You are on page 1of 54

Philosophical Magazine, 1 Nov–1 Dec 2003

Vol. 83, Nos. 31–34, 3475–3528

Multiscale modelling of nanomechanics and
micromechanics: an overview

Nasr M. Ghoniemy
Mechanical and Aerospace Engineering Department, University of California,
Los Angeles, California 90095-1597, USA

Esteban P. Busso
Department of Mechanical Engineering, Imperial College, London, UK

Nicholas Kioussis
Department of Physics, California State University Northridge,
Northridge, California 91330-8268, USA

Hanchen Huang
Department of Mechanical, Aerospace and Nuclear Engineering,
Rensselaer Polytechnic Institute, Troy, New York 12180-3590, USA

[Received 20 April 2003 and in final form 22 June 2003]

Abstract
Recent advances in analytical and computational modelling frameworks to
describe the mechanics of materials on scales ranging from the atomistic, through
the microstructure or transitional, and up to the continuum are reviewed. It is
shown that multiscale modelling of materials approaches relies on a systematic
reduction in the degrees of freedom on the natural length scales that can be
identified in the material. Connections between such scales are currently
achieved either by a parametrization or by a ‘zoom-out’ or ‘coarse-graining’
procedure. Issues related to the links between the atomistic scale, nanoscale,
microscale and macroscale are discussed, and the parameters required for the
information to be transferred between one scale and an upper scale are
identified. It is also shown that seamless coupling between length scales has not
yet been achieved as a result of two main challenges: firstly, the computational
complexity of seamlessly coupled simulations via the coarse-graining approach
and, secondly, the inherent difficulty in dealing with system evolution stemming
from time scaling, which does not permit coarse graining over temporal events.
Starting from the Born–Oppenheimer adiabatic approximation, the problem
of solving quantum mechanics equations of motion is first reviewed, with
successful applications in the mechanics of nanosystems. Atomic simulation
methods (e.g. molecular dynamics, Langevin dynamics and the kinetic Monte
Carlo method) and their applications at the nanoscale are then discussed.
The role played by dislocation dynamics and statistical mechanics methods
in understanding microstructure self-organization, heterogeneous plastic
deformation, material instabilities and failure phenomena is also discussed.

y Email: ghoniem@ucla.edu.

Philosophical Magazine ISSN 1478–6435 print/ISSN 1478–6443 online # 2003 Taylor & Francis Ltd
http://www.tandf.co.uk/journals
DOI: 10.1080/14786430310001607388

3476 N. M. Ghoniem et al.

Finally, we review the main continuum-mechanics-based framework used today
to describe the nonlinear deformation behaviour of materials at the local (e.g.
single phase or grain level) and macroscopic (e.g. polycrystal) scales. Emphasis is
placed on recent progress made in crystal plasticity, strain gradient plasticity and
homogenization techniques to link deformation phenomena simultaneously
occurring at different scales in the material microstructure with its macroscopic
behaviour. In view of this wide range of descriptions of material phenomena
involved, the main theoretical and computational difficulties and challenges are
critically assessed.

} 1. Introduction
Computational modelling of materials behaviour is becoming a reliable tool to
underpin scientific investigations and to complement traditional theoretical and
experimental approaches. In cases where an understanding of the dual nature of
the structure of matter (continuous when viewed at large length scales and discrete
when viewed at an atomic scale) and its interdependences are crucial, multiscale
materials modelling (MMM) approaches are required to complement continuum
and atomistic analyses methods. On transitional (or microstructure) scales
(in between continuum and atomistic), continuum approaches begin to break
down, and atomistic methods reach inherent time and length-scale limitations.
Transitional theoretical frameworks and modelling techniques are being developed
to bridge the gap between length-scale extremes. The power of analytical theories lies
in their ability to reduce the complex collective behaviour of the basic ingredients of
a solid (e.g. electrons, atoms, lattice defects, and single-crystal grains) into insightful
relationships between cause and effect. For example, the description of deformation
beyond the elastic regime is usually described by appropriate constitutive equations,
and the implementation of such relationships within continuum mechanics generally
relies on the inherent assumption that material properties vary continuously
throughout the solid. However, certain heterogeneities linked to either the micro-
structure or the deformation per se cannot be readily described within the framework
provided by continuum mechanics: dislocation patterns, bifurcation phenomena,
crack nucleation in fatigue, some non-local phenomena, etc. Some examples will
be discussed next to illustrate the role of MMM in nanomechanics and micro-
mechanics research.
Recent interest in nanotechnology is challenging the scientific community to
analyse, develop and design nanometre to micrometre-sized devices for applications
in new generations of computers, electronics, photonics and drug delivery systems.
These new exciting application areas require novel and sophisticated, physically
based approaches for design and performance prediction. Thus, theory and model-
ling are playing an ever-increasing role in these areas to reduce development costs
and manufacturing times. An important problem which concerns the microelectronic
industry is the reliable operation of integrated circuits, where the lifetime is limited
by the failure of interconnect wires in between submicrometre semiconducting chips.
In some cases, the nucleation and growth of even a single nanovoid can cause
interconnect failure. Statistical mechanics cannot adequately address this situation.
Future electronic and optoelectronic devices are expected to be even smaller, with
nanowires connecting a nano-sized memory and information storage and retrieval
nanostructures. Understanding the mechanics of such nano-engineered devices will

Multiscale modelling of nanomechanics and micromechanics 3477

enable high levels of reliability and useful lifetimes to be achieved. Undoubtedly,
defects are expected to play a major role in these nanosystems and microsystems
owing to the crucial impact on the physical and mechanical performance.
The potential of MMM approaches for computational materials design is also
great. Such possibility was recently illustrated on a six-component Ti-based alloy,
with a composition predetermined by electronic properties, which was shown to
exhibit an entirely new twin- and dislocation-free deformation mechanism, leading
to ‘unexpected new properties, such as high ductility and strength, and Invar and
Elinvar behaviour’ (Saito et al. 2003). Such an alloy would be unlikely to be found by
trial and error. This points to a paradigm shift in modelling, away from reproducing
known properties of known materials and towards simulating the behaviour of
possible alloys as a forerunner to finding real materials with these properties.
In high-pay-off, high-risk technologies, such as the design of large structures in
the aerospace and nuclear industries, the effects of ageing and environment on failure
mechanisms cannot be left to conservative approaches. Increasing efforts are now
focused on developing MMM approaches to design new alloys and material systems
in these areas. An illustration of an MMM based strategy for the development of
large components surrounding the plasma core of a fusion energy system is shown
in figure 1. The development of ultrastrong and yet ductile materials by combining
nanolayers with different microstructures also requires detailed understanding of
their mechanical properties. Such materials, if properly designed, may be candidates
for many demanding applications (e.g. microelectronics, optoelectronics, laser
mirrors, aircraft structures, rocket engines and fuel cells).
Appropriate validation experiments are also crucial to verify that the models
predict the correct behaviour at each length scale, ensuring that the linkages between
approaches are directly enforced. However, current nanoscale and microscale
mechanical experiments have been mostly limited to indentation (Loubet
et al. 1986, Wu et al. 1993) and bulge tests (Baker 1993, Small et al. 1994), and
to non-contact tests such as X-ray residual stress measurements (James 1980,

m Continuum
Mechanics
Statistical
mm Mechanics
Space

Dislocation
Dynamics
Monte
µm Carlo Z

σ = 165 Mpa
X Y

Classical 5000

MD
4000

3000
z(a)

2000

Quantum 1000

nm Mechanics
0

1000

2000
y(
a) 3000 3000
2000

)
1000
0

x(a
S S 4000 4000

3 3 5000 5000

b

fs ps ns s m
ms s ks
Time
Figure 1. Illustration of a MMM approach for the design of radiation-resistant materials for
fusion energy structures: MD, molecular dynamics.

3478 N. M. Ghoniem et al.

Segmueller and Murakami 1988). Multiscale interconnected approaches will need to
be developed to interpret new and highly specialized nanomechanical and micro-
mechanical tests. One of the advantages of these approaches is that, at each stage,
physically meaningful parameters are predicted and used in subsequent models,
avoiding the use of empiricism and fitting parameters.
As the material dimensions become smaller, its resistance to deformation is
increasingly determined by internal or external discontinuities (e.g. surfaces, grain
boundaries and dislocation cell walls). The Hall–Petch relationship has been widely
used to explain grain-size effects, although the basis of the relationship is strictly
related to dislocation pile-ups at grain boundaries. Recent experimental observations
on nanocrystalline materials with grains of the order of 10–20 nm indicate that
the material is weaker than would be expected from the Hall–Petch relationship
(Erb et al. 1996, Campbell et al. 1998). Thus, the interplay between interfacial or
grain-boundary effects and slip mechanisms within a single-crystal grain may result
in either strength or weakness, depending on their relative sizes. Although experi-
mental observations of plastic deformation heterogeneities are not new, the signifi-
cance of these observations has not been addressed until very recently. In the same
metallic alloys, regular patterns of highly localized deformation zones, surrounded
by vast material volumes which contain little or no deformation, are frequently seen
(Mughrabi 1983, 1987, Amodeo and Ghoniem 1988). The length scale associated
with these patterns (e.g. typically the size of dislocation cells, the ladder spacing
in persistent slip bands (PSBs) or the spacing between coarse shear bands) controls
the material strength and ductility. As it may not be possible to homogenize such
types of microstructure in an average sense using either atomistic simulations or
continuum theories, new intermediate approaches will be needed.
The issues discussed above, in addition to the ever-increasingly powerful
and sophisticated computer hardware and software available, are driving the
development of MMM approaches in nanomechanics and micromechanics. It is
expected that, within the next decade, new concepts, theories and computational
tools will be developed to make truly seamless multiscale modelling a reality.
In this overview article, we briefly outline the status of research in each component
that makes up the MMM paradigm for modelling nanosystems and microsystems:
quantum mechanics (QM), molecular dynamics (MD), Monte Carlo (MC) method,
dislocation dynamics (DD), statistical mechanics (SM) and, finally, continuum
mechanics (CM).

} 2. Computational quantum mechanics

2.1. Essential concepts
There is little doubt that most of the low-energy physics, chemistry, materials
science and biology can be explained by the QM of electrons and ions and that,
in many cases, the properties and behaviour of materials derive from the quantum-
mechanical description of events at the atomic scale. Although there are many
examples in materials science of significant progress having been made without
any need for quantum-mechanical modelling, this progress is often limited as one
pushes forward and encounters the atomic world. For instance, the understanding of
the properties of dislocations comes from classical elasticity theory, but even in these
cases, until recently, very little was known about the core of a dislocation or the
effect of chemistry on the core, precisely because this part of the dislocation requires

Multiscale modelling of nanomechanics and micromechanics 3479

detailed quantum-mechanical modelling. The ability of QM to predict the total
energy and the atomic structure of a system of electrons and nuclei enables one to
reap a tremendous benefit from a quantum-mechanical calculation. Many methods
have been developed for solving the Schro¨dinger equation that can be used to
calculate a wide range of physical properties of materials, which require only a
specification of the ions present (by their atomic number). These methods are usually
referred to as ab initio methods.
In considering the motion of electrons in condensed matter, we are dealing with
the problem of describing the motion of an enormous number of electrons and nuclei
(about 1023) obeying the laws of QM. Prediction of the electronic and geometric
structure of a solid requires calculation of the quantum-mechanical total energy of
the system and subsequent minimization of that energy with respect to the electronic
and nuclear coordinates. Because of the large difference between the masses of
electrons and nuclei and the fact that the forces on the particles are the same, the
electrons respond essentially instantaneously to the motion of the nuclei. Thus, the
nuclei can be treated adiabatically, leading to a separation of the electronic and
nuclear coordinates in the many-body wave function

ðfRI , rn gÞ ¼ el ðfrn g; fRI gÞ nuc ðfRI gÞ, ð1Þ

which is the so-called Born–Oppenheimer approximation (Parr and Yang 1989,
Dreizler and Gross 1990). The adiabatic principle reduces the many-body problem
in the solution of the dynamics of the electrons in some frozen-in configuration {RI}
of the nuclei; this is an example of the philosophy behind systematic degree-of-
freedom elimination which is itself at the heart of multiscale modelling.
Even with this simplification, the many-body problem remains formidable. The
Hamiltonian for the N-electron system moving in condensed matter with fixed nuclei
is (in atomic units)

X
N X
N XX 1
H¼ 12=2i þ vi ðri Þ þ 12 , ð2Þ
i¼1 i¼1 i j6¼i
jri  rj j

where the first term is the kinetic energy operator, v ðrÞ is the one-electron spin-
dependent external potential (e.g. the electron–nucleus interaction), and the third
term represents the effect of the electron–electron interactions which poses the most
difficult problem in any electronic structure calculation.
Density functional theory (DFT) (Hohenberg and Kohn 1964, Kohn and
Sham 1965, Jones and Gunnarsson 1989, Parr and Yang 1989, Dreizler and Gross
1990) has been proven to be a very powerful quantum-mechanical method in
investigating the electronic structure of atoms, molecules and solids. Here, the
electron density ðrÞ, or spin density  ðrÞ, is the fundamental quantity, rather
than the total wave function employed for example in Hartree–Fock theory
(Ashcroft and Mermin 1976). The DFT makes successful predictions of ground-
state properties for moderately correlated electronic systems (Perdew 1999). DFT
can provide accurate ground-state properties for real materials (such as total energies
and energy differences, cohesive energies of solids and atomization energies of mole-
cules, surface energies, energy barriers, atomic structure, and magnetic moments)
and provides a scheme for calculating them (Jones and Gunnarsson 1989, Fulde
1993, Perdew 1999).

The exchange– correlation energy is ‘nature’s glue’. Ghoniem et al. formally.3480 N. then taking the functional derivative with respect to the density would produce an exchange– correlation potential that included the effects of exchange and correlation exactly. It is largely responsible for the binding of atoms to form molecules and solids. The Kohn–Sham equations must be solved self-consistently so that the occupied electron states generate a charge density that produces the electronic potential that was used to construct the equations. The complexity of the real many-body problem is contained in the unknown exchange–correlation potential vxc ðrÞ. Indeed the simplest . 1992). of an electron gas (even in the presence of a static external potential) is a unique functional of the electron density. These are much more efficient than the traditional diagonalization methods. ð5Þ k where ðxÞ is the step function and  is the chemical potential. #  vxc ðrÞ  : ð4Þ d ðrÞ Here. ð3Þ jr  r0 j where k ðrÞ are the Kohn–Sham orbitals. including exchange and correlation. Kohn and Sham (1965) then showed how it is possible. Hohenberg and Kohn (1964) proved that the total energy. Payne et al. and the spin-dependent exchange– correlation potential is defined as dE xc ½" . New iterative diagonalization approaches can be used to minimize the total-energy functional (Car and Parrinello 1985. Gilan 1989. The minimum value of the total-energy functional is the ground-state energy of the system. We next give a brief discussion of the essential concepts of DFT. The Kohn–Sham equations represent a mapping of the interacting many- electron system on to a system of non-interacting electrons moving in an effective non-local potential owing to all the other electrons. If the exchange–correlation energy functional were known exactly. making simple approximations. to replace the many-electron problem by an exactly equivalent set of self-consistent one-electron equations  ð 0  1 2 3 0 ðr Þ  2= þ v ðrÞ þ d r þ vxc ðrÞ k ðrÞ ¼ k k ðrÞ. #) is found by summing the squares of the occupied orbitals: X  ðrÞ ¼ j k ðrÞj2 ð  k Þ. and the density that yields this minimum value is the exact single-particle ground-state density. Exc ½" . #  is the exchange–correlation energy and the density  ðrÞ of electrons of spin (¼ " . M. ð7Þ k  where ð ð ðrÞðr0 Þ U½ ¼ 1 2 d r d 3 r0 3 ð8Þ jr  r0 j is the Hartree self-repulsion of the electron density. the total energy is X Xð 3 E¼ h k j  12=2 j k ið  k Þ þ d r v ðrÞ ðrÞ þ U½ þ Exc ½" . we can hope to circumvent the complexity of the problem. Nevertheless. # . The total electronic density is ðrÞ ¼ " ðrÞ þ # ðrÞ: ð6Þ Aside from the nucleus–nucleus repulsion energy.

have been proven (Kohn and Sham 1965. # ¼ d3 r f ð" . These parametrizations use interpolation formulae to link exact results for the exchange–correlation energy of high-density electron gases and calculations of the exchange–correlation energy of intermediate and low-density electron gases. Recent work has shown that this success can be partially attributed to the fact that the LDA gives the correct sum rules for the exchange–correlation hole (Hariss and Jones 1974. non-local information about the charge or spin density could be provided from the gradient terms. von Barth and Hedin 1972) very successful. GGAs have made DFT popular in quantum chemistry. Perdew et al. The LDA or LSDA. in principle. 1996) ð GGA   Exc " . the Perdew–Wang generalized gradient approximation (GGA) (Perdew 1991. These type of approximation are in general referred to as gradient expansion approximations (Perdew 1999). =" . In the LSDA the exchange–correlation energy of an electron system is constructed by assuming that the exchange–correlation energy xc per electron at a point r in the electron gas is equal to the exchange–correlation energy per electron in an electron gas of uniform spin densities " . ð LSDA   Exc " . nothing has been said about the way that k ðrÞ itself will be determined. =# Þ ð10Þ starts from the second-order gradient expansion of the exchange–correlation hole density and then cuts off the spurious long-range (jr  r0 j ! 1) parts to restore the sum rules. Many methods share the common feature that the wave function is presumed to exist as a linear combination of some set of basis functions which might be written generically as X ðrÞ ¼ n n ðrÞ.2. that is the local-density (LDA) or local-spin-density (LSDA) approximation. 1980. # ¼ d3 r ðrÞunif xc ð" ðrÞ# ðrÞÞ: ð9Þ The LSDA assumes that the exchange–correlation energy functional is purely local. Several parametrizations exist for the exchange–correlation energy of a homogenous electron gas all of which lead to total-energy results that are very similar (Kohn and Sham 1965. The GGA is typically more accurate than LSD. 1999. Hedin and Lundqvist 1971. The most popular. In this way. # . # . Considering the inexact nature of the approximation.3. it is remarkable that calculations performed using the LDA have been so successful. 2. Basis functions In the discussion of the Kohn–Sham equations earlier. especially for the rapidly varying densities of atoms and molecules. Langreth and Perdew 1977). ð11Þ n . A number of attempts to improve the LDA or LSDA use gradient expansions of the charge or spin density. Gunnarsson and Lundqvist 1976. Multiscale modelling of nanomechanics and micromechanics 3481 possible approximation. namely. The local density approximation This simplest method of describing the exchange–correlation energy of an electron system has been the workhorse of condensed-matter physics for almost 30 years. Perdew and Zunger 1981). In recent years. 2. Vosko et al. ignores corrections to the exchange–correlation energy at a point r due to nearby inhomogeneities in the electron gas.

chemisorption of molecules on surfaces (Du¨rr et al. Ruberto et al. such as those containing point. 1992. Similar arguments hold with respect to the functional forms of the n ðrÞ. 2002). grain boundaries (Wu et al. Denteneer and van Haeringen 1985).4. Yildrim et al. There are several powerful techniques in electronic structure calculations. 2001) and quantum dots (Wang et al. 1999. Janotti et al. interfaces (Batirev et al. or quantum dots if a periodic supercell is used. 2001). M. such as the full-potential linear-augmented-plane- wave method (Wimmer et al. 1981). 1999). where n is the weight associated with the nth basis function n ðrÞ. the so-called plane-wave basis set (Payne et al. The implementation of the DFT within the LDA or the GGA for the exchange– correlation energy. a compromise will have to be made between high-accuracy results. structural and magnetic proper- ties of surfaces (LaBella et al. alloys (Kent and Zunger 2001. The solution of the Kohn–Sham equations in turn becomes a search for unknown coefficients rather than unknown functions. the full-potential linear muffin-tin-orbital method (Price and Cooper 1989. 2001. leading to the unzipping of the nanotube wall (Scudder et al. 1992) and the full-potential Korringa–Kohn–Rostoker method (Papanikolaou et al. The independence of defects in neighbouring cells can be checked by increasing the volume of the supercell until the calculated defect energy is converged. 2. Representative successful applications of DFT for a wide variety of materials ranging from metals to semiconductors to ceramics include the electronic. It is essential to include enough bulk solid (or vacuum) in the supercell to prevent the defects in neighbouring cells from interacting appreciably with each other. Lu and Kioussis 2001). Periodic boundary conditions are applied to the supercell so that the supercell is reproduced throughout space.3482 N. the development of new linearized methods for solving the single- particle Schro¨dinger equations and the use of powerful computers allow for a highly efficient treatment of up to several hundreds of atoms and has led to an outburst of theoretical work in condensed-matter physics and materials physics. The various methods may be divided into those which express the wave functions as linear combinations of some fixed basis functions. Alternatively. Price et al. Lu et al. and computational costs. 1999. In deciding on a particular set of basis functions. Part of the variety associated with the many different implementations of DFT codes is associated with the choice of basis functions and the form of the effective crystal potential. C nanotubes (Chan et al. 1994. 2002). one can study even molecules (Rappe et al. Nanomechanics applications Electronic structure calculations based on DFT also can be applied for studies of non-periodic systems. 2002). say linear combination of atomic orbitals localized about each atomic site (Eschrig 1989) or those which use a free-electron-like basis. The supercell contains the defect sur- rounded by a region of bulk crystal or vacuum for the case of a surface. For example. provided that the supercell is large enough that the interactions between the molecules are negligible. 1999). which favour small basis sets. 1992). Using the supercell geometry. which require a large basis set. Ghoniem et al. 2003). there are those methods which employ matching of partial waves. planar or line defects. This remarkable hydrogen-induced unzip- ping mechanism lends strong support to the recent experimental observations . recent ab initio calculations predict that two rows of hydrogen atoms chemisorbed on selective sites exterior to an armchair carbon nanotube catalyses the breaking of the nearest-neighbour C–C bond of the nanotube through the concerted formation of C–H bonds.

