Advanced Engineering Materials Progress Report on Computational Materials Science

PROGRESS REPORT
Don't Trust your Simulation -
Computational Materials Science
on Its Way to Maturity?
By Dierk Raabe*

Progress reports are a new type of article in Advanced Engineering Materials, dealing with the hottest
current topics, and providing readers with a critically selected overview of important progress in these
fields. It is not intended that the articles be comprehensive, but rather insightful, selective, critical,
opinionated, and even visionary. We have approached scientists we believe are at the very forefront of
these fields to contribute the articles, which will appear on an annual basis. The article below describes
the latest advances in Computational Materials Science.

1. A Model is a Model is a Model companied by scrutinizing the model ingredients and bound-
ary conditions of that calculation in the same critical way as
The title of this article is of course meant to provoke. Why? an experimentalist would check his experimental set-up.
Because a menace of confusing models with reality always In the following I will address some important aspects of
exists. Does anyone nowadays refer to ªfirst principles simu- computational materials science. The selection is of course
lationsº? This point is well taken. However, practically all of biased (more structural than functional; more metals than
the current predictions in this domain are based on simulat- non-metals; more mesoscale than atomic scale). I will try to
ing electron dynamics using local density functional theory. reach a balance between fundamental and applied topics. The
These simulations, though providing a deep insight into ma- article focuses particularly on topics and publications of 2001
terials ground states, are not exact but approximate solutions of and 2002.
the Schrödinger equation, whichÐwe should not forgetÐis a
model itself.[1] Does anyone still refer to ªfinite-element simu-
lationsº? This point is also well taken. However, also in this 2. Some Semantics of Modeling and Simulation
case one has to admit that approximate solutions to large sets
of nonlinear differential equations formulated for a (non-ex- Before continuing in medias res, let us revisit some basic
isting) continuum under idealized boundary conditions is thoughts on the semantics of modeling and simulation. The
what it is: a model of nature but not reality. words modeling and simulation are often distinguished by
But let us calm down and render the discussion a bit more somewhat arbitrary arguments or they are simply used syno-
seriousness: current methods of ground-state calculations are nymously. This lack of clarity reflects that concepts in compu-
definitely among the cutting-edge disciplines in computa- tational materials science develop faster than semantics. To
tional materials science and the community has learnt much establish a common language in this field, a less ambiguous
from it during the last years. Similar aspects apply for some
continuum-based finite-element simulations. After all, this ar- ±
ticle is meant to attract readers into this exciting field and not [*] Prof. D. Raabe
to repulse them. And for this reason I feel obliged to first Max-Planck-Institut für Eisenforschung
make a point in underscoring that any interpretation of a re- Max-Planck-Str. 1, D-40237 Düsseldorf (Germany)
search result obtained by computer simulation should be ac- E-mail: [email protected]

ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5 Ó WILEY-VCH Verlag GmbH, D-69469 Weinheim, 2002 1438-1656/02/0505-0255 $ 17.50+.50/0 255
 Raabe/Computational Materials Science on Its Way to Maturity?

definition of both concepts might be helpful. In current un- advantageous material properties. Following Haasen,[5] mi-
PROGRESS REPORT

derstanding the word modeling (model (Latin, Italian): copy, crostructure can be understood as the sum of all thermody-
replica, exemplar) often covers two quite different meanings, namic non-equilibrium lattice defects on a space scale that
namely, model formulation and numerical modeling. The latter ranges from Angstrùms (point defects) to meters (sample
term is frequently used as a synonym for numerical simula- surface). Its temporal evolution ranges from picoseconds
tion (simulare (Latin): fake, duplicate, mimic, imitate). I think (dynamics of atoms) to years (fatigue, creep, corrosion, diffu-
that the general synonymous use of the terms modeling and sion). Haasens's definition clearly underlines that microstruc-
simulation is not an ideal choice. Rosenblueth and Wiener[2] ture does not mean micrometer, but non-equilibrium.
offered an elegant comment on this point, which underlines Some of the previously suggested size- and time-scale
that the creation of models encompasses a much more gener- hierarchy classifications group microstructure research into
al concept than simulation. According to their work the gen- macroscale, mesoscale, and microscale (recently, nanoscale
eral intention of a scientific inquiry is to obtain an under- has become very popular, too) (Fig. 1). They take a somewhat
standing and a control of some parts of the universe. different perspective at this topic in that they refer to the real
However, most parts of the real world are neither sufficiently length scale of microstructures (often ignoring the intrinsic
obvious nor simple that they could be grasped and controlled
without abstraction. Scientific abstraction consists in replac-
ing the part of the real world under investigation by a model.
This process of designing models can be regarded as the gen-
uine scientific method of formulating a simplified imitation
of a real situation with preservation of its essential features.
In other words, a model describes a part of a real system by
using a similar but simpler structure. Abstract models can thus
be regarded as the basic starting point of theory. However, it
must be underlined that there exists no such thing as a uni-
fied exact method of deriving models. This applies particular-
ly for the materials sciences, where one deals with a large
variety of scales, mechanisms, and parameters. The notion
simulation, in contrast, is often used in the context of running
computer codes that have been designed according to a cer-
tain model. A more detailed discussion of these admittedly
philosophical aspects has been published before.[3,4]

3. Microstructure or the Hunt for Mechanisms

While the evolutionary direction of microstructure is pre-
scribed by thermodynamics, its actual evolution path is
selected by kinetics. It is this strong influence of thermody-
namic non-equilibrium mechanisms that entails the large
variety and complexity of microstructures typically encoun-
tered in materials. It is an essential observation that it is not
those microstructures that are close to equilibrium, but often
those that are in a highly non-equilibrium state that provide Fig. 1. Example of a multi-scale simulation for the automotive industry.

Dierk Raabe graduated at the Institut für Metallkunde und Metallphysik at RWTH Aachen in 1990.
After his Dissertation under the guidance of Kurt Lücke in 1992 he worked in the fields of computational
materials science, textures, and composites. After his Habilitation in 1997 he joined the Department of
Materials Science and Engineering at Carnegie Mellon University in Pennsylvania and the National
High Magnetic Field Laboratory in Florida. Since 1999 he is a director of the Max-Planck-Institut für
Eisenforschung in Düsseldorf. Among other recognitions he received a Heisenberg scholarship, the
Adolf-Martens Award, the Masing Award, the Best Paper Award 1995 of Steel Research, the European
Materials Science and Technology Prize, and the Dr. Meyer±Struckmann Award. The author is teaching
in the fields of computational materials science, textures, and polycrystal mechanics at RWTH Aachen
and at Carnegie Mellon University.

