wedge, 386 Dislocation network, 421
hyperelasticity, 207 Dislocation pinning, 346
initiation time, 267 Dislocations, 10, 11, 341
parallel glide dislocations, 404 interaction, 424
versus diffusion wedge, 406 Displacement, 97
Cracks, 10 DREIDING, 56
Cross-slip, 352 Ductile failure, 13
Crystal structure, 9 Dundur’s parameter, 393
Dynamic materials failure, 5
Deformation
diffusive, 378 EAM, 54, 166, 357, 359
elastic, 5 Edge dislocation, 329, 390
nanocrystalline materials, 378 Elastic regime, 5
plastic, 5, 373 Electron gas, 55
thin films, 430 Electron volt, 155
Deformation map, 434 Electronic properties, 68
Deformation mechanisms, 373 Embedded atom potential, 48
Deformation tensor, 104 Embedded-atom method, 54
Deformation-mechanism map, 374 Empirical potentials, 50
Density functional theory, 48 Energetic elasticity, 122
DFT, 164 Energy approach to elasticity, 105
Differential beam equations, 116 Energy length scale
Differential multiscale modeling, 159 characteristic, 188, 234, 244, 299
Index 485
Energy method, 85 Hall-Petch, 373

Energy minimization, 43 Hamiltonian, 37
Energy release rate, 190, 196, 447, 453 Hardening, 424
Entropic elasticity, 122 Harmonic potential, 54, 128, 142
Equations of motion, 39 HCP, 9
Ergodic hypothesis, 36, 41 Heat bath, 41
Experiments Hierarchical multiscale methods, 157
polycrystalline films, 409 High-energy grain boundary, 410, 414
Homologous temperature, 400
Failure, 5 Hooke’s law, 97
Failure processes, 32 Hybrid models, 169, 304, 359
FCC, 9 Hyperelasticity, 62, 108, 260
FCC lattice, 142
FCC packing, 13 Image force, 392
FEAt, 163 Image stress, 406
Flaw tolerance, 446, 457 Insects, 452
Flaws, 10 Interface
Fleischer mechanism, 351 crack-grain boundary, 419
Force calculation, 71 Interface effect, 381
Force field, 48 Interfaces, 287, 436
Fracture, 12 dissimilar materials, 287
Fracture instability, 197 Interfaces and geometric confinement,
Fracture surface energy, 140 436
Fracture surface energy 3D, 148 Interfacial dislocations, 382, 418
Free energy minima, 176 Intersonic mode I cracks, 242
Friedel-Escaig mechanism, 351 Interstitial tubes, 351
Inverse Hall-Petch effect, 375, 376
Gecko, 452, 453 Isobaric-isothermal ensemble, 40
Geometric analysis, 85
Jog dragging, 412
Geometric confinement, 14, 373, 381
Jogs, 413
Geometrically necessary dislocations,
389 kcal/mole, 155
Glassy phase, 398
Glide Langevin dynamics, 41
parallel glide dislocations, 409 Large-scale computing, 78
Grain boundary, 401 Leap-frog algorithm, 39
dislocation source, 414, 415 Length-and time scale
jogs, 402 Classical molecular dynamics, 341
stability, 402 Lennard-Jones, 48, 52
Grain boundary processes, 378 Limitations, classical molecular
Grain boundary structure dynamics, 68
elevated temperature, 400 LINUX
Grain boundary traction relaxation, supercomputers, 82
410, 415 Liquid-like grain boundary, 398
Grain boundary tractions, 430 Loading
Grain triple junction, 408 strain field, 399
Grand canonical ensemble ensemble, 40 Lomer-Cottrell locks, 355
Green-Kubo relations, 76 Long-time limit, 179
Griffith condition, 13, 266 Low-energy grain boundary, 410
486 Index
MAAD, 163 Nanoscale, 381