Alternatively. Other systems with anomalous electronic and magnetic properties are heavy-fermion compounds. This approach represents a combination of an ab initio treatment of the interactions across the slip plane and an elastic treatment of the continua on either side of the slip plane. The resultant approach can then be applied to problems associated with dislocation core structure and the cross slip process. and certain insulating transition-metal oxides. The  surface for both the pure aluminium and the aluminium þ hydrogen system. not only facilitates dislocation emission from the crack tip but also enhances the dislocation mobility dramatically. the correlations among the electrons are stronger than might be expected from the local spin densities. All these approaches are based on LDA as a starting point and introduce additional terms intended to treat Coulomb correlations between electrons. In addition. the self-interaction correction method (Perdew and Zunger 1981. Svane and Gunnarson 1990). this approach is particularly useful for studying the interaction of impurities with dislocations when empirical potentials are either not available or not reliable to deal with such multielement systems. 2001). etc. as in the case for 4f (5f) states in rare-earth (actinide) atoms and sometimes for 3d states of transition-metal atoms. which is in disagreement with experiment. shown in figures 2 (a) and (b). 2001. in turn. there has been a growing sense that realistic LDA-like approaches may be generalized to include the most significant correlation effects for strongly correlated electron systems of the Mott–Hubbard and charge- transfer variety (Anisimov 2000). One of the many successful applications of this approach was the elucidation of the atomistic mechanism responsible for the hydrogen-enhanced local plasticity in aluminium (Lu et al. Lichtenstein et al. Therefore. 2000. only limited information about spectroscopic properties is available from such ground-state calculations. colossal magnetoresistance manganites. In some systems. 2002). This occurs in systems where the electrons partially preserve their localized atomic-like nature. which LDA predicts to be metallic. which neither atomistic nor conventional continuum models could handle separately. This. copper-oxide-based superconductors. as they . 2. ab initio calculations have been recently applied to study the dislocation core properties of isolated h111i screw dislocations in bcc molybdenum and tantalum (Woodward and Rao 2002). Limitations of the density functional theory The DFT within the LSDA and GGA generally works well for ground-state properties of moderately correlated electron systems. Quasiparticle excitations. however. 1998). In the past few years. Typical examples include rare-earth metals. 1995). Furthermore. respectively. Among these are the LDA þ U method (Anisimov et al. the LDA þ dynamical mean-field theory method (Georges et al. 1996) and the optimized effective potential method (Gross et al. Finally. 1991. it allows us to study general trends in dislocation core properties and to correlate them with specific features of the underlying electronic structure. Multiscale modelling of nanomechanics and micromechanics 3483 (Nikolaev et al. where LDA predicts a strong peak of 4f-orbital in the density of states at the Fermi level. by providing a means to incorporate information obtained from ab initio calculations directly into continuum models (Lu et al. indicate an overall reduction in energy in the presence of hydrogen.5. 1997) for the coalescence of single-walled nanotubes in the presence of atomic H. the Peierls–Nabarro model has come to serve as a link between atomistic and continuum approaches.

(After Lu et al.1 0 (b) Figure 2. Ghoniem et al.1 0 (a) 0.2 0.5 0. A major achievement in the field was given by the GW approx- imation (where G stands for the Green’s function and W for the screened self energy .3 0. The g (J m2) surfaces for displacements along a (111) plane for (a) pure aluminium and (b) aluminium þ hydrogen systems.3 0. 0. which correspond to electron–hole excitations in the electron system. A time-dependent extension of DFT (Gross et al.3484 N. but so far the method is limited to finite systems. Excitation properties have also been studied using the configuration interaction method (Dykstra 1998) but. this method is very much restricted to small systems such as small molecules or clusters. In fact. the band structures given by the LDA are often in distinct disagreement from experimental data.4 0. similarly.) occur in photoemission and tunnelling experiments. (2001). optical spectra. showing systematic deviations of band dispersions and band gaps. 1998) can treat excitation properties. M. cannot be described by straightforward use of DFT or other ground-state theories. In partic- ular.4 0. two-particle excitations are not obtained with satisfying accuracy either.2 0. In addition to the failures of the band structure.5 0. are not fully described by the Kohn–Sham eigenvalues.

Goedecker 1999) for . For example. which includes physical quantities such as the lattice constant. 2. 1993. Highly accurate band structures for real materials have been obtained by this method. metals. Li et al. DFT-based generalized pseudopotential theory (Moriarty 1988. such as transition metals. TB MD implements an empirical parametrization of the bonding interactions based on the expansion of the electronic wave functions on a very simple basis set. Empirical potentials involve the fitting of parameters to a predetermined experi- mental or ab initio database. reliable interatomic potentials usually are not available for multielement materials and for systems containing substitutional or interstitial alloying impurities. 1984) or the Finnis–Sinclair (1984) type in metals. Moriarty et al. However.7. the vacancy formation energy and the surface energy. have been extremely useful in investigating generic phenomena in simple systems. which allows for a very accurate evaluation of the self-energy on the basis of results from a preceding DFT calculation. thus making the GWA a standard tool in predict- ing the electron quasiparticle spectrum of moderately correlated electron systems (Louie 1996). the elastic constants. semiconductor sur- faces. This is especially so for directionally bonded systems. Mehl and Papaconstantopoulos 1996) and the self-consistent-charge density functional TB MD approach is becoming widespread in the atomistic simulation community. thereby allowing large-scale atomistic simula- tions for millions of atoms and a simulation time of nanoseconds. insulators. Ordejon et al. Recently. they may not provide the desired physical accuracy for many real complex materials of interest. Quantum-based interatomic potentials for transi- tion metals that contain explicit angular-force contributions have been developed from first-principles. that can be applied to large-scale atomistic simulations. 1993. including bulk semiconductors. novel analytic bond-order potentials have been derived for atomistic simulations by coarse graining the elec- tronic structure within the orthogonal two-centre TB representation (Pettifor 1989. 2000). Pettifor and Oleinik 1999. 1994. 1998). employing either the embedded-atom method (EAM) type (Daw and Baskes 1983. 2002). because it allows one to evaluate both ionic and electronic properties (Frauenheim et al. atoms. derived from QM. Moriarty and Widom 1997. Thus. Multiscale modelling of nanomechanics and micromechanics 3485 correction) (Hedin and Lundqvist 1969) for the self-energy. and the Stillinger–Weber (1985) type (Ackland and Vitek 1990) or the Tersoff (1986) type in covalent materials. Linear scaling electronic structure methods A wealth of efficient methods. The success of TB MD stands on a good balance between the accuracy of the physical representation of the atomic interactions and the resulting computational cost. Connections to interatomic potentials The direct use of DFT methods to perform full quantum MD simulations on real materials is limited to a hundred or so atoms and a simulation time of a few picoseconds. and for chemically or structurally complex systems. The tight-binding (TB) (Cohen et al. defects and clusters. the next step of coarse graining the problem is to remove the electronic degrees of freedom by imagining the atoms to be held together by some sort of glue or interatomic potential.6. There is consequently growing need to develop more accurate interatomic potentials. at the same time. such as intermetallics and alloys. have recently been developed (Yang 1991. Such simulations. 2. the so-called order-N (O(N)) methods.

Based on the electronic structure database.e. the molecular crash simulation of C60 fullerenes colliding with a diamond surface (Galli and Mauri 1994). or alternatively using experimental measurements of specific properties. Most O(N) algorithms are built around the density matrix or its representation in terms of Wannier functions (Marzari and Vanderbilt 1997) and take advantage of its decay properties. Ghoniem et al. To determine the properties of an ensem- ble of atoms larger than can be handled by computational QM. where N can be defined to be the number of atoms or the volume of the system. All the algorithms that would allow us to treat very large basis sets within DFT have certain shortcomings. they also generate a database of properties that can be used to construct effective (empirical) interatomic potentials. The basic idea is to eliminate all electronic degrees of freedom and to assume that the electrons are glued to the nuclei. the widely used Carr–Parinello (1985) MD methods and various minimization tech- niques (Payne et al. electrons). Large-scale atomistic simulations The ab initio methods presented in } 2 are rigorous but limited by present-day computers to systems containing a few hundred atoms at most. 1998) and the geometric structure of large bimolecules (Lewis et al. For simplicity the localization region is usually taken to be a sphere. the dynamic evolution of all atoms can be fully determined by numerical integration. force and temperature). the irradiation- mediated knockout of carbon atoms from carbon nanotubes (Ajayan et al. The use of O(N) methods within DFT methods is not yet widespread. the interaction between two atoms is represented by a potential function that depends on the atomic configuration (i. the surface reconstruction properties of silicon nanobars (Ismail-Beigi and Arias 1999). Firstly. To obtain linear scaling. once the positions and velocities of atoms in the finite ensemble within the simu- lation cell are known. Thus. they provide direct information on the response of materials to external environments (e. In principle. The implementation and practice of MD simulations are more involved than the conceptual description alluded to here. This introduces the concept of a localization region. Thus. calculating the electronic properties of an extended system that require an amount of computation that scales linearly with the size N of the system. } 3. Only inside this localization region is the quantity calculated. According to classical Newtonian mechanics. even though the optimal shape might be different. 1996). one has to cut off the exponentially decaying quantities when they are small enough. Such methods serve two important purposes. The MD method is developed to enable studies of the properties of material volumes containing millions to billions of atoms with effective interatomic potentials. M. these methods are an essential tool for the calculation of the electronic structure of large systems containing many atoms. An illustration of the wide range of areas where O(N) methods have made possible the study of systems that were too large to be studied with conventional methods include the 90 partial dislocation in silicon (Nunes et al. approximate effective potentials can be constructed. all thermodynamic properties can be readily extracted. outside it is assumed to vanish. Secondly. the description of the atomic interactions must be approximated. O(N) methods have become as essential part of most large-scale atomistic simulations based on either TB or semiempirical methods. A successful simulation depends on three .g. 1997). 1992) are limited because the computational time required scales as N3 .e. On the contrary. relative displacement) and the local environment (i.3486 N.

The numerical integration of the equations of motion is performed by either explicit or implicit methods. and stress and temperature control to mimic physically meaningful thermodynamic ensembles. The following presentation outlines the basic equations and princi- ples of the MD method. employment of appropriate boundary conditions. this set of equations is converted for small time increment dt to Ri ðt þ dtÞ ¼ Ri ðtÞ þ dtVi ðt þ 12dtÞ. . In the explicit Verlet’s leapfrog method. the equation of motion is given by d2 Ri Mi ¼ Fi ¼  =Ri F. and (iii) the analysis of massive data resulting from computer simulations. the equation of particle motion is split into two first-order equations: dRi dVi F ¼ Vi . we briefly present key concepts in the numerical implementation of MD methods and then discuss salient aspects of the last two topics. Solving these second-order ordinary differential equations for all the atoms in a simulation cell may appear simplistic. ð12Þ dt2 which is derived from the classical Hamiltonian of the system: X Mi Vi2 H¼ þ F: ð13Þ 2 An atom of mass Mi moves as a rigid particle at the velocity Vi in the effective potential FðRi Þ of other particles. Another popular implicit integration method for MD simulations is the predictor-corrector scheme. 3. In the following. before proceeding to its numerical implementation. The molecular dynamics method The MD method is about the simultaneous motion and interaction of atoms (or molecules). in particular the Gear (1971) algorithm. The dynamic evolution of the system is governed by classical Newtonian mechanics. The atomic force Fi is obtained as the negative gradient of the effective potential: Fi ¼  =Ri F.1. ð15Þ Fi Vi ðt þ 12dtÞ ¼ Vi ðt  12dtÞ þ dt : ð16Þ Mi The Verlet algorithm is very popular in MD simulations because it is stable and memory-efficient and allows for a reasonably large time-step. A physical simulation involves proper selection of numerical integration scheme. Multiscale modelling of nanomechanics and micromechanics 3487 major factors: (i) the computational implementation of the MD method. ¼ i : ð14Þ dt dt Mi Based on the half-step leapfrog method. where for each atom i. A proper numerical integration scheme should be both numerically stable and computationally efficient. The simple Euler scheme is not appropriate for MD simulations because it lacks numerical stability. (ii) the construction of accurate interatomic potentials.

in which atoms follow Newtonian dynamics according to the MD method outlined earlier. When one dislocation is included in the simulation cell. 1998). is more commonly used in dislocation simulations. the direct linking of atomistic and continuum regions (Oritz and Phillips 1999). Hoagland et al. M. In contrast with the two approximate boundary conditions. the initial strain field. boundary atoms are not allowed to relax. According to this condi- tion. However. The . The innermost region is the MD region. Woo and Puls 1976. which are more complex. The rigid boundary condition is probably the simplest. Hoagland 1976. because of limited computational power. For simple dislocation configurations. allowing the simulation of a single straight dislocation. Ghoniem et al. Simulations for 109 atoms represent the upper limit of computation today with simple interatomic potentials (for examples see Lennard- Jones (1924)). Various boundary conditions are used in mechanics simu- lations.3488 N. In the 1970s. the other rigorous scheme. At present. 2000). This rigidity of the boundary condition naturally leads to inaccuracy in the simulation results. the periodic boundary condition (Kido et al. Since a periodic boundary condition is applied along the dislocation line. it is difficult to extend such modifications to general dislocation config- urations. periodic boundary conditions are applied along the dislocation line. 1987). say of an infinitely long straight dislocation. Total simulation times are typically less than 10 ns as a result of short integration time steps in the femtosecond range. 2000). The boundary of a simulation cell is usually within close proximity to each simulated atom. In these simulations. Since dislocations dominate the linking of nanomechanics and micromecha- nics. namely the flexible boundary condi- tion. Rao et al. the direct linking scheme should be the most reliable and desirable boundary con- dition. the two- dimensional version of this boundary condition was implemented (Sinclair 1971. and the flexible (Green’s function) boundary condition (Sinclair 1971. Woo and Puls 1976). The boundary of a simulation cell may not be a real physical interface. The inaccuracy also exists in the application of the periodic boundary condition. Once numerically robust and efficient. During subsequent simulations. boundary atoms are fixed during MD simulations. making the displacement two dimensional. the simulation cell is divided into three regions. This scheme has gone through the implementations in two and three dimensions. the following presentation focuses on those boundary conditions relevant to the long-range strain fields of dislocations. an infinite number of its images are included in the entire space because of the infinite repetition of the simulation cell. These include the rigid boundary condition (Kuramoto et al. On the plane normal to the dislocation line. and to subtract the image effects. This idea has been extended to include direct linking of atomistic and electro- nic structure regions (Abraham et al. in which the force on each atom is calculated and then used to generate displacements of all atoms in the simulation cell according to the Green’s function of displacement. line forces are calculated in the flexible region. the simulation cell is repeated periodically to fill the entire space. proper use of boundary conditions is crucial to obtaining a physically meaningful MD simulation. Similar to the numerical integration scheme. The intermediate region is the flexible region (or Green’s function region). the direct linking and flexible boundary conditions can be rigorous. Modifications have been made to include dipoles in a simulation cell. 1976. is imposed on all atoms in the simulation cell. rather it is merely the interface of the simulation cell and its surroundings. The direct linking of atomistic and continuum region (Qrtiz and Phillips 1999) aims at seamless bridging of two different scales. Here.