256 ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5

Raabe/Computational Materials Science on Its Way to Maturity?

time scales). This might oversimplify the situation and sug- classifications and decide how to integrate them into macro-

PROGRESS REPORT
gest that we can linearly isolate the different space scales scopic constitutive concepts or subject them to further de-
from each other. In other words the classification of micro- tailed investigation. In other words, the mechanisms that gov-
structures into a scale sequence merely reflects a spatial ern microstructure kinetics do not know about scales. It was
rather than a crisp physical classification (Table 1). For in- often found in materials science thatÐonce a basic new mech-
stance, small defects, such as dopants, can have a larger influ- anism or effect was discoveredÐa subsequent avalanche of
ence on strength or conductivity than large defects such as basic and also phenomenological work followed, opening the
precipitates. Think, for instance, about the highly complex path to new materials, new processes, new products, and
phenomenon of shear banding. This can be initiated not only sometimes even new industries. Well-known examples are
by interactions among dislocations or between dislocations the dislocation concept, transistors, aluminum reduction, aus-
and point defects but also by a stress concentration intro- tenitic stainless steels, superconductivity, or precipitation
duced by the local sample shape, surface topology, or pro- hardening. The identification of key mechanisms, therefore,
cessing method. has a bottleneck function in microstructure research and com-
However, if we accept that everything is connected with putational materials science plays a key role in it. This applies
everything and that linear scale separation could blur the particularly when closed analytical expressions cannot be
view on important scale-crossing mechanisms, what is the formulated and when the investigated problem is not easily
consequence of this insight? One clear answer to that is: we accessible to experiments.
do what materials scientists always didÐwe look for mecha-
nisms. Let us be more precise. While former generations of
materials researchers often focused on mechanisms or effects
that pertain to single lattice defects and less complex mecha-
nisms amenable to basic analytical theory or experiments 4. Models for Microstructure
available in those times, present materials researchers have
4.1. Who Needs Electrons and Atoms?
three basic advantages for identifying new mechanisms. First,
ground-state and molecular dynamics simulations have ma- In the introduction we mentioned the predictive power of
tured to a level at which we can exploit them to discover pos- ground-state and molecular dynamics calculations. But what
sible mechanisms at high resolution and reliability. This about possible collective mechanisms above this scale? Under-
means that materials theory is for the first time standingÐin standing microstructure mechanisms at these scales requires
terms of the addressed time and space scaleÐon robust quan- the use of adequate simulation methods. This domain, which
titative grounds, allowing us insights we could not get before. is sometimes referred to as computational microstructure
It is hardly necessary to mention the obvious benefits arising science, is rapidly growing in terms of both improved compu-
from increased computer power in this context. Second, ex- tational methods and scientific harvest. Among these, espe-
perimental techniques have been improved to such a level cially the various Ginzburg±Landau-based phase-field kinetic
thatÐalthough sometimes only with enormous effortsÐnew models, discrete dislocation dynamics, cellular automata,
theoretical findings can be critically scrutinized by experi- Potts-type q-state models, vertex models, crystal plasticity fi-
ment (e.g., microscopy, nanomechanics, diffraction). Third, nite-element models, and texture component crystal plasticity
due to advances in both theory and experiment more com- finite-element models deserve particular attention.
plex, self-organizing, critical, and collective nonlinear mecha- The predictive power of these approaches depends to a
nisms can be elucidated, which cannot be understood by large extent on their approach to deal with the transition from
studying only one single defect or one single length scale. the quantum and atomic scale to the continuum scale. In
All these comments can be condensed to the statement that other words, only three basic theoretical standard approaches
microstructure simulation, as far as a fundamental under- exist currently to map matter in a model, namely the quan-
standing is concerned, consists in the hunt for key mechanisms. tum scale (dynamics of electrons, relaxation of nuclei), the
Only after identifying those we can (and should) make scale atomic scale (atom and molecular dynamics), and the conti-
nuum scale. An essential challenge in any computational ma-
terials approach, therefore, lies in bringing these three con-
cepts into agreement and avoid contradictions. This means
Table 1. Classification of microstructures.
that simulation approaches that are formulated at different
scales should provide very similar results when tackling the
Scale Quantum Molecular Mesoscopic Macroscopic same problem. For instance, some dislocation mechanisms
mechanics dynamics have recently been simulated using phase-field theory,[6] line
Time 10±16±10±12 10±13±10±10 10±10±10±6 >10±6
tension modeling,[7] and even Monte Carlo simulation.[8] In
scales [s] the following, I shall enter the field of continuum simulations
far above the quantum and atomic scale and review recent
Length 10±11±10±8 10±9±10±6 10±6±10±3 >10±3
scales [m]
progress achieved by some of the important methods in this
domain.

ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5 257

 Raabe/Computational Materials Science on Its Way to Maturity?

4.2. Ginzburg±Landau-Type Phase Field Kinetic Models recent contribution was by Wang et al.,[6] who formulated a
PROGRESS REPORT

phase-field approach for tackling the evolution of sets of dis-

The capability of predicting equilibrium and non-equilibri-
locations in an elastically anisotropic continuum under an ap-
um phase transformation phenomena at a microstructural
plied external stress. Jin et al.[14] recently performed simula-
scale is among the most challenging topics in materials
tions of martensitic transformation in polycrystals. Mullis
science. This is due to the fact that a detailed knowledge of
and Cochrane[15] investigated the origin of spontaneous grain
the structural, topological, morphological, and chemical char-
refinement in undercooled metallic melts using the phase-
acteristics of microstructures that arise from transformations
field method. Wen et al.[16] tackled the coarsening kinetics of
forms the basis of most microstructure±property models. It
self-accommodating coherent domain structures using a con-
becomes more and more apparent that classical thermody-
tinuum phase-field simulation.
namic models are increasingly limited when it comes to the
prediction of new materials solely on the basis of free-energy
equilibrium data. This is due to the strong dominance of the 4.3. Discrete Dislocation Dynamics
kinetic boundary conditions of many new materials that have
microstructures far away from equilibrium. While the ther- Discrete continuum simulations of crystal plasticity, which
modynamics of phase-transformation phenomena only pre- use time and the position of each portion of dislocation as
scribes the general direction of microstructure evolution, with variables, are of great value for investigating mechanical
the final tendency to eliminate all non-equilibrium lattice de- properties in cases where the spatial arrangement of the dis-
fects, the kinetics of the lattice defects determines the actual locations is of relevance.[3] In these approaches the crystal is
microstructural path. The dominance of kinetics in structure usually treated as a canonical ensemble, which allows one to
evolution of technical alloys has the effect that the path to- consider anharmonic effects such as the temperature and
wards equilibrium often leads the system through a series of pressure dependence of the elastic constants and at the same
competing non-equilibrium states. For this reason, simulation time treat the crystal in the continuum approach using a line-
methods in this field that are exclusively build on thermody- ar relation between stress and the displacements gradients.
namics will be increasingly replaced by approaches that use Outside their cores, dislocations can then be approximated as
both the thermodynamic potentials, considering also elastic linear defects embedded in an unbounded homogeneous lin-
and electromagnetic contributions, and the kinetic coefficients ear elastic medium. The dislocations are the elementary car-
of the diffusing atoms and of the lattice defects involved. riers of displacement fields from the gradients of which the
Among the most versatile approaches in this domain are the strain rate and plastic spin can be derived. Despite the con-
Cahn±Hilliard and Allen±Cahn kinetic phase-field models, siderable success of current 3D simulations some extensions
which can be regarded as metallurgical derivatives of the the- of the underlying models are conceivable to render them
ories of Onsager and Ginzburg±Landau. These models repre- more physically plausible and in better accord with experi-
sent a class of very general and flexible phenomenological mental data, particularly when considering thermal activa-
continuum field approaches, which are capable of describing tion. One conceivable extension is the replacement of isotro-
continuous and quasi-discontinuous phase-separation phe- pic elasticity by anisotropic elasticity. Another aspect is the
nomena in coherent and non-coherent systems. The term qua- replacement of the phenomenological viscous law of motion
si-discontinuous means that the structural and/or chemical by the assumption of dynamic equilibrium and the solution
field variables in these models are generally defined as con- of Newton's equation of motion for each dislocation segment.
tinuous spatial functions, which change smoothly rather than Further challenges lie in the introduction of climb and related
sharply across internal interfaces. The models can work with diffusional effects into dislocation dynamics simulations on a
conserved field variables (e.g., chemical concentration) and physical basis. Most, if not all, of the suggested model refine-
with non-conserved variables (e.g., crystal orientation, long- ments decrease the computational efficiency of dislocation
range order, crystal structure). While the original Ginzburg± dynamics simulations. However, in some cases their consid-
Landau approach was directed at calculating second-order eration might be of relevance for further predictions. Impor-
phase-transition phenomena, metallurgical variants are tant publications in this field have been published re-
capable of addressing a variety of transformations in matter cently.[7,17±20] Apart from these classical segment and line
such as spinodal decomposition, ripening, non-isostructural models, a Monte Carlo model[8] and a phase-field approach[6]
precipitation, grain growth, solidification, and dendrite for- have been recently successfully introduced to this field.
mation in terms of corresponding chemical and structural
phase-field variables.[9±11]
4.4. Cellular Automata in Microstructure Simulation
Important contributions to this field in the last months
were published by Wen et al.[12] who investigated the influ- Cellular automata are algorithms that describe the spatial
ence of an externally imposed homogeneous strain field on a and temporal evolution of complex systems by applying de-
coherent phase transformation in the Ti±Al±Nb system. A re- terministic or probabilistic transformation rules to the cells of
lated simulation study by Dreyer and Müller[13] addressed a lattice. The space variable in cellular automata usually
phase separation and coarsening in eutectic solders. Another stands for real space, but orientation space, momentum