Materials failure confinement, 391
ductile, 85 deformation phenomena, 435
nickel, 85 Nanoscale adhesion, 453
Materials in small dimensions, 381 Nanostructured materials, 376
Mathews-Freund-Nix mechanism, 386 strain rate, 379
Mean square displacement function, 75 yield stress, 379
Measurement, 37 Nanostructures, 446
Mechanical properties, 142 Nanotechnology, 3, 381
Medium-range-order analysis, 90 Navier-Bernouilli, 114
Melting temperature Newton, 155
copper, 399 Newton’s laws, 96
Mesoscopic simulations, 50, 434 Nickel, 85
Message passing, 80 nanocrystalline, 379
Metropolis-Hastings algorithm, 45 Nonlinear elasticity, 108
Microcanonical ensemble, 40 NPT, 40
Microelectronic devices, 381 NVE, 40
Microscopic configurations, 41 NVT, 40
Microstructure, 9, 76
Miniaturization, 381 On-the-fly concurrent multiscale
Mirror-mist-hackle, 197 methods, 158
Mode I fracture Organic materials, 56
Mother-daughter mechanism, 289 Oxidation, 357
Mode II fracture, 243, 294
Mode III fracture, 142, 299 Pair potential, 48, 50
Model materials, 35, 341 Parallel glide dislocations, 167, 384, 419
Modeling, 32 experimental evidence, 386
Modeling and simulation, 32, 33, 90 minimum film thickness, 403
Molecular dynamics, 37 nucleation, 393, 396, 402
Molecular statics, 44 nucleation mechanism, 394
Monte Carlo, 37, 45 Parallel molecular dynamics, 80
Morse potential, 53 Parrinello-Rahman, 41
Mother-daughter mechanism, 289 Partial dislocations, 348, 350, 422
Mother-daughter-granddaughter Partial point defects, 413
mechanism, 294 Pascal, 155
MPI, 80 PBC, 70
Multi-body potential, 54 Peach-Koehler force, 392
Multi-scale phenomena, 35 Periodic boundary conditions, 70
Multi-scale simulations Petaflop computers, 80
hierarchical, 432 Phonons, 164
Multiparadigm modeling, 158, 170, 357 Pinning potential, 399
Multiscale, 157 Plane strain, 399
Multiscale modeling and simulation, Plastic deformation, 5
157 Plasticity, 341
atomistic modeling, 414
Nanocrystalline copper, 376 nanocrystalline materials, 422
Nanocrystalline materials, 15, 353 polycrystalline thin films, 414
Nanoindentation, 167 thin films, 409
Nanomaterials, 26, 438, 446 Point defect generation, 413
Index 487
Polycrystalline films, 409 Spring constant ratio, 236

Polycrystalline thin films Stacking fault, 358
atomistic modeling, 415 State transition, 179
Polycrystalline thin metal films, 381 Statistical mechanics, 36
Polymers, 9, 56 Strain, 97
Post-processing, 85 Strain rate, 266, 400
Potential, 48 Stress, 97
Property calculation, 73, 93 Stress intensity, 393
Protein unfolding, 92 Stress intensity factor, 194, 266, 390,
Proteins, 9, 56 396
parallel glide dislocations, 402
Quantization Stress tensor, 99
stress, 392 Sub-micron scale, 381
Quantum mechanics, 32, 35, 68 Sub-nano structure, 422
Quasicontinuum method, QC, 165 Submicron thin films, 385
Quasicrystals, 201, 343 Suddenly stopping crack, 260
Brittle fracture, 201 mode I, 268
Dislocations, 343 mode II, 278, 280
Ductile failure, 343 mode III, 303
Super-Rayleigh fracture, 242, 274
Radial distribution function, 74 Supercomputers, 79
Reactive potentials, 59 Supercomputing, 78, 80, 201, 342
ReaxFF, 59, 62, 169, 304, 359 Supersonic fracture, 207, 210, 280, 281,
Reduced units, 155 289
Reference units, 155 Supersonic mode I cracks, 289
Relaxation, 399 Supersonic mode II cracks, 243
Relaxation mechanisms, 430 Surface diffusion, 177
Rice-Thomson model, 404 Surface diffusivity, 177
Richard Feynman, 35 Surface effects, 381, 436
Rigid boundaries, 403 Surface steps, 178, 417
Rise time, 41
Robustness, 459 Temperature, 73
Temperature accelerated method, 177
SC, 9 Tersoff potential, 61
Screw dislocation, 329 The strongest size, 376
Self-folding, 439 Thermodynamics, 41, 105, 121
Sessile locks, 353 Thin films, 167, 381
Shielded Coulomb potential, 64 deformation map, 434
Shock loading, 364 yield stress, 384, 433, 435
Silicon, 55 Threading dislocations, 382, 383, 386,
Simulation, 32 402, 417
Simulation techniques, 49 versus parallel glide dislocations, 416,
Single atoms, 179 430
Single edge dislocations, 390 Three-dimensional molecular dynamics
Size effects, 373, 381, 438, 452 simulations, 142
Slip planes, 353 Threshold stress, 433
Slip vector, 88, 341, 410, 412 Tight-Binding approach, 162
Small-scale materials, 199 Tight-binding potential, 48
Speedup, 81 Tilt grain boundary, 399
488 Index
Time scale, 407 Verlet algorithm, 39