2001) has been developed. However. 1998). this response is consistent with linear elasticity. This transition problem exists in other tempera- ture control mechanisms. success- ful MD simulations rely on the proper control of thermodynamic variables: stress and temperature. when they interact with the simulation cell boundaries or borders. as derived from interatomic interactions. The latter aspect is solely determined by prescribed interatomic potentials and is more crucial in obtaining physically meaningful results. Using frictional forces to add or subtract heat. Multiscale modelling of nanomechanics and micromechanics 3489 outermost region contains atoms that serve as background for force calculations in the flexible region. The driving force of the imaginary atoms is the imbalance of internal and external stresses. elastic waves are emitted and. such a transition is generally too fast compared with realistic processes. In general. a tabulation and interpolation scheme (Golubov et al. As a result of renewed interest in DD. the method says nothing about how simulated atoms interact with each other. Cai et al. During dislocation motion. When the temperature is uniform in space and constant in time. the strength of the frictional force dictates the transition time. is not balanced by the external stress. In this formulation. the simulation cell deforms accordingly. Flexible or Green’s function bound- ary conditions have clear advantages in allowing for full relaxation of one or a few dislocations in a simulation cell. little difficulty exists.2. the Nose (1984)–Hoover (1985) method provides a mechanism of controlling the temperature of a simulation cell. they are reflected and can lead to interference or even resonance in the simulation cell. At low stress levels. However. is controlled through the kinetic energy or velocities of all atoms. However. This large fluctuation may introduce artefacts in the simulation of kinetic process. When the temperature changes. the three vectors describing the shape of simulation cell are equated to position vectors of three imaginary atoms. they are not applicable to truly dynamic simulations. the lattice constant may fluctuate by about 0. but a fully satisfactory solution is still not available. Recently. With this improvement. Interatomic potentials A proper MD method is a necessary condition for physically meaningful simu- lations. Approximate approaches (Ohsawa and Kuramoto 1999. This formulation works well when an equilibrated stress state is sought. Under general stress conditions. 2000) have been proposed to damp the waves at the boundary. A fast transition is usually necessary because of the short time reachable in MD simulations. However. When the internal stress. The other thermo- dynamic variable. flexible boundary condi- tions have been extended to three dimensions and are often referred to as the Green’s function boundary conditions (Rao et al. improving the computational efficiency by two orders of magnitude and yet maintaining the accu- racy of linear elasticity. there is a compromise between the potential rigorousness and the computational . and caution should be exercised. and the movements of those atoms are governed by Newton’s equations of motion. the Green’s function boundary conditions work well for static simulations of dislocations. Green’s function calculations are time-consuming limiting applications of the Green’s function boundary condition. temperature. however. such as velocity scaling or introduction of random forces. without suffering from the artefacts of image dis- locations. the formulation of Parrinello and Rahman (1981) provides a mechanism to track the deformation.5% around its equilibrium value. large fluctuations in the deformation are unavoidable according to the Parrinello–Rahman method. However. In addition to the numerical integration and boundary condition issues. 3.

It is worth pointing out that these interatomic potentials apply for specific classes of materials. pair potentials are applicable to simulations of rare- gas behaviour. can be enhanced by the action of stress (Woo et al. Strictly speaking. are obtained from ab initio calculations and reliable experiments. efficiency. carbon and other covalent systems. which can be easily determined from crystal properties. On the other hand. However. With the increasing demand on accuracy and available computational power. this advantage is accom- panied by the inability to describe covalent systems. have very few fitting parameters. their applicability in studying defects is by default questionable. the bond order potential (Pettifor 1989) and the inversion method (Chen 1990. like the EAM. 2002). Potentials that have a QM basis. Since these potentials are fitted to only a few perfect crystal properties. In studying silicon. We start with point defects. Applications of atomistic simulations in mechanics This section focuses on the applications of MD simulations to mechanics problems at the nanoscale. pair potentials. diamond. The stress effect on a point defect. use several or many defect properties to determine the EAM functions. include the well-known Stillinger–Weber (1985) potential and the Tersoff (1986) potential. the Tersoff (1986) potential. particularly the force matching approach (Ercolessi and Adams 1994). The diffusion process of a point defect has been investigated by a combination of molecular statics (MS) and MD simulations. many-body potentials such as the Finnis–Sinclair (1984) potential and the EAM (Daw and Baskes 1983. these potentials suffer from non-transferability. the presentation is focused on nano- mechanics of point defects.3. in either analy- tical or tabular form. The more sophisticated potentials. which is intrinsic in low-symmetry crystals. . particularly the Stillinger–Weber (1985) potential.3490 N. The EAM type of potential has proven to be a good choice for simple metals. For high computational efficiency. Zhang et al. Angle-dependent potentials. 1997) have also been used for greater accuracy. in particular its formation and migration energies. 1994). Materials properties. However. which are also many-body. but this is usually not a first-order effect in nanomechanics. such as the generalized pseudopotential (Moriarty 1988. such as the Lennard- Jones (1924) and the Morse (1929) potentials are used. in which angular dependence dominates. and the bond-order potential (Pettifor 1989) are among the leading candidates. the mod- ification by Pasianot et al. applications to simple metal systems can also provide qualitative guidance. the normal diffusion process governed by point defect motion is polarized by the action of the stress field. silver and iron (except its magnetic properties) have been very successful. like the Lennard-Jones potential. 1984) have been commonly used. angular dependent potentials. Applications to metals such as aluminium. Ghoniem et al. interatomic potentials are empirical or semiempirical and thereby have fitting parameters. 3. In the same category are effective-medium and glue models (Ercolessi et al. Furthermore. and applied to SiC by Huang et al. Simpler potentials. A modification of the EAM has been proposed to include the angular dependence by Baskes (1987). (1991) is similar. Diffusion anisotropy. 2003) but will not be treated here. (1995). of either the perfect crystal or the defec- tive structures. dislocations and interfaces (free surfaces and grain boundaries) under applied stress. is crucial to the analysis of kinetic processes. copper. 1986). Mechanics of nanotubes have also received much attention (Dumitrica et al. Under stress. In general. The radial function form of the EAM is computationally advantageous. M.

In dislocation statics. Later. nucleation of dislocations from surfaces (Liu et al. Multiscale modelling of nanomechanics and micromechanics 3491 The second type of defect. In the transonic velocity range. During dislocation motion. b) and dislocation nucleation in nano- crystals (Van Swygenhoven et al. Li and Shi 2002). The availability of fast dislocations also enabled the study of disloca- tion dipole stability at the atomic level (Wang et al. Cleri et al. just to name a few. dislocation–dislocation (Xu and Moriarty 1998. to serve as direct con- firmation of the elasticity analysis (Huang et al. (2001b). 2001b). MD simulations have been applied in studies of dislocation emission from crack tips (Bulatov et al. Rao and Woodward 2001. Rodney and Phillips 1999. In addition to the static and dynamic behaviours. which is a direct result of the finite speed of wave propagation. dislocation–point-defect (Justo et al. 2001) and dislocation–grain boundary (Ortiz and Phillips 1999) interaction processes. 2001a). One of the challenging Figure 3. Duesbery and Vitek 1998. 2002a. showing the stabilization of a dipole despite the dynamic overshoot. In the following. Zhou et al. has received much attention in nanomechanics. namely dislocations. 1995). 1998). Gumbsch and Gao (1999) demonstrated that transonic dislocations are possible. 2001. 1998. it was shown that a transonic dislocation can cross the sound barrier back and forth (Shi et al. MD simula- tions have revealed new mechanisms and possibilities. 1999). the simulations further reveal that a dislocation dipole can be stabilized even with an overshoot. 1997). Wang et al. As shown in figure 3. many processes are involved. Three snapshots of a pair of edge dislocations in a dipole configuration. MD simulations have provided much insight to details of kink–kink (Bulatov et al. the nucleation of dislocations is another important area of nanomechanics. The discovery of threefold symmetry of a screw dislocation in bcc metals is a beautiful example of applications of dislocation statics (Vitek 1974. dislocation damping. Rodney and Martin 2000). Xu and Moriarty 1996.) . dislocation intersection and dislocation clustering. (After Wang et al. one is primarily concerned with the final core structure of one or a few dislocations. we analyse the nanomechanics of dislocations in terms of statics and dynamics. 2002.

1999) have investigated mechanisms of grain-boundary interactions. Taking thin slices of polycrystalline nanograins. Wolf. At the same time. Ghoniem et al. sur- faces are nucleation sites. (1997. It is usually necessary to separate this interaction from the true dynamics of a moving dislocation. Figure 4. On the nanoscale. but the problem is still not satisfactorily solved. (1999). However. they are more crucial in facilitating or blocking other deformation mechanisms. Using a multiscale approach. courtesy of D. (After Schonfelder et al. Although grain boundaries also have this function. and L a low-angle grain boundary. for materials with grain sizes below a few nano- metres.) . MD simulations of nanoscale grain-boundary evolution under externally applied stress. 2000. Several recent studies have attempted to damp out boundary effects (Ohsawa and Kuramoto 1999.3492 N. and this evolution in turn affects the mechanical deformation. Wang et al. materials strength begins to decline when the grain size decreases (Van Swygenhoven 2002). H indicates a high-angle grain boundary. problems that remain is treating the interaction of boundaries with elastic waves emitted by a moving dislocation. Cai et al. M. Schonfelder et al. Ortiz and Phillips (1999) have studied the interaction of a dislocation with a grain boundary. the grain structure evolves as well. 2001b). which shows that the strength of materials is inversely proportional to the square root of its grain size.g. surfaces and grain boundaries) respond to mechanical loading. As shown in figure 4. interfaces (e. and quadruple junctions are not stable and are reduced to triple junctions. Time is labelled in each frame. Similar to point defects and dislocations. high-angle grain boundaries shrink. grain boundaries facilitate mechanical deformation through grain-boundary sliding. In connection with dislocations. It is well established that grain boundaries serve as barriers to dislocation motion and give rise to the Hall–Petch effect.

In the following. Saxlov`a et al. 1997. There is still not a universal characterization method that fits all purposes. Hamilton et al. Zaiser et al. . Among the three types of defects. The identification is straightforward if the dislocation configuration is simple. kink-pair formation energy and dislocation core energy). 1998. Thomson et al. 1998). let us examine the type of information that can be passed on to larger length scales. In these early efforts. or tracking the atomic level stress or energy (Hamilton 1997. Finally. Here. Information on defect energetics and diffusion tensors obtained by MD or MS simulations provide parameters to rate equations and MC methods describing microstructure evolutions. the topology of interacting dislocations within the system must be included (El-Azab 2000). was initially motivated by the need to understand the origins of heterogeneous plasticity and pattern formation. unambiguous identification of dislocation cores may require a combi- nation of several of the methods discussed above. 2001). 1989. 1999. An early variant of this approach (the cellular automata) was first developed by Lepinoux and Kubin (1987). Peierls stress and dislocation mobility are used as input parameters in dislocation dynamics simulations at the mesoscopic level (Bulatov et al. using disregistry functions (such as the centro-symmetry parameter). 1998.g. b). dislocations and interfaces. and interfaces (particularly grain boundaries). This physical understanding relies on identification and analysis of point defects. 1998. Chang et al. Ha¨hner et al. Before closing this section. 2002). One may rely on construct- ing the Burgers circuit. Mesomechanics and the role of defects Studies of the mechanical behaviour of materials on a length scale larger than what can be handled by direct atomistic simulations and smaller than what is poss- ible with continuum mechanics represent particular difficulties. and that was followed by the proposal of DD (Ghoniem and Amodeo 1988a. dislocations. The main challenge in this regard is that. evolution equations for statistical averages (and possibly for higher moments) are to be solved for a complete description of the deformation problem. dislocation ensembles were modelled as infinitely long and straight in an isotropic infinite elastic medium. When complex dislocation configurations are involved. } 4. Gregor and Kratochvil 1990. Amodeo and Ghoniem 1990a. In these developments. dislocations are probably the most difficult to characterize or identify. The first approach. Wang and LeSar 1995). Two complementary approaches have been advanced to model the mechanical behaviour in this meso- scopic length scale. that of DD. Barts and Carlsson 1995. Kratochvil and Saxlov`a 1992. El-Azab 2000. Lubarda and Needleman 1993. with applications demonstrating simplified features of deformation microstructure. The method was further expanded by a number of researchers (Guluoglu et al. Dislocation energetics (e. b. we review the main theoretical and computational advances in mesomechanics. The second approach to mechanical models on the mesoscale has been based on SM methods (Walgraef and Aifantis 1985. dislocation nucleation conditions and interaction mechanisms with other defects (dislocations or grain boundaries or other interfaces) from MD simulations are critical to DD simulations. Multiscale modelling of nanomechanics and micromechanics 3493 The presentation in this section has so far focused on physical aspect of nano- mechanics. we consider three types of defects: point defects. unlike the situation encountered in the development of the kinetic theory of gases.

If the PDF pðxÞ cannot be easily inverted analytically. Here. however.g. Then we obtain trial values for x with the inversion technique following p0 ðxÞ. and time scales less than tens of nanoseconds. that is CðxÞ ¼ . 4. While early research in defect mechanics focused on the nature of the elastic field arising from defects in materials. two useful results emerge. the field has steadily moved from conceptual theory to prac- tical applications. we replace the original distribution function pðxÞ by an approximate form p~ ðxÞ for which the inversion method can be applied. MD simulations in the nanomechanics area have been limited to sizes less than 100 nm. Because the motion of all atoms in the material is not relevant. These observations have been instrumental in the emergence of mesomechanics on the basis of defect interactions. recent computational modelling has shifted the emphasis to defect ensemble evolution. if the distribution function is normalized to obtain a probability density function (PDF) pðxÞ we can determine the probability that the random variable x0 is less than an arbitrary x by integrating the PDF from the minimum value to x. which has its roots in the pioneering work of Eshelby (1957). bubbles. dislocations) and surface defects (e. Mura (1968. M. When the CDF is equated to a uniformly distributed random number . 4. sampling can be performed by the Von Neumann rejection technique. At present. we accept trial . A simple method for generation of random numbers according to a given distribution function is the inversion method (James 1990). Because of the many computational advances during the past two decades. These are the MC.3494 N.2. The kinetic Monte Carlo method The MC technique is a statistical method for solving deterministic or probabil- istic problems. We review here some of the more popular methods of computational mesomechanics that have been developed during the last two decades. vacancies and interstitials). one can just follow material regions around defects. Secondly. the number of degrees of freedom required to describe their topological evolution is many orders of magnitude smaller than those associated with atoms. b). The integral of the PDF is called the cumulative distribution function (CDF) CðxÞ. The mechanics problem can thus be greatly simplified if all atomic degrees of freedom were adiabatically eliminated (similar to the Born–Oppenheimer approach).g. Accurate descriptions of material volumes in the micro- metre range where continuum descriptions break down are not yet attainable with MD techniques. Another method to achieve the same result is known as importance sampling.1 Defect mechanics on the mesoscale So far. Ghoniem et al. defect interactions can be accurately described by long-range elastic forces transmitted through the atomic lattice. In this approach. 1982) and Kossevich (1999). the discrete DD and the SM methods. cracks and grain boundaries). we have dealt with simulations of the atomistic degrees of freedom. where various levels of approximations have been introduced to give tractable and yet rigorous solutions of the equations of motion. Fortunately. Finally.e. Firstly. Kro¨ner (1958a. voids. ranging in size from atomic dimensions (i. and only those associated with defects are retained. and is a combination of the previous two methods. by sampling from random distributions utilizing concepts of prob- ability theory. Since the density of defects is many orders of magnitude smaller than the atomic density. and only atoms around defects determine the mechanical properties. the mechanical behaviour of materials is primarily determined by topological defects. the resulting solution for x gives the desired distribution function. to line defects (e.

First. However. The method is based on a hierarchy of approximations that enable the solution of relevant problems with today’s computational resources. the time increment itself. In the Metropolis et al. For example. self-intersti- tial atom (SIA) clusters are produced by atomic collisions in the crystal. (1953) scheme. Thus. ð17Þ i¼0 i¼0 i¼0 i¼0 where ri is the rate at which event i occurs (r0 ¼ 0) and is a random number uniformly distributed in the range [2 ð0. DD has now become an important computer simulation tool for the description of plastic deformation on the microscale and mesoscale (i. Heinisch 1995). as follows: X m 1 X M X m X M ri ri < < ri ri . Beeler 1982. the table of all possible events is updated (Battaile and Srolovitz 1997). where p0 ðxÞ is a constant (James 1980). the size range of a fraction of a micrometre to tens of micrometres).e. The main advantage of this approach is that it can take into account simultaneous different microscopic mechanisms and can cover very different time scales. Hence. They can be represented as rigid small prismatic dislocation loops that can migrate randomly in the drift field of dislocations (Osetsky et al. Multiscale modelling of nanomechanics and micromechanics 3495 values with the probability proportional to the weight w. the system is changed. The results of KMC simulations shown in figure 5 (Ghoniem et al. (1975). because it depends on the corresponding event probability. Dislocation dynamics Since it was first introduced in the mid-1980s independently by Lepinoux and Kubin (1987) and by Ghoniem and Amodeo (1988a). When the configurational space is discrete. illustrate the stages of SIA motion and clustering in the stress field of dislocations. corresponding to each step is variable. we can choose and execute a single change to the system from the list of all possible changes at each MC step. known as the n-fold way algorithm and introduced by Bortz et al. we tabulate the rates at which each event will occur anywhere in the system: ri . no events will happen. and the fastest events are only possible in certain rare situations. Once an event is selected. After a particular event is selected. a fixed time increment is chosen such that at most one event occurs during a time step. and the list of events that can occur at the next step is updated. under high-energy ion or neutron irradiation conditions. An alternative technique.3. The rejection technique has been shown to be a special case of importance sampling. this approach is inefficient since. 2000). This is the general idea of the kinetic Monte Carlo (KMC) method (Doran 1970. 1Þ]. 4. The probability of event occurrence is defined as the rate at which the event takes place relative to the sum of all possible event rates. . as X M t ¼  ln ð Þ ri : ð18Þ i¼1 This method is particularly useful in cases where the events occur on very different time scales. given by w ¼ pðxÞ=p~ðxÞ. in many time steps. one event denoted by m is randomly chosen from all of the M events that can possibly occur. and all rates at which they occur can be determined. 2002). dt. ensures that one event occurs somewhere in the system.

Amodeo and Ghoniem 1988. M. Ghoniem 1999. several research groups extended the DD methodology to the more physical. Kubin 1993. 0. DeVincre and Condat 1992. Kubin et al. 1989. 1996. Wang and LeSar 1995). 1990. 2001). In its early versions. 2000. 1992. Ghoniem and Sun 1999. yet considerably more complex three-dimensional simulations. 1998. Results of KMC simulations for SIA cluster agglomeration and interaction near dislocation segments. Ghoniem et al. Schwartz 1997. The sequence of figures are for t ¼ 0. Schwarz and Le Goues 1997. Rhee et al.3496 N. Guluoglu et al.043. Barts and Carlsson 1995. Recently. 1998. Ghoniem and Amodeo 1988a. the collective behaviour of dislocation ensembles was determined by direct numerical simulations of the interactions between infinitely long straight dislocations (Lepinoux and Kubin 1987. DeVincre and Kubin 1994) has resulted in greater confidence in the ability of DD to simulate more complex deformation microstruc- ture. Schwarz and Tersoff 1996. . 1998. Figure 5. Lubarda and Needleman 1993. DeVincre et al. Kubin and Canova 1992.189 and 0. The method can be traced back to the concepts of internal stress fields and configurational forces. 1991. More rigorous formulations of three-dimensional DD have contributed to its rapid development and applications in many systems (Hirth et al. Ghoniem 1992. Ghoniem et al. 1992.416 ns. 1992. 1990a. Zbib et al. Groma and Pawley 1993. The more recent development of three-dimensional lattice DD by Kubin and co-workers (Canova et al. b. 0.