258 ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5

Raabe/Computational Materials Science on Its Way to Maturity?

space, or wave-vector space occur as well. Space is defined crostructures formed during recrystallization. The method

PROGRESS REPORT
on a regular array of lattice points. The state of each lattice was applied to starting data obtained by a preceding crystal
point is given by a set of state variables. These can be particle plasticity finite-element simulation (Fig. 2). Various nuclea-
densities, lattice-defect quantities, crystal orientation, particle tion scenarios were discussed also with respect to macroscop-
velocity, or any other internal variable the model requires. ic effects such as friction (Fig. 3), shear localization, and yield
Each state variable defined at a lattice point assumes one out surface evolution during recrystallization. Further studies
of a finite set of possible discrete states. The opening state of addressed the simulation of mesoscale microstructures and
an automaton is defined by mapping the initial distribution damage evolution.[28,29]
of the values of the chosen state variables
onto the lattice. The dynamical evolution of
the automaton takes place through the appli-
cation of deterministic or probabilistic trans-
formation rules (switching rules) that act on
the state of each point. The temporal evolu-
tion of a state variable at a lattice point given
at time (t0+Dt) is determined by its present
state at t0 (or its last few states t0, t0±Dt, etc.)
and the state of its neighbors. Due to space
being discrete, the type of neighboring af-
fects the local transformation rates and the
evolving morphologies. Cellular automata
work in discrete time steps. After each time
interval the values of the state variables are
updated for all lattice points in synchrony.
Owing to their features, cellular automata
provide a discrete method of simulating the
evolution of complex dynamical systems,
which contain large numbers of similar com-
ponents, on the basis of their interactions.
During the last years, cellular automata have Fig. 2. Simulation of primary static recrystallization within a single deformed grain. The upper graphs show
increasingly gained momentum for the the change in the distribution of the simulated dislocation density during recrystallization. The lower graphs
indicate different orientations by different colors (the white areas are as-deformed). The initial deformation sim-
simulation of microstructure evolution in the ulation was performed using a crystal plasticity finite-element model. The set-up consisted of a single alumi-
[3,21,22]
materials sciences. The enormous ver- num grain with face centered cubic crystal structure and {111} <110> slip systems, which was embedded in a
plastic continuum and had the elastic±plastic properties of an aluminum polycrystal with random texture. The
satility of the cellular automaton approach sample was plane strain deformed to 50 % thickness reduction. The resulting data (dislocation density, orienta-
for microstructure simulations, particularly tion distribution) were used as input data for a cellular automaton recrystallization simulation. Details of the
simulation have been published before [23].
in the fields of recrystallization, grain
growth, and phase-transformation phenom-
ena, is due to its flexibility in using a large
variety of state variables and transformation
rules.
Important fields where microstructure
based cellular automata have been recently
successfully used in the materials sciences
are primary static recrystallization and re-
covery, formation of dendritic grain struc-
tures in solidification processes, as well as
related nucleation and coarsening phenom-
ena.[23] For instance, Ding and Guo[24] have
recently addressed the complicated field of
dynamic recrystallization by using cellular
automaton simulations. Gandin[25] used a 3D
cellular automaton±finite-element model for Fig. 3. Two subsequent stages of a 2D isothermal simulation of primary static recrystallization in a deformed
the prediction of macrostructures formed in aluminum polycrystal on the basis of crystal plasticity finite-element data. The upper figures show the change
in dislocation density. The lower figures show the change in microtexture. The gray areas in the upper figures
casting. Raabe[23,26,27] conducted cellular indicate a stored dislocation density of zero, i.e., these areas are recrystallized. Details of the simulation have
automaton simulations of textures and mi- been reported before [26].

ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5 259

 Raabe/Computational Materials Science on Its Way to Maturity?

4.5. Potts-Type Multi-Spin Models for Microstructure contain networks consisting of large sets of interconnected
PROGRESS REPORT

