Progress in Materials Science 73 (2015) 127–158

Contents lists available at ScienceDirect

Progress in Materials Science

journal homepage: www.elsevier.com/locate/pmatsci

Ab initio calculations of mechanical properties:

Methods and applications
J. Pokluda a,b,⇑, M. Černý a,b,c, M. Šob d,c,e, Y. Umeno f
a
Central European Institute of Technology, Brno University of Technology, Technická 3058/10, CZ-616 00 Brno, Czech Republic
b
Faculty of Mechanical Engineering, Brno University of Technology, Technická 2, CZ-616 69 Brno, Czech Republic
c
Institute of Physics of Materials, Academy of Sciences of the Czech Republic, Žižkova 22, CZ-616 62 Brno, Czech Republic
d
Central European Institute of Technology, CEITEC MU, Masaryk University, Kamenice 5, CZ-625 00 Brno, Czech Republic
e
Department of Chemistry, Faculty of Science, Masaryk University, Kotlářská 2, CZ-611 37 Brno, Czech Republic
f
Institute of Industrial Science, The University of Tokyo, 4-6-1 Komaba Meguro-ku, Tokyo 153-8505, Japan

a r t i c l e i n f o a b s t r a c t

Article history: This article attempts to critically review a rather extended field of
Received 30 November 2014 ab initio calculations of mechanical properties of materials. After a
Received in revised form 17 February 2015 brief description of the density functional theory and other approx-
Accepted 24 February 2015
imations utilized in a majority of ab initio calculations, methods for
Available online 16 April 2015
predictions of elastic constants and moduli are presented. A rela-
tively large space is devoted to computations of theoretical
Keywords:
Ab initio methods strength under various loading conditions. First we focus on results
Elastic moduli for perfect crystals and make an overview of advanced approaches
Intrinsic hardness to crystal stability. As case studies, elastic stability conditions
Stability analysis defined according to both the adopted definition of elastic coeffi-
Theoretical strength cients and the kind of applied loading are shown for isotropic ten-
Intrinsic brittleness/ductility sile loading of molybdenum crystal and a model of microscopic
deformation is illustrated for a soft phonon found in the dynamic
stability analysis of isotropic loading of platinum crystal.
Collected values of ideal strength under uniaxial/isotropic tension
and simple shear for selected metallic and covalent crystals are dis-
cussed in terms of their comparison with available experimental
data. Further attention is paid to results of studies on interfaces
and grain boundaries. Applications of computed values of the mod-
uli and the theoretical strength to prediction of intrinsic hardness
and brittle/ductile behavior of crystalline materials and simulation
of pop-in effect in nanoindentation tests are also included. Finally,

⇑ Corresponding author at: Central European Institute of Technology, Brno University of Technology, Technická 3058/10,
CZ-616 00 Brno, Czech Republic.
E-mail address: [email protected] (J. Pokluda).

http://dx.doi.org/10.1016/j.pmatsci.2015.04.001
0079-6425/Ó 2015 Elsevier Ltd. All rights reserved.
128 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

remarks about possible topics for future ab initio studies and chal-
lenges for further development of computational methods are
attached.
Ó 2015 Elsevier Ltd. All rights reserved.

Contents

1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
2. Computational methods. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
3. Mechanical response of crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.1. Elastic coefficients and moduli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
3.2. Theoretical strength of perfect crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
3.2.1. Mechanical stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
3.2.2. Superposition of stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135
3.2.3. Comparison with experiments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
3.3. Properties of interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4. Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.1. Intrinsic hardness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.2. Intrinsic brittleness and ductility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.3. Nanoindentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
4.3.1. FCC metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.3.2. BCC metals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5. Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Appendix A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A.1. Numerical experiment for elastic constants and moduli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
A.2. Model of microscopic deformations related to soft phonons . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155