(After Ghoniem et al. where s is an arbitrary unit vector.. excluding the stress contribution of the loop itself. The mechanical power during the virtual motion is composed of two parts: (i) the change in the elastic energy stored in the medium upon loop motion under the influence of its own stress (i. p ðRÞ ¼ Xi R3 . Consider the virtual motion of the dislocation loop. of differential arc length dl with components dlk . The radius vector R connects a source point on the loop to a field point. Parametric representation of dislocation segments. a more efficient form of equation (19) can be expressed as a line integral performed over the closed dislocation loop (deWit 1960): þ þ   bi 1 1 ui ¼  A dl þ  bR þ  b R dlk . The symmetric part of ui. ð20Þ 4p C k k 8p C iki i . mi where  and are the shear modulus and Poisson’s ratio respectively. pp 1  kmn n . x0 Þnn dSðx0 Þ. (2001). which can be expressed as (Volterra 1907. and the vector potential Ak ðRÞ ¼ ijk Xi sj =RðR þ R  sÞ satisfies the differential equation pik Ak. S is the surface capping the loop. x0 Þ are the Green’s functions at x due to a point force applied at x0 . b is the Burgers vector with the Cartesian components bi . Mura 1968) ð o ui ðxÞ ¼  Cjlmn bm 0 Gij ðx. caused by the dislocation loop.. j can be obtained from equation (19) by differentiation.e. successive partial derivatives R. as shown in figure 6.) . Multiscale modelling of nanomechanics and micromechanics 3497 When the Green’s functions are known.: and magnitude R. the elastic field of a dislocation loop can be constructed by a surface integration. The starting point in this calculation is the displacement field in a crystal containing a dislocation loop. nn is a unit normal to S and bm is the Burgers vector. with Cartesian components Ri . In an elastically isotropic and infinite medium. The elastic distortion tensor ui. ijk. the change in the loop self-energy): (ii) the work done on moving the loop as a result of the action of external and internal stresses. The line integrals are carried along the closed contour C defining the dislocation loop. j (elastic strain tensor) is then used in the stress–strain relationship to find the stress tensor at a field point x. Gij ðx. g3 = b b g2 t 0 g1 e Q 1 1 1x 1y Figure 6. ð19Þ S o xl where Cjlmn is the elastic constants tensor.

The approach is reminiscent of the single-electron simplification of the many-electron problem in QM. Secondly. 2001). g2 ¼ t. as shown in figure 6. Firstly. 1y . surface integrals can be replaced by line integrals along the dislocation. and they become functions of both source and field points (i. Therefore. for example. the starting point in most DD simulations so far is a description of the elastic field of dislocation loops of arbitrary shapes by line integrals of the form proposed by deWit (1960) : þ    1   1   ij ¼ R  dl þ imn dlj þ  R  ij R. The line integral is discretized. Now define the three vectors ðg1 ¼ e. they are moved sequentially. A major simplification is that this many-body problem is reduced to the single-loop problem. In addition. 1z g. closed-form solutions for Green’s functions and their spatial derivatives are available only for elastically isotropic materials. and the stress field of dislocation ensembles is obtained by a summation process over line segments. These two components constitute the Peach–Koehler (1950) work . as follows: T ¼ dl=d!. one can obtain the field by numerical integration over the scalar parameter that represents the segment. Recently. One of these segments is described by a parameter ! that varies. a great reduction in the level of required computations ensues. if we introduce the tangent vector T. The main idea of DD is to derive approximate equations of motion from the principle of virtual power dissipation of the second law of thermodynamics (Ghoniem et al. the unit tangent vector t and the unit radius vector e. g3 ¼ b=jbjÞ as a covariant basis set for the curvilinear segment. Green’s functions and their derivatives have analytical solutions. t ¼ T=jTj. M. Let the Cartesian orthonormal basis set be denoted by 1  f1x .3498 N. Ghoniem and Sun (1999) and Ghoniem et al. if dislocation loops are discretized into curved parametric segments. and  denotes the tensor product. by finding virtual Peach–Kohler forces that would result in the simultaneous displacement of all dislocation loops in the crystal. with the motion of each one against the collective field of all other loops. rigorous three-dimensional DD in finite anisotropic systems (e. In finite systems. mpp jmn i 1  kmn . Thus. ð21Þ 4p C 2 . (2001) have shown that. The differential stress field is given by dr VjTj  1      ¼ g  g1 þ g1  g1 þ ð1  Þ g2  g2 þ g2  g2  3g1  g1 þ I : d! 4pð1  ÞR2 ð22Þ .g. e ¼ R=R. If the material is assumed to be elastically isotropic and infinite.e. where ij is the mixed Kronecker delta and V ¼ ðg1  g2 Þ  g3 the volume spanned by the vector basis. ppm dlk . 1. I ¼ 1  1 as the second-order unit tensor. and their contravariant reciprocals as gi  gj ¼ ij . instead of moving all the loops simultaneously. from 0 to 1 at end nodes of the segment. In this simplification. Green’s functions are not invariant under x ! x0 transforma- tions. ijm where  and are the shear modulus and Poisson’s ratio respectively. The segment is fully determined as an affine mapping on the scalar interval 2 ½0. thin films and quantum dots) are very demanding and have not yet been implemented. This does not allow the reduction of the surface integral of equation (20) to a simpler line integral by application of the Stokes theorem. Ghoniem et al. not functions of the radius vector alone any more).

q2 . C ¼ ½C1 ð!Þ. and the line integral is carried along the arc length of the dislocation ds. 1}. . Following Ghoniem et al. Substituting equation (26) into equation (25) we obtain X Ns ð   dQ QT  CT f   CT C  jdsj ¼ 0 ð27Þ j¼1 Gj dt Let ð ð T  fj ¼ C f jdsj. . Ci ð!Þ. Based on these methods for evaluations of the interaction and self-forces. f  ¼ ½f1 . mÞ are shape functions dependent on the parameter (0 4 ! 4 1). . The Peach–Kohler (1950) force exerted on any other dislocation segment can be obtained from the total stress field (external and internal) at the segment as FPK ¼ b  t: ð23Þ The total self-energy of the dislocation loop is determined by double line inte- grals. thus allowing parametrization of the form: r ¼ CQ: ð26Þ Here. . Fkt are the components of the resultant force. kj ¼ CT Cjdsj: Gj Gj . . In each segment j. . The resistivity matrix (inverse mobility) is Bk . the weak variational form of the governing equation of motion of a single dislocation loop was developed by Ghoniem et al. f ¼ . a closed dislocation loop can be divided into Ns segments. (2000): ð  t  Fk  Bk V drk jdsj ¼ 0: ð24Þ  Here. consisting of the Peach–Koehler force FPK (generated by the sum of the external and internal stress fields). C2 ð!Þ. t ¼ : a a B Here. . the self- force Fs and the osmotic force FO (in case climb is also considered (Ghoniem et al. f3 T . V are the velocity vector components. (2000). the stress field everywhere is obtained as a fast numerical quadrature sum (Ghoniem and Sun 1999). r2 . . f2 . . Multiscale modelling of nanomechanics and micromechanics 3499 Once the parametric curve for the dislocation segment is mapped on to the scalar interval {! 2 ½0. qm T . and Q ¼ ½q1 . r3 T . . Hence equation (24) can be rewritten in dimensionless matrix form as ð   dr   d r  f    ds  ¼ 0: T ð25Þ G dt Here. Cm ð!Þ.  the shear modulus and t is time. where qi are a set of generalized coordinates. which are all dependent on the dimen- sionless time t . However. a is lattice constant. Gavazza (1976) have shown that the first variation in the self-energy of the loop can be written as a single line integral. . . 2. 2000). let us define the following dimensionless parameters: r F t r ¼ . ði ¼ 1. we can choose a set of generalized coordinates qm at the two ends. and r ¼ ½r1 . and that the majority of the contribution is governed by the local line curvature. To simplify the problem.

the manner by which the elastic field and self-energies are calculated. (1998) are moved in a rigid-body fashion along the normal to their midpoints. developed by Hirth et al. Wang et al. Ghoniem 1999. coarse resolution is obtained by the lattice method. j¼1 j¼1 then. additional protocols or algorithms are used to treat (i) strong dislocation interactions (e. This has two advantages: firstly. 2000. However. where straight dislocation segments (either pure screw or edge in the earliest versions.g. The differential stress method developed by Schwarz and Tersoff (1996) and Schwarz (1997) is based on calculations of the stress field of a differential straight line element on the dislocation. and some addi- tional details related to how boundary and interface conditions are handled. 2000). the dislocation loop can be geometrically repre- sented as a continuous (to second derivative) composite space curve. Peach–Koehler forces on all other segments are determined . The Brown (1967) procedure is then utilized to remove the singularities associated with the self-force calculation. very drastic variations in dislocation curvature can be easily handled without excessive remeshing. Straight dislocation segments of mixed character in the the force method. we assemble the equa- tion of motion for all contiguous segments in global matrices and vectors as X Ns X Ns F¼ fj . It is based on the Khachaturyan–Shatalov reciprocal space theory of the strain in an arbitrary elastically homogeneous system of misfitting coherent inclusions embedded into the parent phase. Ghoniem et al. (1997). The method of the phase field microelasticity (Khachaturyan 2000. developed by Kubin et al. consideration of individual segments of all dislocation lines is not required. 2001) presented above. The advantage of this method is that the explicit information on the elastic field is not necessary. there is no abrupt variation or singularities associated with the self-force at the joining nodes in between segments. Instead. junctions or tight dipoles). The suitability of each method is determined by the required level of accuracy and resolution in a given application. Thus. from equation (27). 2001c) is of a different nature. Generally. Other approximation methods have been developed by a number of groups. (1996) and Zbib et al. Ghoniem et al. (ii) dislocation generation and annihilation and (iii) adaptive meshing as dictated by large curvature variations (Ghoniem et al. (1992) and Moulin et al. we obtain dQ K ¼ F: ð28Þ dt The solution of the set of ordinary differential equations (28) describes the motion of an ensemble of dislocation loops as an evolutionary dynamic system. or of a mixed character in more recent versions) are allowed to jump on specific lattice sites and orientations. 2000. secondly. since closed-form solutions for the interaction forces are directly used. the temporal and . These approaches differ mainly in the representation of dislocation loop geometry. In the parametric method (Kukta and Freund 1998. M. Using numerical integration. but they are not tied to an underlying spatial lattice or grid.3500 N. K¼ kj . Following a similar procedure to the finite-element method.

without loss of dislocation lines. Multiscale modelling of nanomechanics and micromechanics 3501 Figure 7. forming a regular pattern with a wavelength of approximately 1 mm. The dislocation microstructure is shown in figure 7 at different total strain (Wang et al.) spatial evolution of several density function profiles (fields) are obtained by solving continuum equations in Fourier space. leading to the formation of dislocation junction bundles and what appears to a stabilization of a cellular structure. the segments that lie outside the simulation boundary are periodically mapped inside the simulation volume to preserve translational strain invariance. (2003a). within a distance of 400a (where a is the lattice constant). In the parametric DD method. The total number of degrees of freedom starts at 6000 and is increased to 24 000 by the end of the calculation. and is based on a binary tree search. the number of interacting degrees of freedom is determined by a nearest-neighbour criterion. Results of computer simulations for dislocation microstructure deformation in copper deformed to increasing levels of strain (shown next to each microstructure).2 mm side length. 2003a). It is observed that fine slip lines that nucleate at low strains evolve into more pronounced slip bundles at higher strains. Figure 7 shows the results of such computations for simulation of plastic deformation in single-crystal copper at a constant strain rate of 100 s1 . The initial dislocation density of  ¼ 2  1013 m2 has been divided into 300 complete loops. Conjugate slip is also observed. The num- ber of nodes on each loop starts at 5 and is then increased adaptively proportional to the loop length. The slip bundles are well separated in space. However. Each loop contains a random number of initially straight glide and superjog segments. the shape of loop ensembles is evolved using equations of motion for generalized coordinates representing the position. (After Wang et al. . with a maximum number of 20 nodes per loop. tangent and normal vectors of nodes on each loop. When a generated or expanding loop intersects the simulation volume of 2.

Several models have been recently advanced. the mobile disloca- tion density m is divided into two subfamily densities representing dislocations gliding in the direction of the Burgers vector (þ m ) or in the opposite direction ( þ  m ) (with m ¼ m þ m ). Kratochvil and Saxlov`a 1992. 1998. Dislocation patterns consist of alternating dislocation-rich and dislocation-poor regions usually in the micrometre range (e. avalanches and percolation). The model is formulated as a set of reaction–diffusion equations with spatial derivatives that lead to inhomogeneous dislocation distributions and will thus be considered first here.g. and the mobile dislocation density for dislocations gliding between obstacles is defined as m . To attain these objectives. Saxlov`a et al. Ghoniem et al. Zaiser and Ha¨hner 1997. Statistical models of dislocation cells. Walgraef and Aifantis 1985. walls and channels). consider systems oriented for single slip. Thomson et al. (ii) It will shed light on the intriguing physics of both self-organization phenomena and the behaviour of critical-state systems (e. the local values of strain rates associated with intermittent dislocation avalanches are estimated to be of the order of (1–10)106 times greater than externally imposed strain rates (Neuha¨user 1983.4. Hence. veins. subgrain walls or boundaries. and to enable a direct link with continuum deformation theory. the SM approach has been advanced (Neumann 1971. Zaiser 2001). dislocation cells. Statistical mechanics Two of the most fascinating features of microscale plasticity are the spontaneous formation of dislocation patterns. On the other hand. Gregor and Kratochvil 1990. Ghoniem et al. Understanding such collective phenomena is of paramount importance in two respects. where b is the length of Burgers vector. These dislocation densities are related to the strain rate via the Orowan relation _ ¼ bm vg . Even if direct computer simulations are successful in the description of these collective phenomena. the static dislocation density. Hence. 1998. direct computer simulations of inhomogeneous plastic deformation and dislocation patterns are still unattainable.g. is defined as s . Ha¨hner 1996. are linear objects of considerable topological complexity.3502 N. For simplicity. The Walgraef–Aifantis (1985) statistical model has gained interest. unlike particles. The fundamental difficulty here is that dislocations. LeSar and Richman 2003). and will be reviewed briefly here. El-Azab 2000. it is very desirable to obtain global kinetic or thermodynamic principles for understanding the self-organization associated with plasticity on the microscale. 4. formed by the immobilized dislocations of the forest. because of its ability to predict dislocation pattern formation under cyclic deformation. when concepts of statistical mechanics and the theory of rate processes are used. Because of the high density of dislocations and the strong nature of interactions. shear bands or other organized dislocation structures attempt to develop continuum equations for dislocation den- sities that contain spatial gradients that drive instabilities and self-organization. M. 1997. Ha¨hner et al. and the highly intermittent and spatially localized nature of plastic flow. 2002. 1990. m the total mobile dislocation density and vg the glide velocity in the primary . subgrains. (i) It will allow materials design approaches that are more fundamentally based to mitigate the detrimental effects of plastic deformation. bundles. In this model. Zaiser et al. some level of phenomenological description is unavoidable. fatigue and fracture on material failure. PSBs.

Multiscale modelling of nanomechanics and micromechanics 3503 slip plane. Moreover. the dislocation densities are related to the internal stress by the relation b i ¼ þ b 1=2 s ð29Þ 2p.

In the last equation. the first contribution comes from obstacles such as precipitates or pre-existing walls sepa- rated by an effective spacing . where  is the shear modulus and is a constant.

Another class of models is based on the evolution of dipolar loops. and their mutual interaction processes. The essential features of DD are their mobility. and ¼ 0 vg e . or at least their order of magnitude. it is the flux caused by their edge component. the glide velocity is related to the effective stress via appropriate phenomenological relations expressing the fact that individual dislocation motion is initiated when the effective stress acting on a dislocation exceeds the yield stress. ð31Þ 2 ot    þ m ¼ =  J þ s  vg Gðs Þm  vg m ðs þ m Þ. 1997. It is shown that PSB formation is triggered by the clustering of dislocations or dislocation dipoles. which includes thermal diffusion and climb. Essmann and Mughrabi 1979). 2 where is the characteristic separation length between dislocations for spontaneous annihilation (Essmann and Mughrabi 1979). triggered by their interaction with gliding dislocations (Kratochvil and Saxlov`a 1992. the resulting dynamic system can be written as (Walgraef and Aifantis 1985) ot s ¼ =  Js þ vg m 1=2 2 s  vs ds  vg m s  s þ vg Gðs Þm . is the frequency of a dislocation freeing from the forest and is proportional to vg =d where d is the characteristic dipole destabilization length which is inversely proportional to the effective stress. thermal activation and climb. the forest dislocations mobility is represented by a diffusive current J ¼ Ds =s . . Kubin and Kratochvil 2000). The internal stress i reduces the effective stress e defined as e ¼ a  i . The proposed statistical models are of the reaction–transport type and focus on the feedback between the evolution of glide dislocations and the dipole density. as conceptualized earlier by Mughrabi (1981). and. which become finally immobile and arrange themselves in regularly spaced walls of high dislocation density (Walgraef and Aifantis 1985). ð30Þ kT 0 where 0 is the yield stress and m > 1. the second part is the contribution from the static dislocation population which also opposes dislocation motion. Finally. in the form vg / ðe =0 Þm . By taking into account these mechanisms. This can be put more explicitly as    e m vg ¼ v0 exp  . in the present case. which represents the effective diffusion within the forest. with a representing the applied stress. Stability and numerical analyses of the previous set of equations have provided information on the conditions for formation of PSBs in fatigued specimens. ot þ þ þ  m ¼ =  Jþ þ s  vg Gðs Þm  vg m ðs þ m Þ. Saxlov`a et al. The current of mobile dislocations is taken as J ¼ vg  m and represents the flux caused by gliding dis- locations. Because of mutual interactions. may in principle be evaluated from microscopic analysis (Neumann 1971. The different characteristic lengths introduced here. d is the characteristic length of sponta- neous dipole collapse.

. self- organized patterns of dipolar loops may be attainable.3504 N. . . . . The result is the sweeping of dipole loops by screw dislocations. Considering only glide motion. equation (34) represents the balance equation for the continuous distribution of loops described by the density %ðrÞ.e. ð32Þ dt j6¼i where B is the loop mobility. Ghoniem et al. rN Þ ¼ N j6¼i ðr  ri Þ. In order to obtain a continuous field description. the problem of loop– loop interaction was shown to be represented as a diffusion equation in a drift field:   d d d%ðrÞ d %ðrÞ ¼ D þ f ext . dipole generation and interactions play a secondary role and are introduced in an ad hoc and qualitative way. Summing the above equation for all loops. Then. it can be replaced by a weighted integral. v. and the result is differentiated with respect to r. f ext is the force caused by stress gradients and glide dislocations and f int is the interaction force between dipoles. the tangent vector is described . t. In an attempt to generalize the statistical description of dislocations. Kratochvil and co-workers utilized a Taylor series expansion of %ðrÞ around r to obtain finally ð d%ðrÞ fðrÞ ¼  ðs  rÞfint ðs  rÞds þ f ext : ð36Þ dr From the balance equation (34) and approximation (36). equation (32) is multiplied by the delta function ðr  ri Þ. For the ith loop. one obtains d d %ðrÞ þ ½BfðrÞ%ðrÞ ¼ 0. . as conjectured by Mughrabi (1981). Since the interaction force between loops fint is short range. i ¼ 1. El-Azab (2000) described the dislocation content of a slip system by introducing the distribu- tion ðiÞ ðx. N. thus we have   " ! # d dri d XN ðr  ri Þ ¼ B f int ðri  rj Þ þ f ext ðr  ri Þ : ð33Þ dr dt dr i¼1 P By introduction of the discrete density function %ðr. . . ð37Þ dt dr dr dr where D is an effective diffusion coefficient. To illustrate briefly the statistical approach of Kratochvil and co-workers. which initiates the formation of dislocation walls. the equation of motion can be then expressed as ! dri XN ¼B f int ðri  rj Þ þ f ext . ð34Þ dt dr where ð fðrÞ  f int ðs  rÞ%ðsÞ ds þ f ext : ð35Þ As a first approximation it can be assumed that % in equation (34) and definition (35) represents the local averaged loop density %ðrÞ independent of detailed loop positions ri . M. and taking the sum on the right-hand side of equation (33). tÞ. r1 . either interstitial or vacancy). given by ð D ¼ B%ðrÞ sEf int ðsÞ ds: ð38Þ When the motion of screw dislocations is coupled with equation (37). let us consider N loops of one type (i. In this approach.