Simulation boundary segments that meet at vertices. The system dy-

namics lies in the motion of the junctions and boundary seg-
The application of the Metropolis Monte Carlo method in
ments. The dynamics of these interface portions and of the
microstructure simulation has gained momentum particular-
vertices determines the evolution of the network. The motion
ly through the extension of the Ising lattice model for model-
of the boundaries and vertices is usually described in terms
ing magnetic spin systems to the kinetic multistate Potts lat-
of a linearized first-order rate equation. Depending on the
tice model. The original Ising model is in the form of a 1/2
physical model behind such a simulation, the boundary seg-
spin lattice model, where the internal energy of a magnetic
ments and vertices can be described in terms of characteristic
system is calculated as the sum of pair-interaction energies
activation enthalpies of their mobility coefficients.
between the continuum units, which are attached to the
The calculation of the local driving forces in most vertex
nodes of a regular lattice. The Potts model deviates from the
models is based on equilibrating the line energies of grain
Ising model by generalizing the spin and by using a different
boundaries abutting at junctions according to Herring's equa-
Hamiltonian. It replaces the Boolean spin variable where only
tion. The assumption of local mechanical equilibrium at these
two states are admissible (spin up, spin down) by a general-
nodes is for obvious topological reasons in most cases only
ized variable, which can assume one out of q discrete possible
possible by allowing the neighboring interfaces to curve.
ground states, and accounts only for the interaction between
These curvatures in turn act through their capillary force,
dissimilar neighbors. The introduction of such a spectrum of
which is directed towards the center of curvature, on the
different possible spins enables one to represent domains dis-
junctions. In summary, this algorithm entails a displacement
cretely by regions of identical state. For instance, in micro-
of both the nodes and the grain boundaries. Most vertex mod-
structure simulation such domains can be interpreted as areas
els use switching rules, which describe the topological recom-
of similarly oriented crystal material. Each of these spin ori-
bination of approaching vertices, for instance, in the case of
entation variables can then be equipped with a set of charac-
domain shrinkage. This is analogous to the use of phenomen-
teristic state variable values quantifying the lattice energy,
ological annihilation and lock-formation rules used in dislo-
the surface energy, the dislocation density, the Taylor factor,
cation dynamics. As in all continuum models, such recombi-
or any other orientation-dependent constitutive quantity of
nation laws replace the exact atomistic treatment. It is worth
interest. Lattice regions that consist of domains with identical
noting in this context that the recombination rules, particular-
spin or state can be interpreted as crystal grains. The values
ly the various critical recombination spacings, can substan-
of the state variable are entered in the Hamiltonian of the
tially affect the topological results of a simulation. Depending
Potts model. The most characteristic property of the energy
on the underlying constitutive continuum description, vertex
operator used in microstructure research is that it defines the
simulations can consider crystal orientation and therefore
interaction energy between nodes with like spins to be zero,
misorientations across the boundaries, interface mobility, and
and between nodes with unlike spins to be one. This rule
the difference in elastic energy between adjacent grains. Due
makes it possible to identify interfaces and to quantify their
to the stereological complexity of grain boundary arrays and
energy as a function of the abutting domains.
the large number of degrees of freedom involved, most net-
The Potts model is very versatile for describing coarsening
work simulations are currently confined to 2D. Maurice has
phenomena. In this field it takes a quasi-microscopic metal-
recently successfully introduced a 3D version of a network
lurgical view of grain growth or ripening, where the crystal
model.[35] Depending on which defect is assumed to govern
interior is composed of lattice points (e.g., atom clusters) with
the dynamical evolution of the system, one sometimes speaks
identical energy (e.g., orientation) and the grain boundaries
of pure vertex models, where only the motion of the junctions
are the interfaces between different types of such domains.
is considered, or of pure network models, where only the
As in a real ripening scenario, interface curvature leads to in-
motion of the boundary segments is considered. Recent im-
creased wall energy on the convex side and thus to wall mi-
portant contributions to this field have been reported.[36±38]
gration. The discrete simulation steps in the Potts model, by
which the progress of the system towards thermodynamic
equilibrium takes place, are typically calculated by randomly
4.7. Crystal Plasticity Finite-Element Models
switching lattice sites and weighting the resulting interfacial
energy changes in terms of Metropolis Monte Carlo sampling. Engineering materials are often in polycrystalline form
Recent microstructure simulations using the Potts model where each grain has a different crystallographic orientation,
have been devoted to recrystallization, grain growth, and shape, and volume fraction. The volume distribution of the
scaling aspects.[30±34] different grains in orientation space is referred to as crystallo-
graphic texture. The discrete nature of crystallographic slip
entails an anisotropic plastic response of polycrystals under
4.6. Grain Boundary Vertex Models
load. While the elastic±plastic deformation of single crystals
Topological network and vertex models idealize solid ma- and bicrystals can nowadays be well predicted, plasticity of
terials or soap-like structures as homogeneous continua that polycrystalline matter is not well understood. This is essen-

260 ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5

Raabe/Computational Materials Science on Its Way to Maturity?

tially due to the intricate interaction of the grains during co- conducted a finite-element investigation of the effect of parti-

PROGRESS REPORT
deformation, leading to strong heterogeneity in terms of cle distribution on the uniaxial stress±strain behavior of parti-
strain, stress, and crystal orientation. One major aim of plas- culate-reinforced metal±matrix composites. Other studies
ticity research consequently lies in mapping and predicting were published by Kraft et al.,[47] who presented finite-ele-
this crystallographic anisotropy during large strain plastic de- ment simulations of die pressing and sintering, and by Soppa
formation of polycrystalline matter. In this context, it is essen- et al.[48] who conducted a study of in-plane deformation of
tial to note that the crystals rotate during deformation owing two-phase materials. Important recent studies on the incor-
to the antisymmetry of the displacement gradients created by poration of statistical texture models into finite-element
crystal slip.[39] (Figs. 4,5). formulations were presented by Aretz et al.[49,50] and Peeters
Crystal plasticity finite-element models provide a direct et al.[51,52]
means for updating the anisotropic material state via integra-
tion of the evolution equations for the crystal lattice orienta-
4.8. Texture Component Crystal Plasticity Finite-Element
tion and the critical resolved shear stress. The deformation
Models
behavior of the grains can at each integration point be deter-
mined by a crystal plasticity model, which accounts for plas- The recently introduced texture component crystal plastic-
tic deformation by crystallographic slip and for the rotation ity finite-element method (TCCP-FEM) is based on directly
of the crystal lattice during deformation. In most crystal mapping a small set of discrete and mathematically compact
plasticity finite-element simulations one assumes the stress orientation components onto the integration points of a non-
response at each continuum material point to be potentially linear crystal plasticity finite-element model.[53±55] The spheri-
given by one crystal or by a volume-averaged response of a cal orientation components can be extracted from experimen-
set of grains comprising the respective material point. The ac- tal data, such as pole figures stemming from X-ray or electron
commodation of the plastic portion of the total local deforma- diffraction. The main progress of this approach when com-
tion gradient is reduced to the degrees of freedom given by pared to the conventional crystal plasticity finite-element
the respective crystallographic slip dyads, rotated into the lo- method lies in identifying an efficient way of mapping a
cal coordinate crystal system, prescribed by the orientation(s) statistical and representative crystallographic orientation distri-
mapped at each integration point. The flow stress is typically bution rather than one single grain orientation on an integra-
described in terms of a viscoplastic law using an appropriate tion point.
hardening matrix, which relates strain hardening on one slip Four different techniques are commonly used to reproduce
system to shear on another.[40±45] Other important finite-ele- the orientation distribution function. The first group is re-
ment studies about the micromechanics of microstructures ferred to as direct inversion methods. These methods approx-
were recently published by BorbØly and Biermann[46] who imate textures by directly integrating the fundamental equa-
tion using a large set of pole figure points. The second group
comprises Fourier series expansions, which use spherical har-
monic library functions. The third method is the approxima-
tion of either of the two aforementioned functions by a set of
discrete and identically shaped single orientation points. The
fourth group, referred to as texture component method, com-
prises schemes that mathematically fit given orientation func-
tions or pole figures using spherical Gauss-type standard
functions.
Crystal plasticity finite-element approaches that update
Fig. 4. 3D crystal plasticity finite-element simulations of nanoindentation. The materi- the texture require a discrete representation of the continuous
al is single crystalline copper of different orientation and the indents are made by a
Berkovich indenter. The color coding shows the crystal orientation change around the
orientation distribution function at each integration point.
indents. This circumstance suggests the use of small groups of non-

Fig. 5. 2D crystal plasticity finite-element simulation (left) and experiment (right) of bicrystal deformation. The material
is aluminum and the grain boundary is a 36 tilt boundary. The sample has been plane strain deformed to 30 % engineer-
ing thickness reduction. The color coding in both diagrams shows the distribution of the accumulated von Mises strain.
The experiments have been performed by use of a stereologic photogrammetric displacement measurement.

ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5 261

 Raabe/Computational Materials Science on Its Way to Maturity?

identical discrete texture components and is based on the use pertise of discovering basic mechanisms and microstructural
PROGRESS REPORT

of symmetrical spherical orientation functions. This approach principles behind simulations rather than getting lost in the
provides a small set of compact texture components, which details.
are characterized by simple parameters of physical signifi-
cance (three Euler angles, full width at half maximum, vol-
ume fraction). Using this method, only a few texture compo- 6. Virtual Laboratories for the IndustryÐTools for
nents are required for mapping the complete texture in a Inverse Engineering
precise mathematical fashion. This is conducted in two steps.
First, the discrete main (center) orientation of each texture Computational materials scientists in industry and public
component is assigned in terms of its Euler triple (j1, u, j2) laboratories as well as some research foundations are often
onto each integration point. It is important in this context that not consequent enough when it comes to the introduction of
the use of the Taylor assumption locally allows one to map new and promising simulation approaches into industry. Re-
more than one orientation on an integration point. Second, all searchers sometimes have the tendency that once they have
main orientations of each component initially assigned to the accomplished a simulation task they move on to the next
integration points are rotated in such a way that their result- problem. This attitude has a healthy and a not so healthy side:
ing orientation distribution reproduces the texture function Of course a good researcher should always move on to the
derived before. The reproduction of a texture (e.g., alumi- next serious scientific challenge and try to solve it. Sticking
num, steel, copper, magnesium) usually requires 1±4 texture too much to solved problems means to neglect a huge set of
components plus a background random portion.[53±55] unsolved problems. On the other hand, researches in the
computational materials science community should tend
more to finish a certain initiative by making their results un-
5. Drowned by DataÐHandling and Analyzing derstandable andÐabove allÐfurther exploitable to possible
Simulation Data subsequent less-well trained users.
New simulation methods can be essential tools in the in-
A very good multi-particle simulation nowadays faces the dustry both with respect to the immediate prediction of a pro-
same problem as a good experiment, namely, the handling, cess or a product and in the form of virtual laboratories. While
analysis, and interpretation of the resulting data. Let us take the former aspect is obvious and often used by industry, the
for a moment the position of a quantum-Laplace-daemon and latter aspect has not yet been sufficiently utilized. Virtual la-
assume we can solve the Schrödinger equation for 1023 parti- boratories can be referred to as computer simulation methods
cles over a period of time that covers significant processes in that have been thoroughly tested and proven reliable and they
microstructure. What had we learned at the end? The answer can be used to explore changes in product quality under dif-
to that is: not much more than from an equivalent experiment ferent virtual boundary conditions of production. This corre-
with high lateral and temporal resolution. The major common sponds to the philosophy of inverse engineering where one
task of both operations would be to filter, analyze, and under- starts from a desired material condition and systematically
stand what we simulated/measured. We must not forget that investigates which steps must be taken for obtaining this state
the basic aim of most scientific initiatives consists in obtaining by use of computer simulation. Often such studies show that
a general understanding of principles that govern processes the possible solutions to an optimization problem are not
and states we observe in nature. This means that the mere unique but they allow the engineer to choose among different
mapping and reproduction of 1023 sets of single data microstructures and processes. Virtual laboratories can also
packages (e.g., 1023 times the positions and momentum of all be used to better interpret property changes of products in
particles as a function of time) can only build a quantitative terms of internal rather than in terms of external variables.
bridge to a basic understanding but it cannot replace it. This physically based method of formulating and exploiting
However, the advantages of this quantitative bridge built by constitutive laws in simulations generally provides deeper
the Laplacian super-simulation introduced above are at hand: insight when compared to empirical laws.
First, it would give us a complete and well-documented his-
tory record of all particles over the simulated period, i.e.,
more details than from any experiment. Second, once a simu-
7. Computational Bio-Materials Science
lation procedure is working properly it is often a small effort
to apply it to other cases. Third, simulations have the capabil- The life sciences are currently one of the most dynamical
ity to predict rather than only to describe. Fourth, in simula- fields of research. Computational methods play an ever larger
tions the boundary conditions are usually exactly known (be- role in this area. This section reviews some of the recent chal-
cause they are mathematically imposed) as opposed to lenges, analogies, and applications of materials simulation
experimentation where they are typically not so well known. methods to biomaterials and biological problems. It is likely,
Therefore, my initial statement about the concern of being due to the dynamics in this field, that interdisciplinary re-
drowned by simulation data aims at encouraging the compu- search between computational materials science and the life
tational materials science community to better cultivate an ex- sciences will play an increasing role in the next decade.

262 ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5

Raabe/Computational Materials Science on Its Way to Maturity?

Typical areas where an overlap exists are the histological the user to conduct macromolecular simulations of complete