Time scale dilemma, 175, 176 Virial strain, 124
Time scale methods, 175 Virial stress, 76, 77, 123
Time step, 43 Virtual internal bond method, VIB, 168
top500.org, 80 Viscoelasticity, 169
Transition region, 171 Visualization, 83, 85, 345, 401
Transport properties, 76 distributed computing, 83
Triangular lattice, 9 movies, 83
Triple junction, 410 Richard Hamming, 84
Twin grain boundary, 424 virtual reality, 83
Twin lamella, 422 Volterra edge dislocations, 388
Two-dimensional lattice, 128
Water formation, 66
UFF, 56 Work-hardening, 354
Unit conversion, 155
Unstable stacking fault energy, 14 Yield stress, 435
Young’s modulus, 97, 120, 133, 142, 204
Vacancy tubes, 351 bilinear, 138, 207
Velocity autocorrelation function, 76 FCC, 142, 144
Velocity verlet, 398
Velocity Verlet algorithm, 40 Zhou’s virial stress, 77

Buehler

Загружено:

Сведения о документе

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Buehler

Загружено:

Авторское право:

Доступные форматы

Atomistic Modeling of Materials Failure

ISBN 978-0-387-76425-2 e-ISBN 978-0-387-76426-9

Printed on acid-free paper

Cambridge, MA Markus J. Buehler

List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . XVII

List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . LXIII

Part II Basics of Atomistic, Continuum and Multiscale Methods

2 Basic Atomistic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

2.2.2 The Concept of Computational Experiments . . . . . . . . 35

3 Basic Continuum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

4 Atomistic Elasticity: Linking Atoms and Continuum . . . . . 121

5 Multiscale Modeling and Simulation Methods . . . . . . . . . . . 157

5.4.2 Concurrent Integration of Tight-Binding,

Part III Material Deformation and Failure

6 Deformation and Dynamical Failure of Brittle Materials . 185

6.6 Crack Limiting Speeds of Cracks: The Signiﬁcance

6.11.5 Dynamical Fracture Mechanisms . . . . . . . . . . . . . . . . . . 311

7 Deformation and Fracture of Ductile Materials . . . . . . . . . . 327

8 Deformation and Fracture Mechanics

8.2.3 Rice–Thompson Model for Nucleation of Parallel

8.9.3 Atomistic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 456

1.1 Illustration of how the characteristic material scales of

1.7 Schematic of brittle (a) vs. ductile (b) materials behavior.

1.12 Mechanical deformation of a red blood cell (RBC) with

1.15 Experimental study of fracture mechanics of bone. Subplot

2.1 Molecular dynamics can be used to study material

2.2 This ﬁgure illustrates the concept of model building. Panel

2.8 Summary of the Metropolis–Hastings Monte Carlo

2.17 Chemical complexity in proteins involves a variety of

2.23 Water production at varying temperature, for constant

2.35 Summary of top 10 of the TOP500 supercomputer list, as of

2.39 Application of the energy method to visualize fracture

3.1 Axial tensile loading of a beam and schematic force–

3.12 Solution ﬁeld for a simply supported beam under dead

4.1 Example to illustrate Cauchy–Born rule in a one-dimensional

4.19 Entropy controlled molecular elasticity. Subplot (a) Coiled,

5.1 A summary of a hierarchical multiscale scheme that can

5.4 Hierarchical modeling of Cybersteel [18]. Subplot (a)

5.7 Application of the quasicontinuum method in the simulation

5.15 Hybrid CADD-Parallel Replica study of mechanical

6.1 Picture of shattered glass, a model for a brittle material . . . . . 185

6.8 Mode II loading experimental setup for studies of dynamic

6.18 Strain ﬁeld of a crack travelling in a material with

6.29 Principal strain ﬁeld at various crack velocities (a) v/cR ≈ 0,

6.36 This ﬁgure shows a continuously increasing hyperelastic

6.46 Calculation results of the Broberg problem. The plot

6.54 Subplot (a) schematic of stiﬀening materials behavior,

6.85 Crack tip velocity history during the mother–daughter–

6.100 Dependence of the average crack velocity in a single crystal

6.107 Crack dynamics in silicon without oxygen (O2 molecules)

7.1 This plot illustrates the diﬀerence between a screw

7.8 Concept of stacking fault energy, considering Peierl’s

7.16 Schematic of diﬀerent dislocation cutting processes. Subplot

7.31 Centrosymmetry analysis of the details of the dislocation

7.39 Summary of critical failure strains for diﬀerent cases studied

8.1 This graph shows a proposed deformation-mechanism map

8.4 Illustration of the maximum in the strength of

8.8 Experimental results of the thin ﬁlm strength dependence at

8.12 TEM micrographs of an unpassivated copper ﬁlm showing

8.24 Change of displacements in the vicinity of the diﬀusion

8.33 Generation of trails of point defects. Subplot (a): Dislocation

8.40 Nucleation of parallel glide dislocations, small grain sizes.

8.46 Simulation results of nanostructured material with twin

8.49 Flow stress σY vs. the ﬁlm thickness hf , as obtained

AEnsemble = ATime . (2.1)