Hartley (2003) introduced the concept of a dislocation distribution vector. of the form   X ðiÞ o þ v  = þ v_  =v ðiÞ ¼ S . El-Azab (2000) finally obtained a set of kinetic equa- tions. tÞ ¼ t  bðiÞ ðiÞ ðx. Multiscale modelling of nanomechanics and micromechanics 3505 by a single scalar parameter: t ¼ tðÞ. The linear stability analysis of Groma (1997) shows that a homogeneous stationary solution is unstable and that density perturba- tions grow. i ¼ 1. defined as q ðtÞ  t ðtÞ ð45Þ so that the magnitude of q ðtÞ is j q ðtÞj ¼ q ðtÞ. N: ð41Þ dt O O P ðiÞ S includes all possible tensorial sources SðiÞ resulting from Burgers vector reac- tions and cross-slip. ð42Þ dt x x v  ot where the terms containing _ o=o cancel each other. . Considering a volume O. . tÞ dx dv d is the dislocation line length contained in the phase space volume dx dv d at time t on the ith slip system. In an effort to connect DD simulations to continuum theories. for which dislocations on the ith slip system contribute: ð ð aðiÞ ðx. The distribution function ðiÞ is defined in a generalized phase space. The approach of El-Azab (2000) is to consider the conservation equation of the disloca- tion density tensor a. i ¼ 1. For interacting dislocations. The time rate of variation in the dislocation density tensor in terms of the distribution function has been worked out by El-Azab (2000) as ð  ð ð ð   d o aðiÞ ðx. tÞ dx ¼ t  bðiÞ þ v  = þ v_  =v ðiÞ dx dv d. such that ðiÞ ðx. ð46Þ . tÞ dv d: ð39Þ v  The condition of compatibility of the total distortion field in the crystal is written as a þ =  bP ¼ 0 (Kro¨ner 1981). for the ith slip system we have ð  ð X d ðiÞ a dO ¼ SðiÞ dO . leading to pattern formation. N: ð44Þ ot It is also shown that this generalized approach can be reduced to the more simplified formulation of Groma (1997) in the special case of a system of parallel edge disloca- tions in a single-slip configuration. Thus. The right-hand side of equation (41) can also be represented by a phase space integral of scalar source functions ð X ð ð ð X ðiÞ S dO ¼ t  bðiÞ SðiÞ dx dv d: ð43Þ O x v  From the last two equations. however. Equation (40) corresponds to the principle of invariance of the total Burgers vector. El-Azab (2000) included additional source terms to the conservation equations. v. v. this can be extended to ð ð A¼ a dO þ =  bP dO ¼ 0: ð40Þ O O P is the plastic distortion tensor.

the instantaneous active slip volume in fcc metals is of the order of 106 only and that. and dislocation velocity v fluctuations: D E D ED E 2 ðdeff Þ ¼ int eff . the time between neighbouring avalanches exceeds the lifetime of an individual avalanche by four to six orders of magnitude. and then connect these distortions to continuum theories. Hahner (1996) proposed that. On that basis. Thus. M. Ghoniem et al. if one neglects dislocation multiplication and annihilation. D E D E . which is the total length per unit volume of dislocation lines with tangents parallel to t. ð47Þ where the sum is carried over all Burgers vectors in all slip systems. by X a¼ b  q . because of the highly intermittent nature of plastic flow. at typical strain rates of 104 s1 . To estimate the magni- tude of such fluctuations. ð49Þ where int and v are internal stresses and dislocation velocities on slip system . Recognizing that the effective stress eff has an external ext plus an internal int component. Zaiser (2001) arrives at the following expressions for the autocorrelation functions of the effective stress eff . the average work done by the internal stress must be zero. The curvature tensor of the deformed volume is given by j ¼ 12 Tr ðaÞ I  a: ð48Þ Thus. one can write D E int v ¼ 0. and that the work of the external stress is dissipated into heat. and with Burgers vector b . Zaiser (2001) noted that. The dislocation distribution vector is related to the Nye tensor a. it may be possible to compute the volumetric distortions associated with the movement of dislocations within a simulation volume.3506 N. he concluded that plastic deformation can be characterized as a slowly driven non-equilibrium system exhibiting large fluctuations. and upon separating the stress and velocity into average (represented by h  i) and fluctuating (represented by d   ) components.

ð50Þ ðdv Þ2 int b2  int .

2 ¼ D E ¼ .

Using typical values for copper in the previous equations. (1998). detailed measurements of the misorientation angles across dislocation walls have been made by Hughes et al. . . Zaiser (2001) formulated a velocity- averaged stochastic version of equation (44) for the distribution function of surplus dislocation segments or geometrically necessary dislocations to show that the prob- ability distribution function for misorientation angles is in good agreement with the experimental measurements of Hughes et al. the relative velocity fluctuation is found to be of the order of 107 Zaiser (2001) in accordance with the experimental measurements of Neuha¨user (1983). Recently. v eff B _ where B is the mobility. and _ is the average shear strain rate. (1998).

Thomson et al. Their analysis shows that percolating slip clusters conform to the universality class of percolation theory (Stauffer and Aharony 1992). an energy expression in terms of the dislocation density tensor (Rickman et al. 1989. it can be effectively used to describe slip transmission in stage III. The basic idea in this approach is to divide space into small averaging volumes to calculate the local multipole moments of the dislocation microstructure and then to write the energy (LeSar and Rickman 2002) or the force (Wang et al. LeSar and Rickman 2002. Numerical implementation of the multipole expansion for the Peach–Kohler force on a test dislocation separated from a volume containing a dense dislocation aggregate revealed that such a technique is highly accurate and . The issue of spatial averaging has been recently dealt with by LeSar and Rickman (2002) and Wang et al. Shim et al. 2003. the solute atoms function essentially as static traps for dislocations. a random distribution of sources within the cell walls and/or. it is of interest to determine the necessary coarse-graining strategies for both temporal and spatial variables. where s is the number of dislocations in the strained cell. D0 is the diffusivity when there are no traps. (2003b). Multiscale modelling of nanomechanics and micromechanics 3507 A percolation-type stochastic dislocation model has been advanced by Thomson and co-workers (Thomson and Levine 1998. 2002). which is limited to stage III deformation. 2003).e. In this case. LeSar and Rickman (2003) found that the dislocation diffusivity is modified by Dðc. 2003) have begun to tackle this issue by investigations of a number of specific problems. and the degrees of freedom associated with individual solute atoms can be eliminated (Rickman et al. LeSar and co-workers modelled slip as a discrete one-dimensional ‘birth-and-death’ stochastic process (Van Kampen 1992). Slip transmission to neighbouring cells is assumed to be of the form s ¼ as. 2001. 2003b) as an expansion over the multipoles. Thus. These examples show that it may be possible to average out fast time variables in DD simulations. and the problem of the long-range interactions of blocks of dislocation ensembles. namely the combined motion of dislocations and solutes. secondly a distribution of locks holding up dislocations within the cell walls. which initiates a cluster of newly strained cells. solute) concentration. dislocation–solute atmosphere pairs can be treated as quasi-particles. The large disparity in time and length scales during plastic deformation limit the reach of direct DD simulations. In this model. and that a well-defined percolation threshold can be identified. in the other extreme limit of low solute mobility compared with dislocations. Rickman et al. an effective mobility can be derived and can be related to the details of such interaction. Furthermore. LeSar and co-workers (Lesar et al. The key parameter a in such a model is assumed to depend on two types of mechanisms: firstly. In the limit where the solute mobility is high relative to that of dislocations. 2003) is derived. a is a pile-up amplification factor and s is the number of dislocations in the previously unstrained cell. ð51Þ D0 c ðt =  1Þ þ 1 where t = ¼ exp ð=kB T Þ is the mean residence time in a trapped state. =kB TÞ: D t =  1 ¼ c . The power of SM can be brought to bear on one of the most important chal- lenges for the multiscale modelling approach. the weakest cell in a band undergoes a burst of strain. However. Starting from the work of Kossevich (1979) on the interaction energy of systems of dislocations. c  Nt =N is the trap (i. Hence.

1. Relative error versus the expansion order for different R/h (volume size. where Einstein summation applies for repeated indices. different orders of the expansion and different R=h ratios (R being the distance from the centre of the volume to the test dislocation and h being the volume size). a  b ¼ ai bi . can result in significant coarse graining in current DD simulations. an overview will first be given of the main CM-based framework used today to describe the nonlinear deformation behaviour of materials at the local (e.5 R/h=3 R/h=4 4 R/h=5 R/h=6 Error(%) R/h=7 3 2 1 0 0 1 2 3 4 5 6 7 Expansion Order Figure 8. The relative error in the multipole expansion. AB ¼ Aij Bjk .g. 5. (ii) constitutive equations. M. and homogenization techniques to link deformation phenomena simultaneously occurring at different scales in the material microstructure with its macroscopic behaviour. It is clear that the multipole approximate solution converges very rapidly. expressed as jmultipole  accurate j=accurate has been tested by Wang et al. that the third-order expansion (quad- ropole) gives a relative error less than 1% and that the fourth-order expansion gives a relative error of less than 0:05%. A : B ¼ Aij Bij . (2003b). L : A ¼ Lijkl Akl and ða  bÞij ¼ ai bj . 6 CubeSize=1µm 5 R/h=2. second-order tensors by bold capital letters and fourth-order tensors by bold script capital letters. the deformation of a body subjected to external forces and prescribed displacements is governed by the (i) equilibrium equations. Standard tensorial notation will be used throughout. Ab ¼ Aij bj . } 5. Also.3508 N. A representative result is shown in figure 8. .g. polycrystal level) scales. 5 mm and 10 mm). (iii) boundary conditions and (iv) initial conditions. Continuum discretization of a boundary value problem In a generic boundary value problem (BVP). Vectors will be described by bold lower case letters. Emphasis will be placed on recent progress made in crystal plasticity. 1 mm). Continuum mechanics methods In this section. strain gradient plasticity. single-phase or grain level) and macroscopic (e. These numerical tests are performed for different sizes (1 mm. Ghoniem et al.

ð52Þ V G for any arbitrary virtual velocity vector field dv compatible with all kinematics constraints. Esche et al. ð54Þ where ð ð fI ¼ BT r dVe . This ‘weak form’ constitutes the basis for obtaining a numerical solu- tion of the deformation problem via. (1997) and Busso et al. satisfies the following equation: ð ð r : d_e dV  t  v dG ¼ 0. Zienkiewicz and Taylor (1994)). t ¼ rn represents the boundary traction forces. In a Newton–Raphson iterative scheme. Multiscale modelling of nanomechanics and micromechanics 3509 The ‘weak’ form of the boundary value problem is obtained when the equili- brium equations and the boundary conditions are combined into the ‘principle of virtual work’. ð56Þ o^v where k represents a generic iteration and or=o^v is the global tangent stiffness or Jacobian matrix of the nonlinear system of equations. Thus. respectively. and B relates the symmetric strain rate tensor with ^v. The global equilibrium relations (equation (54)) represent a set of implicit nonlinear equations which may be solved incremen- tally using a Newton-type algorithm. the nonlinear system (equation (54)) is typically expanded using Taylor series in the neighbourhood of ^v: k k k k orf^v g k k2 rf^v þ ^v g ¼ rf^v g þ k d^v þ Ofd^v g. r the Cauchy stress. The basic features of a generic Galerkin-type discretization framework are given next. ð53Þ i¼1 where ^v denotes the nodal values of the element velocity field and N are the isopara- metric shape functions. Substituting equation (53) into equation (52) leads to the discretized version of the principle of virtual work on the finite element Ve :  r ^v  f I  f E ¼ 0. the principle of virtual work for the structure. and d_e the virtual strain rate associated with the velocity field dv. In the above equation. In the absence of body forces and inertial effects. Consider a structure occupying a domain V in the deformed configuration which is subjected to external forces and displacements on its boundary. the principle of virtual work is the vehicle by which the global equilibrium equations are obtained (for example. n the normal to the surface on which the tractions act. (2000)). Crisfield (1997). in a CM Lagrangian formulation of a quasistatic BVP. in its rate form. Let v be approximated at a material point within an element by X N max i v¼ N i ^v  N ^v. fE ¼ NT t dGe . for example. the finite-element method. G. . The formulation of accurate estimates of the global Jacobian is at the heart of most numerical schemes developed to provide robust algorithms for the use of complex constitutive models with continuum approaches (for example. To solve a complex BVP numerically. ð55Þ Ve Ge are the internal and external global force vectors. the discretization of the principle of virtual work is generally performed using the finite-element method.

Ghoniem et al.3. hardening and strain gradient phenomena (for example. Meissonnier et al. (2000)). 5. Internal state variables are generally introduced in both formulations to represent the evolution of the microstructural state during the deformation process. that is the stress evolution is computed on axes which rotate with the crystallographic lattice. Note that the Cauchy stress is related to T . (2002)) and general anisotropic plastic and viscoplastic behaviour (for example Jordan and Walker (1992). (1983). M. the macroscopic stress state is resolved on to each slip system following Schmid’s law.e. Meric and Cailletaud (1991) and Meric et al. which has been widely reported in the computational mechanics literature (for example Pierce et al. Kalidindi et al. Hill-type approaches (for example. Nouailhas and Freed (1992) and Schubert et al. Anand and Kothari (1996) and Busso and McClintock (1996)). The description of the kinematics of most crystal plasticity theories follows that originally proposed by Asaro and Rice (1977). In constitutive formulations based on crystallographic slip. (2000)) are based on a generalization of the von Mises yield criterion proposed by Hill (1950) to account for the non- smooth yield or flow potential surface required to describe the anisotropic flow stress behaviour of single crystals. Although recent developments in these two approaches have now reached an advanced stage. the most widely used approach is to assume that the material response is hyperelastic. Thus.2. Gurtin (2000). 5. they treat the material as a continuum in order to describe properly the plastic or viscoplastic effects. the major improvements have been made by crystallographic models due to their ability to incorporate complex micromechanisms of slip within the flow and evolutionary equations of the single crystal models. (1992) and Busso et al. Cuitin˜o and Oritz (1992). that is that its behaviour can be derived from a potential (i. Continuum approaches for single-crystal plasticity Constitutive models developed to predict the anisotropic behaviour of single- crystal materials generally follow either a Hill-type or a crystallographic approach. Asaro and Needleman (1985). These include the effects of dislocation interactions (for example.3510 N. As a common feature. (2001) and Stainer et al. ð58Þ and the corresponding objective work conjugate (symmetric) stress (Hill 1978) or second Piola–Kirchhoff stress T. free energy). Such a potential may be expressed in terms of the elastic Green–Lagrange tensorial strain measure T Ee ¼ 12 ðFe Fe  1Þ. It relies on the multiplicative decomposition of the total deformation gradient F into an inelastic component Fp and an elastic component Fe (Lee 1969). (1991)). F ¼ Fe Fp : ð57Þ Although single-crystal laws can be formulated in a corotational frame. Generic local crystallographic framework A generic internal-variable-based crystallographic framework is said to be local when the evolution of its internal variables can be fully determined by the local microstructural state at the material point. under isothermal conditions. A brief outline of the salient features of local and non-local crystal plasticity approaches are given below.

in contrast. m and n are unit vectors defining the slip direction and the slip plane normal on the slip system. is the slip interaction matrix which accounts for latent hardening effects. . Differentiation of equation (59). if _  > 0: ð62Þ ¼1 In the above equation. In rate-dependent formulations. The hyperelastic response of the single crystal is governed by oFfEe g T¼ . S1 . henceforth the focus of the discussions will be on rate-independent approaches. Here the distinction between a multiplicative slip resistance S2 and an additive slip resistance S1 is motivated by the additive and multiplicative use of non-directional hardening variables rather than by mechanistic .  cS S1 . to ensure uniqueness in the mode of slip (for example Anand and Kothari (1996) and Busso and Cailletaud (2003)). . as ! p Xn F_ ¼ _  P Fp . Owing to the severe restrictions placed on material properties. rate-dependent models have been successfully used in quasirate-independent regimes. . X n _ ¼ _c ¼ h _  . and  is the absolute temperature. Sm  S . and the associated difficulties in its numerical implementation. This has been compounded by the fact that. Thus. . such as latent hardening. the use of rate- independent formulations has been somehow restricted and is much more limited than rate-dependent models. h . . by cali- brating their strain-rate sensitivity response accordingly. the time rate of change in the inelastic defor- mation gradient is related to the slip rates _  on each slip system (Asaro and Rice 1977). . whereby the rate of slip is related to the time rate of change in the resolved shear stress   ð¼ T : P Þ. S2 . with P  m  n : ð61Þ ¼1   Here. . and the assumption of small elastic stretches yield T L : Ee : ð60Þ where L is the anisotropic linear elastic moduli. ð59Þ oEe where FfEe g represents the Helmholtz potential energy of the lattice per unit refer- ence volume. ð64Þ where cS is a scaling parameter. flow rules are based on the well- known Schmid law and a critical resolved shear stress c . A useful and generic expression for the overall flow stress in the slip system can be conveniently found by inverting equation (63) with mS ¼ 2:    ¼ f^ v _  . Then. ð63Þ where Si for i ¼ 1. and S1 and S2 represent additive and multiplicative slip resistances respectively. mS denotes a set of internal state variables for the slip system . . . Multiscale modelling of nanomechanics and micromechanics 3511 1 T by r¼ detðFe Þ Fe T Fe . the slip-hardening rates. In rate-independent formulations. The slip rate in equation (61) can functionally be expressed as   _  ¼ b _   .

the precipitate phase size in two-phase single-crystal materials (for example Busso et al. _ n . and of the order of 0. generation of geometrically necessary dislocations is required in these regions of incompatibility (for example see Arsenlis and Parks (2001). . is what causes the additional strengthening of the material. S  . which allows a more physically meaningful interpretation of strengthening phenomena.1–1 mm in single-crystal materials (Busso et al. This introduces a coupling between the viscous term and microstructure which is inconsistent with most strengthening mechanisms. To accommodate these strain gradients. among others. which give rise to localized strain gradients. Most continuum approaches and formulations dealing with these problems are based on strain gra- dient concepts and are known as non-local theories since the material behaviour at a given material point depends not only on the local state but also on the deformation of neighbouring regions.3512 N. . . (2000)). _  . see Busso and Cailletaud (2003). In general. the boundary conditions or the type of loading. the increase in measured microhardness with decreasing indenter size (for example Swadener et al. the local material flow stress is controlled by the actual gradients of strain when the dominant geo- metric or microstructural length scales force the deformation to develop within regions of less than approximately 5–10 mm wide in polycrystalline materials. work by Meric and Cailletaud (1991) and Busso et al. M. . 5. The crystallographic formulation is completed with the evolutionary relations for the Si internal slip system variables. By expressing the flow stress in the slip system in the way shown in equation (64).g. Recently. _ 1 . . Non-local approaches The study of experimentally observed size effects in a wide range of mechanics and materials problems has received much attention recently. Examples of such phenomena include the particle size effects on composite behaviour (for example Nan and Clarke (1996)). expressed as   S_ ¼ b  i S_ S . . Thus. . . The introduction of these geometrically necessary disloca- tions. . . (2002)). (2000) have proposed flow stress relations with S1 6¼ 0 and S2 ¼ 0. considerations. For a more detailed discussion in these issues. The majority of for- mulations relied on power-law functions for equation (63). in its most general form.  : i 1 mS ð65Þ Note that the dependence of equation (65) on the slip rates on all systems enables cross-hardening effects to be accounted for. Busso et al. average grain size in polycrystal materials). the contributions from viscous effects (first term in equation (64)). The dependence of mechanical properties on length scales can in most cases be linked to features of the microstructure. the conventional crystallographic for- mulations discussed in the previous section will be unable to predict properly the evolution of the local material flow stress. in addition to those stored in a random way (so-called ‘statistically stored’ dislocations). In such cases. gradient-dependent behaviour is expected to become important once the length scale associated with the local deformation gradients becomes sufficiently large when compared with the controlling microstructural feature (e. (1983)). The time rate of change in each internal slip  system variable S_ i is. and hardening mechanisms (second term) can be clearly identified. Ghoniem et al. .4. (2000)). (2000) and Gao and Huang (2003)). which results in S2 6¼ 0 and S1 ¼ 0 in equation (64) (for example Pierce et al. and the effects of decreasing film thickness (for example Huber and Tsakmakis (1999)). . .