PROGRESS REPORT
and mechanical fundamentals of biological and artificial soft potential functions that include force fields for all biomolecu-
tissues, the micromechanics of arterial walls and cells, cardiac lar interactions. The atomic scale simulations in this field can
mechanical properties, vascular mechanics, ventricular me- provide insight into the fluctuations and conformational
chanics, cardiac electromechanical properties, and bypass changes of proteins, nucleic acids, protein±nucleic acid, and
mechanics. More areas with an overlap are in arterial regen- protein±lipid combinations. These calculations are still essen-
eration processes, flow properties of blood, rheology of blood tially limited to the nanosecond regime, which prohibits im-
in microvessels, viscoelasticity of collagen, DNA-based mo- portant processes such as large conformational transitions in
lecular design, molecular self-assembly mechanisms, biomo- proteins to be studied. Many techniques, such as coarse-
lecular motors, design and properties of bio-sensors, anti- graining and combination with finite-element methods,
body-antigen binding, and organ micromechanics. The might soon extend the length and time scale of these simula-
understanding and development of synthetic materials for tions. Several overviews of some important aspects of the cur-
the human body, including its cellular and genetic composi- rent state in this field have been published.[56±62]
tion, mechanical properties of muscles, patient-specific expert
and simulation systems for diagnosis and treatment, dynamic
micromechanical simulations of artificial diarthrodial joints,
the mechanics of articular cartilage layers, bone mechanics 8. Scaling, Coarse Graining, and Renormalization
including anisotropy and failure mechanisms, as well as
in Computational Materials Science
macromechanics of the musculoskeleton also pertain to these
overlap areas. When realizing that quantum mechanics is not capable of
Many of these challenges can be tackled by using simula- directly treating 1023 particles, the question arises how macro-
tion methods that have long been used in the computational scopic material properties of microstructured samples can
materials sciences. These methods include for instance elec- nonetheless be recovered from first principles. Numerous
tron dynamics and molecular dynamics approaches, switch- methods have been suggested to tackle such scale problems.
ing models such as the q-state Potts model or cellular automa- They can be classified into two basic groups, namely, multi-
ta, phase-field models, and finite-element models with scale methods and scale-bridging methods. The first set of ap-
incorporation of appropriate micromechanical constitutive proaches (multi-scale) consists in repeatedly including
descriptions. Since this section cannot provide an overview of parameters or rules that were obtained from simulations at a
all the computational methods used in this field, I shall only respective smaller scale in the simulation for the larger scale.
shade some light on the atomic regime. For instance, the interatomic potentials of a material can be
Electron and molecular dynamics methods are nowadays approximated using local density functional theory. This re-
routinely used to simulate solvated proteins, protein±DNA sult can enter molecular dynamics simulations by using it for
complex formation, ligand binding, or the mechanics of small the design of embedded atom or tight-binding potentials. The
proteins. The potentials used in this field depend on the prob- molecular dynamics code could now be used say for the sim-
lem addressed. For instance, mixed quantum mechanical± ulation of a dislocation reaction. Reaction rules and resulting
classical molecular dynamics simulations are used to study force fields of reaction products obtained from such predic-
enzymatic reactions in the context of the full protein. More tions could be part of a subsequent elastic discrete dislocation
empirical potential functions (often termed force field ap- dynamics simulations. The results obtained from this simula-
proximations), which include interactions via molecular tion could be used to derive the elements of a phenomenolo-
bonds and those independent of molecular bonds, are em- gical hardening matrix formulation and so on. Scale bridging
ployed to study molecular nanomechanics addressing aspects methods take a somewhat different approach at scaling. They
such as bond-stretch, bending angles, and torsion angles. As try to identify in phenomenological macroscopic constitutive
in the materials sciences, the more empirical potentials are laws those few key parameters that mainly map the atomic
used as a compromise between accuracy and efficiency. The scale physical nature of the investigated material and try to
potential parameters can be fitted in a way to match experi- skip some of the regimes between the atomic and the macro-
mental results and/or ground-state simulations. In experi- scopic scale.
mental terms the potentials typically are designed to repro- Both approaches suffer from the disadvantage that they do
duce structural data obtained from X-ray crystallography, not follow some basic and general transformation or scaling
dynamic data obtained from spectroscopy and inelastic neu- rules but require instead complete heuristical and empirical
tron scattering, as well as thermodynamic data. guidance. This means that both multi-scale and scale-bridg-
An essential limitation of empirical force field potentials is ing methods must be directed by well-trained intuition and
that they do not allow substantial changes in electronic struc- experience. An underlying and commonly agreed methodolo-
ture, i.e., they cannot model events such as bond formation or gy of extracting meaningful information from different scales
bond failure. Recently improved potential force field func- and implementing them into another does simply not exist.
tions have been developed in the form of hybrid quantum A similar but less arbitrary approach to this scale problem
mechanical±molecular mechanical force fields. This enables might lie in introducing suitable averaging methods that are

ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5 263

 Raabe/Computational Materials Science on Its Way to Maturity?

capable of reformulating forces among lattice defects accu- Although the idea of applying the principles of renormali-
PROGRESS REPORT

rately, in terms of a new model with renormalized interac- zation group theory to microstructures might seem appealing
tions obtained from an appropriate coarse graining algo- at first sight it is essential to underline that there exists no
rithm. Such a method could render some of the multi-scale general unified method of coarse graining microstructure
and scale-bridging efforts more systematic, general, and re- problems. Approaches in this context must therefore also first
producible. The basic idea of coarse graining and renormali- take a heuristic view at scaling and carefully check which ma-
zation group theory consists in identifying a set of transfor- terial property might be useful to assume the position of a
mations (group) that translate characteristic properties of a preserved function when translating the system to the respec-
system from one resolution to another. This procedure is as a tive coarser scale. Another challenge consists in identifying
rule not symmetric (semi-group). The transformations are appropriate methods to describe the in-grain and grain-to-
conducted in such a way that they preserve the fundamental grain behavior of the system clusters obtained by coarse
energetics of the system, such as for instance the value of the graining. On the other hand, coarse graining and renormali-
partition function. zation group theory might offer an elegant opportunity to
To understand the principle of renormalization let us con- eliminate empiricism encountered in some of the scaling ap-
sider an Ising lattice model that is defined by some spatial proaches used in microstructure simulations. Another advan-
distribution of the (Boolean) spins and its corresponding par- tage of applying renormalization group theory to microstruc-
tition function. The Ising model can be reformulated by ap- ture ensembles might be to elucidate universal scaling laws,
plying Kadanoff's original real-space coarse graining or block which are common also to other materials problems.[63±65]
spin approach. This algorithm works by summarizing a small Some important contributions to computational materials
subset of neighboring lattice spins and replacing them by one science along these lines were recently published by Thom-
single new lattice point containing one new single lattice spin. son, Levine, and Shim on percolation and self-organized criti-
The value of the new spin can be derived according to a deci- cal behavior in deforming metals,[66±68] by Rose et al. on some
mation or majority rule. The technique can be applied to the universal features of bonding in metals,[69] and by Nguyen
system in a repeated fashion until the system remains invar- and Ortiz on coarse scaling and renormalization of atomic-
iant under further transformation. The art of rendering these scale binding-energy relations with particular respect to frac-
repeated transformations a meaningful operation lies in the ture mechanics.[70]
way that the value of the partition function or some other
property of significance is preserved throughout the repeated
renormalizations. 9. Artificial Neural Networks in Materials
The effect of successive scale transformations is to generate ScienceÐThe Tale of the Surgeon and the
a flow diagram in the corresponding parameter space. The
Veterinarian
state flow occurs in our present example as a gradual change
of the renormalized coupling constant. Reaching scale invar- Doing materials science in the world of real materials pro-
iance finally is equivalent to transforming the system into a duction (and not in a model or laboratory world) means that
fixed point. These are points in parameter space that corre- we have to deal with fuzzy boundary conditions, unclear pro-
spond to a state where the system is self-similar, i.e., it re- duction parameters, insufficient reproducibility, and all kinds
mains unchanged upon further coarse graining and transfor- of similar ill-defined conditions. Making things even worse
mation. The system properties at fixed points are, therefore, we can add to this list of constraints the influence of markets,
truly fractal. Each fixed point has a surrounding region in labor, or ecology. Everyone who has ever been confronted
parameter space where all initial states finally end at the fixed with the task of mapping materials science that takes place
point upon renormalization, i.e., they are attracted by this for instance in a hot rolling mill producing a quarter of a mil-
point. The surface where such competing areas abut is re- lion tons of steels per month (»1034 atoms) into computational
ferred to as a critical surface. It separates the regions of the materials science will understand the problem (Fig. 6). In this
phase diagram that scale toward different single-component context, where physically based methods (the surgeon) often
limits. fail to quantitatively predict correlations between processes,
The approach outlined in this section is called direct or real microstructures, and products, advanced empirical nonlinear
space renormalization. For states on the critical surface, all learning systems (the veterinarian) in the form of artificial
scales of length coexist. The characteristic length for the sys- neural networks have brought substantial progress.
tem goes to infinity, becoming arbitrarily large with respect Artificial neural networks, which are sometimes also re-
to atomic-scale lengths. For magnetic phase transitions, the ferred to as adaptive methods, are a branch of the broad field
measure of the diverging length scale is the correlation of artificial intelligence. They mimic biological neural net-
length. For percolation, the diverging length scale is set by works although the latter are much more complicated than
the connectivity length of magnetic clusters. The presence of adaptive mathematical models. An artificial neural network
a diverging length scale is what makes it possible to apply to is an information processing method that is inspired by the
percolation the technique of real-space renormalization. way biological nervous systems, such as the brain, process

264 ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5

Raabe/Computational Materials Science on Its Way to Maturity?