_  . . even though Cosserat-type models have shown to be well suited to predicting localization phenomena. Consider a crystal- lographic approach and let q be a higher-order stress vector with components qi . _ n . is relatively easy to implement numerically into the finite-element method but is limited in describing strain gradient problems that involve only one material length scale. . Thus. . Busso et al. including rate-independent plastic deformation and viscoplasticity in both single- crystal and polycrystalline materials (for example Acharya and Beaudoin (2000). the functional dependence of the evolutionary laws of the slip system internal vari- ables. by the nature of the formulation. and let p be the work . However. Another class of non-local theories with higher-order fields is that proposed recently by Fleck and Hutchinson (2001) and by Gurtin (2003). . It requires additional boundary conditions. A significant amount of work has been based on the treatment of the solid as a Cosserat continuum (for example Mu¨hlhaus (1989) and Forest (1998)). . . Arsenlis and Parks (2001) and Bassani (2001). they are analytically complex and phenomenological in nature. where the material flow stress is assumed to be controlled not just by the rate of slip but also by the material curvature.  . . . One additional attractive aspect of the these theories is that they are relatively easy to implement numerically and do not require higher-order stresses or additional boundary conditions. In such non-local formulations. thus making it difficult to gain a direct insight into the controlling physical phenomena at the microstructural level. . They contain higher-order stresses and are extensions of the original theory proposed earlier by Fleck and Hutchinson (1997). . . =_ n . . They rely on internal state variables to describe the evolution of the obstacle or dislocation network within the material and generally introduce the strain gradi- ent effects directly in the evolutionary laws of the slip system internal variables without the need for higher-order stresses. let the work conjugate to the slip rate gradients be =_  . are required. will now include an additional dependence in the gradient = _  of the slip rates. namely the work conjugates of the slip and slip rate gradients. Multiscale modelling of nanomechanics and micromechanics 3513 One of the first non-local theories was that proposed by Aifantis (1987) and Zbib and Aifantis (1988) to describe the formation of shear bands. the balance laws are based on a principle of virtual work where additional field variables.   S_ i ¼ b  S_ i S1 . However. (2001). . such as the general form given for the slip resistance in equation (65). =_  . . This type of formulations rely on first and second derivatives of strain linked to the flow rule to describe strain gradient effects without the need to use higher-order stresses. (2000) and Arsenlis and Parks (2001)). . Sm  S . it provides a limited mechanistic insight into the non-local phenomena. . _ 1 . . (2000). in some of these formulations. . . A more physically intuitive continuum approach to describing strain gradient effects is that followed by constitutive theories such as those developed by Acharya and Beaudoin (2000). Furthermore. . =_ 1 . Then. ð66Þ This class of theories has been shown capable of providing considerable physical insight into the effects of microstructure on the observed macroscopic phenomena. Busso et al. one limitation of these types of theory is that they are unable to describe problems which may require non-standard boundary conditions such as the boundary layer problem modelled by Shu et al.

it can be shown that. equation (67) also implies that the following relations be satisfied at the boundary G: t ¼ rn or v. 5. when no slip rate gradients are present.g. ð68Þ the following microforce balance is obtained:  ¼   þ =q : 1.3514 N. ð ! X X      r: d_e þ ð   Þ d_ þ q  Jd_ dV V   ð ! X  r n  dv þ q  n d_  dG ¼ 0: ð67Þ G  By integrating equation (67) by parts. When distinct heterogeneities exist at different microstructural levels. They must satisfy D L: ð71Þ . once the relevant scales in the heterogeneous microstruc- ture are identified. The numerical implementation of these typically highly nonlinear theories require the development of complex numerical algorithms. equation (68) gives p ¼   and equation (67) resolves to the local version of the principle of virtual work given by equation (52). Such schemes are based on the assumption that the mechanical behaviour of individual constituents can lead to the description of the mechanical response of a macroscopic aggregate through either suitable interaction laws or a numerical averaging process. the introduc- tion of higher-order stresses and the required boundary conditions to satisfy the additional higher-order field equations introduced by some non-local theories often precludes and limits the use of these formulations to in-house programmes (e. 1999)). Homogenization approaches for heterogeneous microstructures The bridging between the mechanical behaviour of heterogeneous materials and that of their individual constituents remains a topic of major interest and is at the heart of homogenization schemes developed to predict the behaviour of materials at different scales. or _  : ð70Þ It can be seen that. it is then necessary to select a RVE of the microstructure at the level of interest. and q  n. in addition to the usual equilibrium relation in Vgiven by div r ¼ 0. it is always possible to identify at each level the smallest possible representative volume element (RVE) of the microstructure which contains all the information concerning the distribution and morphology of the material’s heterogeneities. ð69Þ where 1 is the second-order unit tensor and   is the resolved shear stress previously defined. Then. Ghoniem et al. M. Furthermore. However. conjugate to the slip rate _  . see implementation of a coupled stress theory by Shu et al. Irrespective of the homogenization schemes. Typically.5. the representative length scales in the microstructure are the average size D of the largest heterogeneity at that particular level and the size L of the RVE.

The former assumes strain uniformity and can only fulfil compatibility at grain boundaries. in cases of low strain-rate sensitivity and large property mismatch. but not equilibrium. A critical aspect of the self-consistent method is that the strain distribution within the inclusion is assumed uniform when in reality. In contrast. r_ and e_ . Delannay and Samajdar (1999)) and on mean-field approaches. . One of the simplest self-consistent frameworks is that proposed by Hill (1965). then the homogenized equivalent medium is said to be generalized. Multiscale modelling of nanomechanics and micromechanics 3515 The value of D can be. such as self-consistent (for example Hill (1965)) and variational (for example Ponte Castan˜eda (1991)) methods. Here. Some of the most widely used approaches to link local fields with macroscale phenomena. the interaction between a single-crystal grain and its neighbours is treated as that between an inclusion with the same properties as those of the grain. the average grain size in a polycrystal with randomly oriented grains or can be defined by the size of the precipitates relative to their mean spacing. compatibility and equilibrium between grains are satisfied at both the local and the macroscopic levels. through an elastic accommodation tensor Le   T_  r_ ¼ Le : e_  E_ : ð72Þ The determination of suitable interaction relations between the inclusion and HEM is at the centre of most self-consistent approaches. thus a high constraint is imposed on the inclusion Matrix Incl. The cases to be discussed here are those where loading in the RVE is homogeneous. (1992)) or Sachs-type (for example Leffers (2001)) models. Homogenization schemes require not only constitutive models for the individual constituents but also appropriate rules to make the transition between scales. and special kinematics and equilibrium equations would apply (for example Besson et al. In equation (72). embedded in an homogeneous equivalent medium (HEM) which has the same (unknown) properties as those of the macroscopic aggregate (figure 9 (a)). where the stress-rate tensor T_ and strain-rate tensor E_ in each phase or grain are related to those of the HEM. HEM HEM (a) (b) Figure 9. Among the most successful recent methods proposed to overcome the limitations of the assumed plastic strain or stress uniformity are those based on the relaxed constrained method (for example van Houtte. Sachs-type models assume homogeneous stresses and ignore local compatibility at grain boundaries. an elastic interaction between the grain and the polycrystal aggregate is implicitly assumed. If the loads applied on the RVE were inhomogeneous. In the self-consistent averaging approach. for instance. such as the large defor- mation and texture evolution of polycrystals. (2002)). it is seldom the case. Incl. Comparison between the assumptions behind (a) standard and (b) generalized self-consistent approaches. are Taylor-type (for example Kalidindi et al.

referred to as a general- ized or a three-phase scheme. among other applications. whereby the effective stress potentials of nonlinear compo- sites are expressed in terms of the corresponding potentials for linear composites with similar microstructural distributions. a composite sphere made up of two phases (i. whereby a microstructural ‘window’ of the constituents is periodically arranged and subjected to homogeneous far-field loading. thus enabling a precise model of the microstructure to be made and a frame- work to predict accurately the inhomogeneity of the deformation at the level of the individual phases. For further details. to generate the corresponding bounds. An alternative homogenization method to the classical self-consistent approach is that derived from a variational procedure. their numerical implementation require Newton-type iterative procedures to solve the highly nonlinear systems of equations. This method has the advantage . self-consistent schemes are well suited to plastic or viscoplastic aggregates which can be treated as elastically rigid. Here. see the references given herein. for the nonlinear case. As a result of the ever-increasing computer power available. is illustrated in figure 9 (b). the average stress and strain responses of the unit cell can be obtained numerically. As self-consistent schemes are generally implicit. Herve and Zaoui (1993) showed that such scheme provides a framework for statistical analyses and demonstrated this by determining the effective behaviour of a random assembly of arbitrary-size spheres. by the surrounding elastic aggregate. The additional requirement in this case is that the average strain in the two-phase composite sphere must be the same as the strain prescribed in the far field. Similarly. M. such a high constraint is partially relaxed by the plastic deformation of the polycrystalline aggregate. Alternatively. By defining appropriate periodic boundary conditions on the smallest window or RVE which contains all the information about the local heterogeneities. the finite-element mesh can be designed so that the element boundaries correspond to phase boundaries. equation (72) becomes  p  T_  r_ ¼ Lp : e_ p  E_ .3516 N. One of the most powerful is that based on periodic unit-cell concepts. Even though these methods are computationally intensive. the tangent approach method proposed by Molinari (2002) enables approximate solutions for nonlinear material behaviour problems to be obtained while preserving the structure of the Eshelby linear inclusion solution. Pitakthapanaphong and Busso (2002) used a similar generalized self-consistent approach to determine the elastoplastic properties of functionally gradient materials. they offer the capability for digitized images of typical microstructures to be easily superimposed on to a regular finite-element mesh. A more elaborate self- consistent approach (for example Herve and Zaoui (1993)). Ghoniem et al. The variational formulation proposed by Ponte Castan˜eda (1991) relies on the effective modulus tensor of ‘linear elastic comparison composites’. Thus. Generally. ð73Þ where the global and local strain-rate tensors are now the plastic tensors. In reality. matrix and inclusion) is embedded in an infinite matrix which is the HEM of unknown properties. The incorporation of elastic effects within a self-consistent frame- work is more difficult and clear solutions remain elusive. Ponte Castan˜eda’s variational principles have been used successfully to derive elastoplastic relations for metal matrix composites. The work of Berveiller amd Zaoui (1979) addressed this problem by introducing a plastic accom- modation factor Lp .e. it is now becoming possible to replace the approximate mean-field methods for more accurate techni- ques based on numerical homogenization.

} 6. It can be seen that the localization of inelastic strain occurs in the vicinity of the stronger (eutectic) region. For complex three-dimensional microstruc- tures. after a 10% straining along the [010] orientation at a rate of 103 s1 at 950 C. We conclude this article by briefly summarizing the current level of understanding in these areas and discuss our expectations of forthcoming progress. and figure 10 (b) the predicted RVE contour plot of uniaxially equivalent accumu- lated inelastic strain. but the modelling effort required is generally impractical for very complex microstructures. An example of a periodic unit cell of a typical heterogeneous single-crystal superalloy RVE is presented in figure 10 (Regino et al. Homogenization of a typical heterogeneous single-crystal superalloy: (a) RVE.1. links between length scales should bring the overall field of material modelling one step closer to predict. . the final properties of a proposed material. 2002). Figure 10 (a) shows a scanning electron micrograph of the microstructure’s RVE at an intermediate scale. there has not yet been a concerted effort to develop explicit links between atomistic and CM models. realistic finite-element meshes can be built based on Voronoi tesselations (for example Ghosh et al. ab initio. we have discussed the different modelling approaches which address specific phenomena at different length scales and have highlighted the rich variety of physical. Multiscale modelling of nanomechanics and micromechanics 3517 Figure 10. that almost any heterogeneity can be modelled. and (b) predicted contours of accumulated inelastic strain under [010] uniaxial loading (Regino et al. such as those of polycrystal aggregates with randomly oriented grains. Outstanding issues and future prospects In this review. Bridging the length and time scales Even though recent advances in computing power have led to new and vastly improved simulation techniques at the atomistic and continuum levels. 6. (2002) and Barbe et al. When properly reinforced. 2002). (2002)). computational and technological issues within the broad area of nanomechanics and micromechanics which have been successfully addressed.

slip systems and interface structures of the phases and the kinetics of phase transformations during subsequent thermal or thermomechanical processes using atomistic. linking time scales remains an outstanding problem. structures and the derivation of inter- atomic potentials from ab initio calculations. and inter- atomic potentials) needs to be related to the range of compositions predicted after solidification.3518 N. The resulting defect structures and interface kinetics are then fed into the continuum formulations to provide the crystallographic description and slip systems of the individual phases present in the heterogeneous (multiphase) microstructures. .g. Here. crystal energies and structures. systematic and rigorous Lattice energies. As pointed out earlier. The overall behaviour of RVEs of such microstructures can finally be determined using homogenization techniques. Ghoniem et al. kinetics of phase transformation Continuum Range of Compositions / phases Individual single crystal Microstructural heterogeneities resulting phase models and kinetics from solidification of phase transformation at the microscale Microstructure morphologies Average behaviour of Constitutive model for the and compositions the RVE’sm icrostruture RVEs' individual phases Macroscopic Constitutive Behaviour Figure 11. slip Interfacial energies compositions systems. (ii) the prediction of complex microstructural heterogeneities and morphologies resulting from solidification from the surface energies and kinetics predicted ab initio. crystal energies/structures Atomic scale and crystal structures and interatomicp otentials topological modelling of from ad-initio interface/ defect structures Atomistic calculations Kinetics of liquid-solid phase transformation Range of Crystals tructures. A natural sequence to produce a fully integrated multiscale modelling framework will require the following: (i) the prediction of crystal energies. The ab initio results can then be used in MD calculations of deforma- tion mechanisms. DD and topological modelling techniques. the physical properties of the alloy are predicted at the atomistic scale and these are linked to the continuum level via appropriate constitutive and kinetic models. M. Flow chart showing the links between the different modelling approaches. together with the outputs and inputs for each subarea. (iv) the formulation of crystallographic models for each phase to describe the dominant deformation processes and the development of homogenization schemes to obtain the macroscopic mechanical response of the material. While progress on linking length scales has just started. The properties determined by ab initio calculations (e. Interatomic potentials. Figure 11 illustrates the structure and the links between the modelling para- digms. interface structure. (iii) the identification of the crystal structures.

While the standard local spin density and generalized gradient approximations for the exchange–correlation energy can work for certain cases of strong correlation. methods for self-consistent linking of time scales are still lacking. to enable dynamic processes to be simulated over sufficiently long time scales. Multiscale modelling of nanomechanics and micromechanics 3519 reduction in the degrees of freedom that describe material evolution will lead to self-consistent length scale linking. the LDAþU method. It is also possible that the use of quantum MC methods for benchmark calculations could provide a fruitful path to assess and improve density functionals. (iii) computationally faster methods. Thus. matching can be accomplished for static or quasistatic problems. might yield a more reliable description of strong correlations. while microstructure evolution takes place on much longer time scales: seconds to years. while it has not been shown for fully dynamic problems. The introduction of generalized gradient corrections has improved the accuracy of binding energies. when a MD model is directly linked to the continuum. and (iv) more general approaches. Novel functionals. Recasting the electronic structure calculations in an O(N) form which scales linearly with N will have important conceptual and practical implications for the treatment of sufficiently large systems. several other methods have been recently developed to treat the ground-state properties of condensed systems that exhibit strong correlations. As alluded to in } 2. In addition. Two other areas of developments. (ii) larger systems. and the optimized effective potential method. and O(N) methods. relaxation time scales in atomic models cannot be matched by continuum relaxation time scales. Atomistic scales Atomistic simulations have mainly supplemented experimentally obtained infor- mation until now. On the other hand. are faster methods. Furthermore the independence between the . 6. Nevertheless. If the full calculation on a large system could be carried out in a time of O(N). that is where electrons partially preserve their localized atomic-like nature. such as the recently developed self-correlation-free meta-GGA and self-interaction-free hyper-GGA. and hence confident predictions of mechanical properties. Such quantum O(N) methods will have exactly the same scaling as classical empirical potential methods. atomic jumps and nucleation) may have profound effects on microstructure evolution. perhaps the most important ones. In addition. they fail dramatically for others. Since the time evolution of the microstructure is path dependent.g. the LDAþdynamic mean-field theory method. the time-dependent extension of DFT and the GWA will provide a way to treat the excitation properties of moderately correlated electron systems. then the properties of a small region of the large system could be calculated in a time independent of the system size. events that occur in the picosecond to nanosecond time scale (e. Progress in this area is needed. Density functional approaches have evolved rather rapidly during the past decade to address accuracy. surface energies and energy barriers. future developments are expected in four major directions: (i) higher accuracy. these include the self-interaction correction method. Concurrent with the above efforts in developing more accurate functionals within the spirit of DFT. Events at the atomic scale are often in the picosecond to nanosecond range.2.