PROGRESS REPORT
Fig. 6. Simulations of materials processing, microstructures, and resulting properties
close to production reality require the consideration of fuzzy boundary conditions,
unclear production parameters, and insufficient reproducibility.

information. The key element of this paradigm is the novel

structure of the information processing system. It is com-
posed of a large number of highly interconnected simple pro-
cessing elements (neurons) operating in parallel to solve spe-
cific problems. A processing unit is essentially an equation Fig. 7. Basic set-up and functions of a mathematical and of a real neuron.
that typically has the task of a transfer function, i.e., it re-
ceives weighted signals from other neurons, possibly com-
bines them, transforms them, and outputs a numerical result.
The behavior and performance of neural systems, i.e., how
precisely and effectively they map input data to output data,
is determined by their connection structure, the weights of
the interconnections between neurons, the transfer functions
of the neurons (connection strengths), and the processing per-
formed at computing elements or nodes. Most artificial adap-
tive networks have their neurons structured in layers that
have similar characteristics and execute their transfer func-
tions in synchrony. Also, neural networks have neurons that
accept data (input layer) and neurons that map the results
Fig. 8. A neural network that is used outside of its trained regime is of little predictive
(output layer). Artificial neural networks must be generally use. The present back-propagation network was used to predict the function sin(x) out-
trained by using existing data. This means that they use pre- side of the trained area of 2p.
vious examples to establish (learn) the usually nonlinear rela-
tionships between the input variables and the predicted vari- empirical forecasting, and nonlinear multi-parameter optimi-
ables by repeatedly setting and adjusting the weights until zation. A well-trained neural network can be thought of as a
the predicted variables are very similar to the training vari- highly trained though empirically working expert in the cate-
ables. Once these relationships are established (the neural gory of information it has been given to analyze (Fig. 7). Al-
network is trained), it can be confronted with new input vari- though the programming and mathematics behind neural
ables and it will generate predictions within the limits set by networks are complex, using neural network software can be
the training data. According to these features artificial neural quite simple and the results are often robust provided the
networks resemble the brain in two respects: First, they ac- system was properly trained and the forecast remains strictly
quire knowledge through a learning process. Second, they within the trained area.
use the interneuron connection strengths known as synaptic The classic form of neural methods used in the materials
weights to map and store the knowledge (Figs. 7,8). sciences is the back-propagation network. This system is typi-
Artificial neural networks are applied to problems of con- cally designed in a multi-layer form of the hidden layer with
siderable complexity. Their most important advantage is in appropriate and versatile bias terms and momentum. It is
solving problems that are too complex for conventional com- used to discover and empirically forecast structures in time-
putational technologies or trained human experts. This ap- series, to provide information on the values of phenomenolo-
plies in particular for situations where the complexity of the gical internal variables under complex production situations,
investigated physical phenomena are too large or where the to identify possible relationships between chemical composi-
task consists in deriving meaning from large complicated or tion and materials properties in complex multi-component
imprecise data sets. In general, because of their abstraction systems, or to control and balance complex machines or as-
from the biological brain, neural networks are well suited to sembly lines. The Hopfield adaptive network can be used as
problems that people are good at solving, but for which com- an autoassociative system to memorize digital images. The
puters are not. These problems include pattern recognition, pictures are typically stored by calculating a corresponding

ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5 265

 Raabe/Computational Materials Science on Its Way to Maturity?

weight matrix. Thereafter, starting from an arbitrary config- [10] L. D. Landau, Collected Papers of L. D. Landau (Ed: D. Ter
PROGRESS REPORT

uration, the memory will settle on exactly that stored image, Haar), Gordon and Breach, New York 1965.
which is nearest to the starting configuration. Thus, given an [11] A. G. Khachaturyan, Theory of Structural Transformations
incomplete or corrupted version of a stored image, the net- in Solids, John Wiley, New York 1983.
work is able to recall the corresponding original image. The [12] Y. Wen, Y. Wang, L. Q. Chen, Acta Mater. 2001, 49, 13.
system can be used in any area of materials science where the [13] W. Dreyer, W. H. Müller, Int. J. Solid Struct. 2001, 38,
fast interpretation of complex image information is vital, e.g., 1433.
in the fields of quality control and failure detection in a pro- [14] Y. M. Jin, A. Artemev, A. G. Khachaturyan, Acta Mater.
duction environment.[3] 2001, 49, 2309.
[15] A. M. Mullis, R. F. Cochrane, Acta Mater. 2001, 49, 2205.
[16] Y. H. Wen, Y. Wang, L. Q. Chen, Acta Mater. 2002, 50, 13.
10. Conclusions [17] Multiscale Modeling of Materials (Eds: L. P. Kubin, J. L.
Bassani, K. Cho, H. Gao, R. L. B. Selinger), Materials
In this article I have made an approach to highlight some
Research Society Symposium Proceedings 653, Warren-
important current topics in the field of computational materi-
dale, PA 2001.
als science. Some of the chapters have concisely reviewed the
[18] V. S. Deshpande, A. Needleman, E. Van Der Giessen,
fundamentals of mature microstructure methods with partic-
Acta Mater. 2001, 49, 3189.
ular reference to lattice defect ensemble simulations above
[19] B. D. Wirth, M. J. Caturla, T. Diaz de la Rubia,
the atomic scale. Other chapters have taken a more general
T. Khraishi, H. Zbib, Nucl. Instrum. Methods Phys. Res. B
perspective focussing on questions such as data handling or
2001, 180, 23.
the necessity to identify key mechanisms. What I feel is of
[20] V. S. Deshpande, A. Needleman, E. Van Der Giessen,
particular importance for the further scientific success and re-
Scr. Mater. 2001, 45, 1047.
putation of this exciting field is that we avoid creating pure
[21] J. von Neumann, Collected works of John von Neumann
simulation schools that work exclusively with computer
(Ed: A. W. Burks), Vol. 5, Pergamon Press, New York
models and neglect experimental feedback. Of course, specia-
1963.
lization is an essential issue nowadays but joint proposals
[22] S. Wolfram, Theory and Applications of Cellular Automata,
and mixed research programs should always make it possible
Advanced Series on Complex Systems, selected papers
to find a balance between the realm of simulations and ex-
1983±1986, Vol. 1, World Scientific Publishing Co. Pte.
perimental validation. Another tremendous challenge lies in
Ltd, Singapore 1986.
developing smarter and more general methods to bridge the
[23] D. Raabe, Annual Review of Materials Research 2002, in
scales in multi-scale simulation initiatives in an unambiguous
press.
and reproducible fashion. I am convinced that in this field
[24] R. Ding, Z. X. Guo, Acta Mater. 2001, 49, 3163.
much can be improved by adopting methods developed in
[25] C.-A. Gandin, Adv. Eng. Mater. 2001, 3, 303.
the framework of renormalization group theory.
[26] D. Raabe, Adv. Eng. Mater. 2001, 3, 745.
[27] D. Raabe, Comput. Mater. Sci. 2000, 19, 13.
± [28] P. Matic, A. B. Geltmacher, Comput. Mater. Sci. 2001, 20,
[1] K. Ohno, K. Esfarjani, Y. Kawazoe, Computational Mate- 120.
rials Science. From Ab Initio to Monte Carlo Methods, [29] S. G. Psakhie, Y. Horie, G. P. Ostermeyer, S. Yu. Koros-
Springer Series in Solid-State Sciences, Vol. 129, Spring- telev, A. Yu. Smolin, E. V. Shilko, A. I. Dmitriev, S. Blat-
er-Verlag, Berlin 2000. nik, M. Spegel, S. Zavsek, Theor. Appl. Fract. Mech. 2001,
[2] W. Rosenblueth, N. Wiener, Philos. Sci. 1945, 12, 316. 37, 311.
[3] D. Raabe, Computational Materials Science, Wiley-VCH, [30] D. Raabe, Acta Mater. 2000, 48, 1617.
Weinheim, 1998. [31] E. A. Holm, C. C. Battaile, JOM 2001, 9, 20.
[4] R. W. Cahn, The Coming of Materials Science, Pergamon [32] A. D. Rollett, D. Raabe, Comput. Mater. Sci., 2001, 21, 69.
Press, New York 2001. [33] B. Radhakrishnan, G. Sarma, H. Weiland, P. Baggethun,
[5] P. Haasen, Physikalische Metallkunde, Springer-Verlag, Model. Simul. Mater. Sci. Eng. 2000, 8, 737.
Berlin 1984. [34] V. Tikare, E. A. Holm, D. Fan, L.-Q. Chen, Acta Mater.
[6] Y. U. Wang, Y. M. Jin, A. M. Cuitiæo, A. G. Khachatur- 1998, 47, 363.
yan, Acta Mater. 2001, 49, 1847. [35] C. Maurice, Proc. 21st Risù Symposium on Materials
[7] K. W. Schwarz, X. H. Liu, D. Chidambarrao, Mater. Sci. Science (Ed: N. Hansen), Risù, Roskilde, Denmark 2000,
Eng. A 2001, 309±310, 229. p. 431.
[8] S. Swaminarayan, R. LeSar, Comput. Mater. Sci. 2001, 21, [36] F. Montheillet, J. LØpinoux, D. Weygand, E. Rauch, Adv.
339. Eng. Mater. 2001, 3, 687.
[9] a) L. Onsager, Phys. Rev. 1931, 37, 405. b) L. Onsager, [37] S. Shibata, T. Watanabe, Metall. Mater. Transact. 1999,
Phys. Rev. 1931, 38, 2265. 30A, 621.