Hybrid methods of this type will certainly be based on the same notions of locality as O(N) methods and will employ similar techniques. Approaches linking directly atomistic and continuum methods do exist. thus making the study of the mechanics of a single nanotube relatively easy. This would enable calculations on large systems of great interest in areas which are beyond current capabilities. However. It would be desirable to take such a configuration. Such problems can only be approached if one succeeds in linking methods of different accuracies. it will not be possible in the foreseeable future to treat systems containing millions of atoms at a highly accurate DFT level using large basis sets. pure versus segregated grain boundaries) lead to a large number of interaction mechanisms. and applying the high-accuracy method only to regions where the low- accuracy method is expected to fail. when com- pared with bulk dislocations. The second area is dislocation interaction with interfaces. In atomistic MD simulations. In addition to the areas discussed above. Even with O(N) algorithms. Ghoniem et al. A single method is unlikely to meet all goals in multiscale modelling of nano- mechanics and micromechanics. when the interface of a nanotube and a matrix (e. to describe it in MD simulations and then to feed the resulting configuration back to DD simulations in a seamless fashion. a few dislocations may interact at close proximity. For instance. Many of these mechanisms remain to be explored. different regions will allow them to be readily adapted to parallel computations. which focuses on the behav- iour of defects rather than that of atoms. low-angle versus high-angle grain boundaries) and compositions (e. There are common features between the mechanics of nanotubes and that of crystalline solids. to be performed. 6. most of these being of high symmetry. forming a low-symmetry configuration with respect to the simulation cell boundaries.e. two areas are worth exploring. polymer) is involved. A fair understanding of how a dislocation nucleates at a surface or grain boundary (i. M. the 5–7 structure in a nanotube glides in a similar fashion to a dislocation in a crystal.g. thus a comprehensive description of disloca- tion–grain-boundary interaction needs to be developed. The first is the direct linking between atomistic and mesomechanics or microstructure–mechanics simulations. The interaction of a dislocation with an interface depends very much on the nature of the interface. Transitional-continuum scales The significant role that defects play in determining the mechanical properties of metallic materials and the vast progress recently made in computational techniques has led to the emergence of the field of mesomechanics. such as DFT methods with classical force fields. Atomistic simulations are capable of providing interaction rules or mechanisms for a finite number of dislocation configurations. Thus mesomechanics describes the mechanics of heterogeneities and irregularities in materials.g. Different grain-boundary structures (e. such as materials science and biology. as would be necessary for certain materials science applications. including topological .3520 N. A dislocation behaves differently near a surface or at a grain boundary.g. a two-way connection between atomistic and DD simulations has not yet been achieved.3. dislocation absorption or blockage at interfaces) exists. In DD simulations. it is worth pointing out that the mechanics of nanotubes (and related nanocomposites) is within the reach of MD simulations today. the representation of atomic interactions can be a challenging issue. however.

Advances in areas such as the mechanical behaviour of very small volumes and dislocation– interphase interactions are expected to continue in the near future. Numerical and analytical homogenization techniques at the continuum level will be relied upon to a much greater extent than at present to model the behaviour of complex multiphase and polycrystalline microstructures. One of the most powerful mesomechanics methods is DD. Methods for a more direct coupling between DD simulations and continuum methods would also be required to improve the ability of non-local CM approaches to predict complex deformation phenomena. and compositional and structural heterogeneities. surface and volume). such as size effects. Then. microelectronic components and micromechanical elements. and extraction of average lattice curvatures. Theoretical methodologies that enable the proper coarse graining of space should be pursued. The increasing complexity of non-local formulations to predict size effects in multiphase materials and composites will require improved and more robust numerical schemes to be developed. Crystallographic approaches for single-crystal behaviour which rely on internal slip system variables will continue to provide the most powerful framework to incorporate basic mechanistic understanding in continuum models. This will enable the resulting constitutive models to incorporate explicit links between features of the microstructure at different levels and the macroscopic behaviour. New and efficient computational techniques for processing and visualizing the enormous amount of data generated in mesomechanical and continuum multiscale simulations must be developed. Such fundamental information can then be transferred to the continuum level to underpin the formulation of flow and evolutionary behaviour of CM-based constitutive equations. Progress in three-dimensional DD has contrib- uted to a better understanding of the physical origins of plastic flow and has pro- vided tools capable of quantitatively describing experimental observations at the nanoscale and microscale. details obtained by DD simulations can provide information about the . line. For instance. and quantitative predictions have been validated experimentally. grains and heterogeneous microstructures. As previously discussed. especially when a more physical descrip- tion of dislocation interaction with themselves and with grain boundaries or other obstacles is required. such as the properties of thin films. One of the most difficult future challenges is the development of explicit links between the transitional-microstructure scales addressed by mesomechanics approaches and the continuum level. further development of this type of multiscale material design capability will require a few challenges to be overcome. nanolayered struc- tures. thus imposing severe computational limitations. where considerable progress has been made during the past two decades owing to a variety of conceptual and computational developments. In its present stage of devel- opment. Multiscale modelling of nanomechanics and micromechanics 3521 defects (point. However. such as multipolar expansion techniques. DD have already addressed complex problems. as it may require to be dynamically described as quick avalanche events separated by long time intervals. mesomechanics approaches are needed to complement atomistic methods and to provide informa- tion about defect interaction and the kinetics of slip and interphase motion (see for example figure 11). It has moved from a curious proposal to a full and powerful computational method. continuum averaging of slip gradients. the issue of computational efficiency must be addressed so that truly large-scale simulations on thousands of processors can be effectively performed. Another example of the severe computational complexities that can arise with deformation is given by the highly intermittent nature of plastic slip.

A. 1988. Solid State Physics (Philadelphia. I. J. D. J.. J.. Strong Coulomb Correlations in Electronic Structure Calculations: Beyond the Local Density Approximation (New York: Gordon and Breach). F. Plast. Mech.. Plast. J. 2001. M. Amodeo.. 49. Rev. I. J. and Carlsson. S. 453. 309. 41. Mech. E. G. M. Phys. M. Batirev. and Srolovitz.. 309. 15. N. 48... and Lebowitz.. Busso. and to develop exciting new materials with extraordinary properties. J.. J. Mag. 1993. M. J. Phys. Int. and Zaoui. J. Besson. Bernstein. and McClintock. .. Soc. The integration of the approaches discussed in this section is expected to lead to more physically based multiscale formulations and establish materials by design as an approach to optimize the mechanical properties of materials with complex micro- structures. 26. 1. 17. M.3522 N. E. Phil.. Anisimov.. and by the National Science Foundation through grant DMR-0113555. V. 1987.. 6958... S.. 41.. Finnis. Mag. F. N. 77. J. 1996. R. Phys. A. 25. Brown. 1991. Solids. 2666. Mech. 81. Lett. and O’Dowd. 1976. Baker. Phys. 1995. Bulatov. Bortz. 1995. 1990a. C. and Andersen.. 59.. D. 1967. Phys. Office of Fusion Energy. P.. 1979. first edition (Paris: Hermes Science). Mater. 2000.. V. 308. 1987. 2000.. was supported by the US Department of Energy. L. Asaro. E. Proc. 669. 2001.. 1991. Kubin. P. Battaile. 95. 17.. 1985. Modeling of Deformation of Crystalline Solids. and Cailletaud. Broughton. 23. Zaanen. Mag.. Anisimov. F. 44. 525. N... 10. R.. and Parks.. Chaboche. References Abraham. F. and Needleman. A. V. 1998. Pennsylvania: Saunders College). V. 211. 6968. J. 2002. J. Barts. Solids. T. T. Phys.. and Charlier. Rev.. G. Mech. O.. and Beaudoin. Solids. Daehn (Warrendale. Bulatov. Plast. W. 3195. Decker. Acta metall. the work of California State University Northridge by the National Science Foundation through grant DMR-0097187 and by the US Army Research Office through grant DAAD-19-00-1-0049. Res.. Phys. B. B. 52. E. Arsenlis. N. F. E. and Ghoniem. Int. Mech. 1990. and Cailletaud. J. D. L. I.. S. C. 1987. J. and Mermin.... A. S. 1990b.. Ghoniem et al. Rev. Berveiller. 391. R. J.. Solids. Phys. Phys.. Pennsylvania: Metallurgical Society of AIME). 209. and Kaxiras. Busso. F. 943.. W. L. Ajayan.. Mecanique Non-Line´aire des Mate´riaux. 82. and Vitek.. 2333. Alavi. J. M. N. Meissonnier. A. J. 2002. J. Busso. Abraham. Phys. Los Angeles. 538. 12.. Nature. 363. 137. A. D. Rev. J... F. and Deutsch. Phys.. B.. 1979... 751. Acknowledgements The contribution to this work by University of California... 1998. 72.. Rollett and G. B. 1983. 1982.. Barbe. A. F. Kalos. E.. E. 1510. Cailletaud. and Yip. Yip. J.. B. evolution of the material topology. 2003. R.. and the dislocation density tensor which can lead to precise descriptions of lattice curvatures.. through grant DE-FG02-03ER54708. S. Mech. 2000. Phys. V. ibid. G. Phys. and Argon. 33. M. Ravikumar. V. P. Asaro... appl. Res Mech... A. G. T. H. Anand. 41... Phys.. 2213. Phil. 1975. V.. B.. and the contribution of Imperial College London by the Engineering and Physical Research Council of the UK through grant GR/N12312. 255.. Acharya. Rev. Comput. 1997. 1977.. R. Q. 1998. 325.. p. Solids. Rev. 10 324. Paris. Solids... and Forest. Radiation Effects Computer Experiments (Amsterdam: North-Holland). Lett. Phil. M. 82. J.. Phys. J.. Solids.. 125. 1996. A.. J... G. 105. A. Devincre. Lett. Phys. 1999. 48. 44. 1437. Comput.. J. Baskes.-C. 513. Rev.-L. 6293. Ackland. V.. edited by P. L. Mech. J. Phys. 12.. Beeler. A. Phys. D. and Kothari. all of which are gratefully acknowledged. Int. Lowe. Aifantis... 50. Ashcroft. and Rice. Symp. I. IV. Jeulin. T.

Acta mater. Hansen. M. Mater. p. 1. 1970. B. A. E.. L. Mater.. Finnis. Pedersen. 4127. 1992. Dreizler. 1984. Tosatti. 1985.. V. Phys... Non-linear Finite Element Analysis of Solids and Structures. E. 2001. Phys. M. 3265. P. C. Rev.. Yip.. 1984. Rev. Ab initio Calculation of the Structure and Properties of Molecules (Amsterdam: Elsevier). and Parrinello. Microstructures and Physical Properties... 1998. Acta mater. A. Modelling Simulation Mater. S.. P. Mag. S.. Proc. V. 1994.. H.. Ercolessi. and van Haeringen. D.. S.. Chen. S... 61. A. Doran. Sci. Cleri... 2471. Sci. A. 719. 93. Multiscale modelling of nanomechanics and micromechanics 3523 Cai. T. Canova. J. Dumitrica. 2001. 3471. 1992. 2629. Rev. V.. J. J. 1993.. S. Crisfield. Mech. 55.. Lett. R. St.. Gao. Wolf. Elstner. D. Elsner. Cohen. D.. B.. Brechet. Z.. R. Lett.. J. 49.. U. 87.. V. Pennsylvania: The Minerals. V.. Soc. Engng. G. A. p. M. and Mauri. J. G.. 73.. 491. Frauenheim. Phys. U. and Yip.. Rev... 583. Engng... Phys.. Jungnickel. and Mughrabi. and Liu.. 29.. 1193.. J. Phys. 40. Kubin. 1997.. J. M. Canova. B. Brechet.. Fulde. M. W. P.... G. G. Raschke.. G. 1481. M. 1998. Dykstra.. L. 1979. 10. J. Fleck. Leffers. and Smyshlyaev. and Hutchinson. Adv. 40. and Barnett.. 11956.. D. R. and Yakobson. Zugic. Campbell. Acta metall. DeVincre. Mech.. Chan. G.. and Kubin. Mater. R. J. 18. Phil. 1.-P. Lilholt. Europhys.. O. M. T. 1285. M. Mech. Soc. Mehl.. D. Int.. Modeling of Plastic Deformation and its Engineering Applications. 1998... S. Belytscko. 50. Solids. Daw. 1994. Diaz de la Rubia. B. 50. and Kubin... Lolian (New York: Plenum). and Huang. and Vitek. 40.. 1996. fourth edition (New York: Wiley).. N. S. Gear. J. M. 64. Lett. 1997. .. 14 694.. Phys. D. C. G. T. Phys. 91.. Esche. Pontikis. Phys. L. H. 85.. Essmann. New Jersey: Prentice-Hall). T. 4993. Porezag. and Condat. 2000. 160. Duesbery. DeVincre. Car. Eschrig. 1992. B.. Lorentzen. Pehlke. Engng. 57. Treatise on Materials Science and Engineering (Pittsburgh. Phys. El-Azab. V.. 1997. B. 45. Rev. M. M. K.. A. Phys.. Bulatov... Lett. Bilde-Sorensen. 50.. 123. Engng. W.. 171. DeVincre. Phys. P. D. Anderson. 46. 1989. F. Radiat. King.. G. B. P. and Adams. Optimized LCAO Method and the Electronic Structure of Extended Systems (Berlin: Springer).. Lett. H... and Yip. 731.. 1990. H. Lett.. 1986. Sci. and Sieck. Foiles. M. 249. Chen. 3213. L. Brechet. 24. and Kubin. Ercolessi.. T. M. Meth.. 2. Density Functional Theory (Berlin: Springer).. Rev. 26. 249. 1997. Gong. 2001. A. Y.. 376. 45. Juul Jensen. Palumbo. Rev.. 2000. 4577. 1998. J. Denteneer. 118. and Baskes.. Rev. Condat. M. Phys. S. and Sinclair. 1976. Canova. edited by J. and Condat. P. F. edited by M. 2... Cuitin‹ o.. I. D. W. Haugk. chem. T. and Gross. B... Lett. Chang. Numerical Initial Value Problems in Ordinary Differential Equations (Englewood Cliffs. Y. S. Bulatov. J. 46. Solid. Forest. Phys. S. M. and Phillpot. 1988. and Aust. 225. Du«rr. 205 502... Cai. W. B. 9485. and Ho« fer. Y.. E. K. Lassila. C. and B. P. A251.... U. R... A. Scripta metall.. D. 559. 86. X. Roskilde: Risø National Laboratory). Res. 33. Rev.. G.. L. N. Proc. DeVincre. Effects. Vols 1 and 2. 1993.. Lett. B. 113. J. E. H. R. T. N. 48. and Ortiz. Mareschal and B. A309. Y. Microscopic Simulations of Complex Hydrodynamic Phenomena... Rev.. Electron Correlations in Molecules and Solids (Berlin: Springer).... Galli. X. 1957. edited by S. A. 2001. 413. K. V. 1994. Huang. 1971. and Altan. Ralph (Risø.. numer. 1983.. F. 1990. V.. deWit. Gavazza.... Phil. L... Lett. 1994. M. Nikkel. Kinzel. Symp. M. Phys. Eshelby.. Mag.. Solids. Metals and Materials Research Society). M. 1960. D. Erb. B. 2245. 2003.. J. de Koning. F. Sci.. Engng. R.. G. Pontikis.. E.. Rabiet (Geneva: CH- Transtech). 241. M. Hughes. 295.. and Papaconstantopoulos... P. Phys. Rev. Modelling Simulation Mater. 6443. appl. U. 1992. Phys. 2003. Shu. and Parrinello.. Acta mater. Y. 1985. I.

C. L. and Gunnarsson. Engng Sci. O. Rev. Phys. Instrum.. Phys. J. M. 1. F. 2002. Hoagland... J. Rev. J. Phys.. 1950. 1992. 1998. Electron Correlations and Materials Properties. 81.. Phys.. 1170. 61.. J.. Non-Linear Phenomena in Material Science II.. Ha« hner.. 81. 60. 2215. Phys.. C. 23. Appl. 56. S. 26. S. T. 114. 1965. Commun. 1992. Ghoniem. Rev. and Gao. Phys. W.. 4. 913.. 689. B. Hamilton.. A. 283. 689. A. Kern. Comput. Lett.. and Zbib. 1995. and van de Walle. G. 164. O. M. Huang. Quebec City. M. 2001. Ghoniem. N. C. 31. and Baskes. Rev. Solid St. 843.. J.. Martin (Dordrecht: Kluwer).. Rev. Phys.. G.. Diaz de la Rubia. R. J. R.-H.. mod. Z.. Q. Solids.. Scripta metall. B.. R. 1974. 67. M. J. C. edited by N.. 23. 55. 1459.. Wei.. I. 19. Comput. Phil.. edited by D. 48.. Prog. S. 1985. 1976. Phys. W. N. S... 47. Res Mech. Mag.. 1969... Int. 89. Rev. R. Ghoniem et al. J. Phys. 143. fourth edition (Oxford: Clarendon). P. S. Sci. Phys. Commun. andWoo. Mech. and Pawley.3524 N.... I. L. A... Rev... 4. and Kohn. Heinisch. and Brune. Chrzan. Phil. Hughes. Gunnarsson.. Wong. 1999. S. Gillan. 2064. Defects and Material Instabilities. 197. M.. Res. and Amodeo.. N. Z. Phys.. 3–4. 1976. E. 653. K. 2127. Giovannini. and Wang. O. E. Zhang.. Plast. 2001. Huang.. Phys.. and Tsakmakis. A. Mag. Engng Mater. Rev. K. J.. 2000... Jordan. The Mathematical Theory of Plasticity. 1347. 55. Int. M. Huber. 121. 1. J. C. N. Matthews. Z. 965. J.. Krauth. Hirth. Phys. Mag. 2470. 1..-H. Solids. L. H. Meth. and Gehlen. . G. 3. Rev. 307– 311. 13. J. Ghoniem. 1988b. Phys. K. M. Stumpf. Ismail-Beigi. 1085. 68. and Sun. S.. 1996. A. M. V.. I. Kreibich. S.. P. 1997. M. Ghoniem (Dordrecht: Kluwer). 377. Engng Mater. M. 29. and Lee.. E. Rev. K.. Proceedings of the International Conference on Plasticity. Mater. K. Gumbsch. K. Lett. 1989. Patterns.. J. 1999. I.. 989. Phys. Ghoniem.. 312. Engng Mater.. J. Matter. 43.. 1990.. Acta mater. 413.. Modelling Simulation Mater. 1993. D. B. Lein. S. H. and Zaoui. H. Rhee.. M. 1995.. P. Mech. T. Hedin. Phys. T.. 393. A. X. C.. 1695. H. Technol. 4274. M. M. 82. Hill. L. M.. Huang. Srolovitz. Engng Sci. 1990... Singh. 1145. Hamilton.. 121.. J. 29.. 1964. 473.. 102. Zbib. J. Groma.. Huang. Bromann. Goedecker. Phys. P. M. Rev. Canada. N.. and Jones. and Arias. 4488. Georges. N. and Wen. B. Rev. J. F.. P. Janotti.. Liu. N. 13.. J. Rhee. 1990. K.. Lett. and Hansen. M. N. L. Phil. Gurtin. 136. J. Kubin and G. 136. Mag.. D. this issue. N. and Lomdahl. Phys.. Hirth.. Lett.. 1998. C. 1980. 1999. 31. and Walker.. H. Guluoglu. 89. 1971.-Aided Mater. Hoover. H. R.. 1989. S.. Rev. B. 1999. C. M. Technol. Kotliar. O. Phenomena. R7402.. J... Mech. 82.. 5807. Gross. Golubov. S.. Phil. Harris. Phys. 128. Engng. B864.. and Rozenberg.. M. H. C. and Petersilka.. 47. Nucl. Phys. 1999. N. 2002. 34. A.. 1589. 2002. p. M. C.. Lett. M. and Lundqvist. 197. and Amodeo... 1999. 2003.. 1997. 1993.. J.. 2000. Ghosh. Rep. 1996. G. A. of mod. Solid St. Phys.. N... G. 1990..: condens.. Rev. and Sun. LeSar. Huang. Phys.. Phys. R. Ghoniem. Herve.. Gonis and M. 615. B.. A.. Ha« hner. 62. Jones. 303.... Bay. Solids. J. Des... J. Lett.. R. and Hirth. and Lundqvist. Hartley. 4664. Hohenberg. Liu. J. 61. C.. Tong.. Ghoniem. 45. Phys. J. N. 329. edited by L. 1999.. Int J. B. Z. R.. P. and Kratochvil. Ghoniem. Walgreaf and N. 086 403. U. R. Ghoniem. and Zaiser.. 1989. Technol. 1998. Phys. R. D... Comput. P. B. H. Groma.. Phys. Science. James.. Gregor. J. M. Ghoniem. 14. 82. nucl. J. Tong. 137. H. A. I. E. p. Rev. W.. 1988a. Nowak. 2003. 1996. Mech. 71.. K.. mod. 3. 47. 1999.. P. 19. 60. J.. Ciftan (New York: Kluwer-Plenum).

. p. 6880. G. 1993. and Cailletaud.. Mehl. 83. S.. Multiscale Modelling of Materials. Marzari. Ghoniem (Boston.. L.. 72. 2001. A. G. E.. Kido. Lu. R5467. W. and Ciftan.. Mater.. J.. Louie (Boston.. Int. 2613. M.. 26. Phys. B. Comput. G. L. J. D. LeSar. L.. A. 213. 537. and Sham. Lu. 2001... Khachaturyan. I. 62. R.. Lu. P. Diaz de la Rubia. V.. edited by F.. 624. and Kioussis. and Kaxiras. mater.. C. 140. Microindentation Techniques in Materials Science and Engineering. 1979. Li. Shull (Warrendale. Appl. J. and Bulatov. Nabarro (New York: North-Holland). T. Phys. S. Phys. B.. 1924. Massachusetts Kluwer). A. J. P. T. 23. P. 10 891. Kaxiras and N. A. 106.. 54. and Freund. Kossevich. Lewis. N. 947. Woo. Phys. H. 1991. Phil.. Scripta metall. W. 1992. Acta metall. Mech.. Zhang.. C. Blau and B.. 2000. Georges. this issue. Phys. 129... 2000.. Comput. Phys.. Louie. J. B. Kratzer.. H. Technol. J.. mater.. and Shi. 135. Lubarda. Bronkhorst. B. 1989.. 2001. and Vanderbilt. L.. Kratochvil. A. P. Shi. Rev. Chelikowsky and S. 1. 155. angew. 095 501. J. 1987.. J. and Kioussis. Plast.. Kioussis. Kalidindi. 1996. C. M. B. Proc. A.. Cai. Les Houches Session XXXV (Amsterdam: North-Holland). A309. ASTM Special Technical Publication 889. 63. sol. 1996. Data.. P. and Woo. Comput. and Huang.. R. Rev. R. Lawn (Philadelphia. E... 957.. R. and Saxlov'a. Phys. p. P. 89. Q. 36.. Phillips. J. Phys. Maruyama. 1958a. T. Kent. Dislocations in Solids.. Kubin. Massachusetts: Materials Research Society). Bulatov. Meric.. 877c. 2000. M. W. Quantum Theory of Real Materials.. G. S. G... edited by E.. Rev. 5.. A. H. 1997. V. A. Turchi and A. and Chiba. E. 2001. 2001. E. C.... 2989. C. Engng Mater. LaBella.. N. 201. X. Crystal Lattice Phonons. 1992. 1969. A.. 2002b. B. A1133. Q. V. 1991.. G. Phys. G. Scripta metall. Phys. D. J.. S. 1958b. Phys. 15. p. V. L. Rev. Diffusion Defect Data—Solid St.. and Scheffer. V. P.. L. Rev. 56. 144 110. Phys. O. Stat.. edited by J.. B. Devincre.. Blume. Bulatov. E. D. 1999. 17.. p. 12 847. Rev. 171.. 1998. and Vanderbilt. Kioussis. 2003. N. Meric.. and Kubin.. and Anisimov. 455.. Pennsylvania: American Society for Testing and Materials). Leffers. Thibado. Phys. 1993... and Kaxiras.. Lee. J.. Bullock. 59. Soc. Lett. Lu.. Lett. Lu... and Canova. 113. Rev. R.. Rev. Vol. Kubin. 2002.. Physics of Defects. Kubin. 27. Kioussis.. 86... and Needleman. Ohsawa. G. 2002a. 2000. W. B. 891. 87.. and Zunger. Poubanne.. 1997. B. K. B. Proceedings of a Symposium. Q. D. R. 1995. A. F. . 1965. Ordejon.. 47. Lichtenstein. K. N.. J... Phys. G. and Papaconstantopoulos.. B. Rev. Pontikis. and Meille.. and Sankey. G.. E. A. and Kratochvil. V. Meissonnier. Rev. C. N. P.. 217. Langreth. Shi. Y. 41. V. C. M. Engng. 601.. (a).. and O’Dowd. mater. Liu. 463. Mag. 1992. Niita. Acta mater. 99. J. B. Q. V. 3069. Kuramoto. N. D. 64. Sci. and Perdew.. Huang.. B. 23–24. 1999. Rev. 4519. LeSar. T.. 2000... Yang.. 1981. 293. P. Phys. K. Solitons. Rev. 144 103.. 40... 52. Lett. Q. B. J. Erg. edited by V. Phys.-Aided Mater. 55. 80.. Busso. 469.. P. and Rickman. F. Zaanen. 433.. 625.. P. S. Lennard-Jones. Mater. J. 1977. S. J. 80... Li. N. H. Bulatov. 2001. Science of Alloys for the 21st Century: A Hume–Rothey Symposium Celebration. D. 2884. Technol. Scripta metall. P. 833. M. Lett. 1992.. 33. H.. Rev. Kro« ner. Wu.. Rev. Loubet.. p. 65. Sci. Engng Mater. M.. C. N. 1999. 2002. Phys. J. Lepinoux. Pennsylvania: Metallurgical Society of AIME). E.. Najafabadi. B. G. G. Solids. 83. A. 7.. 3. 66. 113. Liu. G... 663. W. L. 1. 162–170.. edited by P. V. Multiscale modelling of nanomechanics and micromechanics 3525 Justo. G. Modelling Engng Sci. Nunes.. P. Phys. Kukta. and Cailletaud. M. L. Des. p.. Rev. R. Lett. T. J. Kubin. appl. R. 21.. and Dislocations (New York: Wiley). Rev. and Srolovitz. P. and Tsutsumi. R.. 024 101. H. Kontinuumstheorie der Versetzungen und Eigenspannungen (Berlin: Springer). and Brechet.... Nucl. Phys.. Condat. Kohn. Mech. I. L. 1993. and Anand. L. 1986. 1. M. Phil Mag. Canova. 17.. W. Phys.. E. J... 3099.. de Koning. 2002. Math..

. appl...... 2001.. Acta mater.. Gonis. 97. 1984. Molinari. Serra. B. Dislocations in Solids. 7905. 1971. A. P. A. T. 6466. R... P. 34. B. M. 276. Ponte Castan‹ eda. and Yang. 36. 31. 6. B.. S. 65. Joannopoulos. A. Rev. A. 59. 1983. 2339.. T. Rosenbluth.. phys. S.. Peach. and Woodward. Eschrig (Berlin: Akademic). Phys. K. Acta metall. 1994. 287. Parrinello. Lett. So«derlind. A.. Mag. M.. 2002. B. A.. Asaro. 1983. Wills. 2825.. Phys. P. Ciftan (New York: Kluwer and Plenum). J. 1996.. 1997. 1991. P. Phys.. I... 436. Phys. 23. C. P. 124. Acta mater.. H.. R.. 179.. edited by F. Pettifor. edited by O. Advances in Materials Research 3 (New York: Interscience). 1981. 84. 114. Belak. 44. 1981. Rao. J. Rosenbluth. L. Hernandez. 3801. C. appl. Arch.. 1. J. Phys. Sci. M. 1983. Mag. D. 2000. 695. H. 2002. and Freed... Phys. G. 1988.. 31 951.. N. 2799. Lett.-W. 1996. Farkas. Moriarty.. 1992. Ordejon. p. Phys. Phys. A. Price. A.. 7182. A. J. N. 15. 1987. H. J. Parthasarathy. Phys.. Rinzler.. C. J. 59. Rev. D.. Metropolis. C. G. Phys. P. Pierce. Rev.. Moriarty.. J. Phys. Mughrabi. J. Phys. Rev. and Koehler. Neuha«user. 11 368. Phys.. D. M. G. 255.. 21. Rev. Ohsawa. Mura. Papanikolaou. Osetsky. Phys. Electron Correlations and Materials Properties. Phys. Mater. 1233. Phys. J. A. 1989. J.. W. Density Functional Theory of Atoms and Molecules (Oxford University Press). and Kuramoto. B. E. Bacon. M. Nan. and Joannopoulos. N. 2002. 52.. M. Matter.. Ghoniem et al. Brulin and R. P. 1989.. 1929. 14 646.. E. 1087. J. 4124. Burke. J. Ziesche and H. A. Nose. Moulin. Mu« hlhaus. L. M. Electronic Structure of Solids ‘91. and Bash. Chem.. and Widom. Rev. 48. 81... 1516. P. Moriarty. A. Nunes. D.. N. Colbert. and Dederichs. 1997. B. and Technol.. 1045.. 77. Matter. Soc. 1317. Lett.. M. J. D.. Int. C. Pasianot.. Am. M. B. 1989.. 19... Bennetto. edited by P.. Y. O. E. Phys. S. 422. edited by A. and Yang. J. B. 1999. 1968. J... Mech.. E. Perdew. 14. Engng Mater. 1993. Rev. Y.. 8487. 1989. Phil. Drabold. and Clarke. and Busso. Vol. Thess. 1982. D. 45.. Proceedings of the Fourth International Conference on Continuum Models of Discrete Systems. 1950. and Vanderbilt. and Woodward... A. T. 1996. D. Rev. 43. Rev. Rev.. J.. B.. mod.. 54. Rappe. 39. J. Morse.. Nikolaev. K. E. Rev. 2000.: condens.. P... Solids. 4945. 2480. Singh.... 1992. P. R. 14.. Phys... 1999. 1998. chem.... Perdew. Allan. 266. Phys.. and Needleman.. 52... A.. M. Mater.. and Teller.. 3199. H. 86... A.. P. and Zunger. P. 231. and Golubov. and Kubin.: condens. Grumbach. 6952. Condat. 39. M. and Cooper. B. F. Ortiz.. Payne. S. A. J.. T.. Micromechanics of Defects in Solids (Dordrecht: Martinus Nijhoff). H. Adv. Pettifor. 1999. Acta metall. nucl. Rev. 45. Phil.3526 N. C. I. 38. Mech. and Savino. D. 77. ibid. Zeller. J. J. H. 64. Nabarro (Amsterdam: North-Holland). Rudd. and Wang. 1991. A. R. J. W. S.. F. D.. A. R. M... 62. and Oleinik. Phys. S. Perdew.. 1981. Teller. A. The Continuum Theory of Dislocations. Solids.. A. J. B. Lett. R... 114. 80.. N... Simmons. J. Techol. Teter. D.. 50. Phys. and Martin. 63. P. and Cooper.. R. D. M. 1992. Parr. D. Engng Mater. 5048. Streitz. P. 1953. K. 46. Rev. Rev. 1992. Engng. 57.. 1997.. Mol. 31. W.. 12 431. Nouailhas. and Phillips. Hsi (Amsterdam: North Holland). B. .. Acta metall. Kioussis and M. A. Arias. R. 85. 16 533. chem. T. P.. D. Rao.. J.. and Smalley. p. 2002. 124. Phys. B. 56. Phys.. Rev. D.. 1999.. G. 11. 1367.. Price. J. Neumann. R. Mech.. appl. Pitakthapanaphong. and Rahman. J. 49. 1991. D.

5. M. Schonfelder. 133. Can. Phys. 245. 1985. J.. 143. Vosko.. Small. J. Mater. C. E... and Sakuma. Rev. V. and Aifantis. Stainer. W. Engng. and Zunger. 95. 1199. Phys. S. Keblinski... 296. E. Imamoto.. A.... F. 1997. Engng. 56. Rev. and Srolovitz. Rodney. C. and Goddard. G. Sinclair. H.. 61. Shu. A. 632. Phys.. J. and Farkas. H. Engng.. A309–A310. 1511.. B. and Nastasi. Nishino. 1972.. Furuta. and Thomson. T. Mater. Crystal Lattice Defects... A. Solids. Part I. 1996. 1997. von Barth.. Schubert and P.. 1990. R. Y. Zbib.. 51. and Weber. Lecomte-Beckers.. I. D. 1985. 27.. 47. K.. J.. M. F. N.. Y. Acta mater. Wang... S. 2002. p. J. Science.. 2003. Stochastic Processes in Physics and Chemistry (Amsterdam: North- Holland). Tersoff. Textures Microstruct. Schwarz. 1999.. Sci.. and Martin. 1971.. Lett. 1988. J. 5. 82.. Wilk. 1999.. Wang. Nagasako... Phys.. 464. and Hedin. Yamada.. Phys. and LeGoues. N. Swadener. and LeSar. 1220. T. 1148.. L. 226 101. L. F. and Woo... F. and Furtkamp. 2001c.. 6... I.. H. 2003. Cuitino.. C. 31.. Proceedings of the International Conference on Dislocations 2000 (Gaithersburg. Materials for Advanced Power Engineering. Shim. L. Busso.. 3. 1992. 1361.. K. R. Lett. and Gunnarsson. and Kioussis. Suzuki.. 2000. Sci. and Nusair. 24. Sci. G. Y. Y. 50. Phys. W. Ito.. H.. S. 78.... D. Y. Introduction to Percolation Theory (London: Taylor & Francis). van Houtte. Super. T. T. 343. 101. A. L. Lu. LeSar. A. 1992. Wolf. Rev. F. and Ortiz.. E´cole Norm... Meth. Saito. Rev. Sci.. Chen. Lett. Rickman.. C.. 2002. P. Modelling Simulation Mater. Phys.. Maryland: National Institute of Standards and Technology). Int. H. M. 2001b. Hirth. L.... Wang. 294. C. Q. Phys.. Phy. 1200.. Thomson. and Steinhaus. 44. N. M.. R. Rev. Scripta metall. Rev. A234. Scudder. Fleury.. D. and Phillips. B. 283. and Needleman. W. Wolf.. and Samajdar. Engng. Forum. Van Kampen. Phil. M.. B (to be published). Numer. 1974. Mech.. Segmueller. Cagin... 3884. Y. ... 1994. Vitek. 58.. R.. Sci.. Mater. 1907.. 49.... Engng. Paris. Cutino. E. R.. R. J.. and Aharony. Daniels. and Levine. M.. and Murakami. M. H. 1986. Kim. 1704.. B. 59. J. J. and Savage. Levine. Solids. Lett. M. Clemens. M.. 2002. A. D. A. and Lundqvist. 1999. J. D.. Ikehata.. Shim. Lett. and Allen. H. Huang.. J... E.. C. F. A.. Engng. 373. Rev. Rev. Rev.. D. 49. Engng. Sci. Sci. 1998. 1999.. 2002. K. King. N. Sci. 1997.. Appl. 1998. 2000. H. Svane. Thomson. M. L. N. Engng.. p. 9. 2000. Phys. Van Swygenhoven. Jin. J. Science. P. and Zatloukal. 2001. K... R.. K. Rhee. J. Lett. Kratochvil... Phys. Comput.. 1629. 2002. 1995. 4785.. B. Acta mater. Phys. J. Woo. J. Phys. G. Shi. Mater.. L. 300. 5. Engng Sci. Schonfelder. B. 1980. Huang. Mag. Shu.. 79.. and de la Rubia. 109. 69. E. Stillinger. Hoagland. C. Mater. H. appl. 401. edited by J. and Fleck... 467. A. V.. Modelling Engng Sci.. T.. and Tersoff. 8714. Volterra. Phys. 2001. O’Dowd. 31. 1877. 205. 5678. G. N. A. O.. Kuramoto. 1351. G. Lett. Rodney. R. and Huang. Treatise Mater. 245. 1847. Lett. 23. Hwang. 107. Strachan. R.. Saxlov' a. Misra. Phys. W. W... and Nix. Wang. P. Modelling Simulation Sci.. Schwarz.. Interface Sci. Phys. H. Van Swygenhoven.. J. J. 340. 1997. Int. J. and Khachaturyan.. D.. M.. Lett.. Multiscale modelling of nanomechanics and micromechanics 3527 Regino. A. D. 66. 81. Wang.. Annal. Fleck.. Carton. 25. B. V. Levine. J. Mater. Seno. 2001. 23. 71. Caro.. Phys. S. J. Y. D. U... 1. Ruberto. 78. B... Schubert. Schwarz. 8.. T. Phillpot. Van der Giessen. A. Rev. A. H.. 2003. Mater. M. H. R.. J. 5262.. Phys. 2001a. 947. 149. A.. N. Delannay. and Phillpot... B. Yourdshahyan. 2002. 42. N. Ennis (Forschungszentrum Julich). 65. A. T... J.. Mech. S. K. A309. Appl. Rev. 309. H... Sci. Ikuhara. Comput. 5321. L. 440.-W. 9. Stauffer. C. M.. Nonaka. 2002... A. 1999.. Walgraef... 88. Y. N.... Res.

. Lomdahl. Int. X. Rev. J. 113. 261. 265. Sci. Zhou. L27. Science. Sci.. Engng (submitted). E. fourth edition (New York: McGraw-Hill). S. and Aifantis.. 2002. E.. Ge. 23. and Freeman. Yildirim.. Xie.. Phys. Appl. A309–A310. and Wang. H. Phys. Comput. M. Phys. E... Phys. L. 304. Zienkiewicz. P.. C.. W.. 1998. Q.. M. Gu« lseren.. B... M. Phys. 1994. 2002. C. Weinert. Proc... 1998. 308. Science. and Sriram. L. M. 133. Lee.. R. S. Freeman. Krakauer... H. 82. 267. The Finite Element Method. Avlonitis. C.. and Chen. Zaiser. 075 404. Lett. Wu. R.. Wang. O. Rev. and Ha« hner. G. W. Rev.. J.. H. H. Woo... sol.. LeSar.. H. Phys. 9.... Soc. and Hirth. Mater. Zbib. I. 2003b (to be published). Woo. and Moriarty. and Puls. O. 88. C.. C. Res. A. Z.. J. R. Science Engng. Wimmer. Mech.. 864. 376. and Olson. W. 1988. 76. C. T. M. Acta mater... N. 24. 64.. Phys.. J. Zhang. N. 1994. 216 402... 1998... D. 2001. 1993.. Woodward. 1991. appl.. and Ghoniem. Q. A. and Beazley. Symp. M.. Zaiser. Zaiser.-K. Preston.. Res. 2001. Zbib... B. M. 348. 4143. S. T. B. 9. M.3528 Multiscale modelling of nanomechanics and micromechanics Wang. and Ciraci.. X. Rev. 2003a. 578. and Zhu. 40. . A. 1981. and Taylor. 1998. Yang.. Mater. S. 46. C.. J.. 1976. J. 1996. J. A. 1997. Phys. Huang. 1438. R. C.. 54. R. J.. 1997. LeSar. Lett. Sci. 66. Mater. R. Xu. 101.. and Rao. Ghoniem. N. Mech... M. Phys. P. Wu. W.. D. Stat. 1525. P. Rev. S. Z.. Rhee. P. Modelling Simulation Mater. and Aifantis. 279.. (b). 199. 6941.