266 ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5

Raabe/Computational Materials Science on Its Way to Maturity?

[38] D. Weygand, Y. BrØchet, J. LØpinoux, Acta Mater. 1999, [54] D. Raabe, Z. Zhao, F. Roters, Steel Res. 2001, 72, 421.

PROGRESS REPORT
47, 961. [55] D. Raabe, P. Klose, B. Engl, K.-P. Imlau, F. Friedel,
[39] P. Dawson, Int. J. Solids Struct. 2000, 37, 115. F. Roters, Adv. Eng. Mater. 2002, 4, 169.
[40] D. Raabe, M. Sachtleber, Z. Zhao, F. Roters, S. Zaefferer, [56] Y. C. Fung, Biomechanics, Springer, New York 1993.
Acta Mater. 2001, 49, 3433. [57] S. J. Zhou, D. L. Preston, P. S. Lomdahl, D. M. Beazley,
[41] D. Raabe, Z. Zhao, S.-J. Park, F. Roters, Acta Mater. 2002, Science 1998, 279, 1525.
50, 421. [58] B. L. Holian, P. S. Lomdahl, Science 1998, 280, 2085.
[42] A. Bhattacharyya, E. El-Danaf, S. R. Kalidindi, R. D. [59] T. C. Bruice, K. Kahn, Curr. Opin. Chem. Biol. 2000, 4,
Doherty, Int. J. Plast. 2001, 17, 861. 540.
[43] B. Peeters, E. Hoferlin, P. Van Houtte, E. Aernoudt, Int. [60] H. J. C. Berendsen, S. Hayward, Curr. Opin. Struct. Biol.
J. Plast. 2001, 17, 819. 2000, 10, 165.
[44] R. J. Morrissey, D. L. McDowell, T. Nicholas, Int. [61] F. G. Gianocotti, Curr. Opin. Struct. Biol. 1997, 9, 691.
J. Fatigue 2001, 23, 55. [62] A. Goldbeter, Biochemical Oscillations And Cellular
[45] S. Bugat, J. Besson, A.-F. Gourgues, F. N'Guyen, Rhythms: The Molecular Bases Of Periodic And Chaotic Be-
A. Pineau, Mater. Sci. Eng. A 2001, 317, 32. havior, Cambridge University Press, Cambridge 1996.
[46] A. BorbØly, H. Biermann, Adv. Eng. Mater. 2000, 2, 366. [63] K. G. Wilson, Rev. Mod. Phys. 1983, 55, 583.
[47] T. Kraft, H. Riedel, P. Stingl, F. Wittig, Adv. Eng. Mater. [64] K. Binder (Ed), in Monte Carlo Methods in Statistical Phy-
1999, 2, 207. sics, Springer Topics in Current Physics, Springer, Berlin
[48] E. Soppa, P. Doumalin, P. Binkele, T. Wiesendanger, 1979.
M. Bornert, S. Schmauder, Adv. Eng. Mater. 2001, 3, 261. [65] R. Balian, J. Zinn-Justin (Ed), Methods in Field Theory,
[49] H. Aretz, R. Luce, M. Wolske, R. Kopp, M. Gördeler, North Holland, The Netherlands 1976.
G. Pomana, G. Gottstein, J. Phys. IV 2001, 11, 115. [66] R. Thomson, L. E. Levine, Y. Shim, Mater. Sci. Eng. A
[50] H. Aretz, R. Luce, M. Wolske, R. Kopp, M. Goerdeler, 2001, 309±310, 320.
V. Marx, G. Pomana, G. Gottstein, Model. Simul. Mater. [67] Y. Shim, L. E. Levine, R. Thomson, Mater. Sci. Eng. A
Sci. Eng. 2000, 8, 881. 2001, 309±310, 340.
[51] B. Peeters, M. Seefeldt, S. R. Kalidindi, P. Van Houtte, [68] R. Thomson, L. E. Levine, Y. Shim, Physica A 2000, 283,
E. Aernoudt, Mater. Sci. Eng. A 2001, 319±321, 188. 307.
[52] B. Peeters, E. Hoferlin, P. Van Houtte, E. Aernoudt, Int. [69] J. H. Rose, J. R. Smith, J. Ferrante, Phys. Rev. B 1983, 28,
J. Plast. 2001, 17, 819. 1835.
[53] Z. Zhao, F. Roters, W. Mao, D. Raabe, Adv. Eng. Mater. [70] O. Nguyen, M. Ortiz, J. Mech. Phys. Solids, in press.
2001, 3, 984.

______________________

ADVANCED ENGINEERING MATERIALS 2002, 4, No. 5 267