1. Introduction

The objective of most ab initio (or first-principles) approaches based on fundamental quantum the-
ory is to calculate stationary states of electrons in the electrostatic field of atomic nuclei, i.e., the elec-
tronic structure. The energy of this ground state can then serve as a basis for investigation of
displacements of the nuclei which leads to determination of many macroscopic properties important
in technology. First attempts to develop applicable theories were made in the late 1920s, a few years
after the derivation of the Schrödinger equation. However, a really successful approach providing a
solution of this equation for many-body problem that is relevant to solids was done almost 50 years
later. The density functional theory (DFT) was invented to include correlation effects without using the
very costly wavefunction methods. Nowadays, most ab initio methods used in materials science and
solid state physics are based on the DFT.
The main advantage of ab initio approach is its independence on experimental data. Unlike in
the case of semi-empirical methods, there is no need for calibration or fitting parameters. Thus,
ab initio methods can also be used for calculations of some structural and mechanical characteris-
tics of hypothetical systems, i.e., for prediction of properties of materials that have not yet been
developed. This is, however, not the main aim of this article. Here we focus rather on the studies
on materials behavior under large deformations far from the equilibrium state. Such investigations
can provide a better understanding of micromechanisms of materials failure. This article may be
considered, on one hand, as an up-date and continuation of an earlier review [1] devoted solely
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 129

to the ideal strength calculations but, on the other hand, its aim is much broader. Here we would
like to present a concise overview of ab initio methods and results of their applications in calcula-
tions of mechanical characteristics and properties of solids that were mostly published during the
last 20 years. We also intend to discuss critically the reliability of ab initio methods, assessing also
limitations of individual approaches employed to determine the values of mechanical characteris-
tics of materials studied.
Owing to the large scope of the review, we do not intend to present any extensive interpretation
of all the results based on underlying principles as, e.g., changes in electronic structure or local den-
sity of electrons. The reader can find more details in articles listed in the relevant references. Also,
the article does not have ambitions to deal with a complete set of mechanical characteristics of
materials as known from engineering practice. After all, even the most advanced ab initio
approaches are by far not able to produce such an extended database. Due to large computational
demands, ab initio approaches may be applied, in most cases, to solid systems with a rather perfect
atomic order. Consequently, only so-called intrinsic mechanical properties such as elastic constants
and moduli, theoretical strength or ultimate strain are typical outputs of ab initio calculations; here
knowledge of the total energy of the system is the most important quantity in this sense. Indeed,
these basic mechanical characteristics can be, in a straightforward manner, derived from the
changes of the total energy during the deformation of a perfect solid. Some other mechanical prop-
erties as, e.g., hardness, intrinsic brittleness or ductility can be related to (and, in particular cases,
sufficiently reliably estimated from) the above mentioned characteristics. Therefore, some space is
allocated to these mechanical properties as well. Since the stability analysis based on computation
of elastic moduli and phonon spectra during the crystal deformation is crucial for a correct deter-
mination of theoretical strength, two appendices are also added. The first one is devoted to an
important related problem: differences between elastic coefficients (and elastic moduli) calculated
according to their various definitions for a nonzero stress. The second one illustrates a model of
microscopic deformations that can serve as a tool to verify the identification of soft phonons in
the frame of dynamic stability analysis.

2. Computational methods

Computational methods in materials science are requested to supply certain quantities that can
be used to predict properties of the investigated material and its behavior during a deformation or
other physical (or chemical) processes. Ab initio methods derive the required quantities from the
computed electronic structure. The interaction of electrons is described solely with the help of fun-
damental laws of quantum mechanics. Unfortunately, the Schrödinger equation that is the principal
tool in quantum mechanics cannot be solved exactly for systems of interest in condensed matter
physics or in materials science. For this reason, a series of approximations must be made in order
to get a solution for models that are sufficiently accurate to provide a reliable description of studied
systems and processes. Of course, the solution must be obtained within an acceptable period of
time and with reasonable demands on computational power. Each approximation we perform
moves us toward such a solution but costs a little bit of accuracy or imposes certain limitations
to the solution.
One of the widely used treatments is the Born–Oppenheimer approximation (also called adiabatic
approximation). It relies on the fact that the mass of the nuclei is three to four orders of magnitude
greater than the mass of electrons. Consequently, any change in positions of nuclei is almost immedi-
ately (adiabatically) followed by electrons. Thus, to a good approximation, the nuclei may be consid-
ered as to be fixed in static configurations and the electrons are treated in the field of the stationary
nuclei. This allows us, in this approximation, to separate the wavefunction into a product of nuclear
and electronic terms. Here the nuclei may be considered to be classical particles and their positions
can be taken as parameters that appear in the potential of the electronic part of the Schrödinger equa-
tion. There are, of course, important phenomena in solids which are evoked by mutual interaction
between electrons and nuclei, e.g. electrical resistivity and superconductivity [2]. These cases are
beyond the Born–Oppenheimer approximation which may also fail, for example, in a non-radiative
130 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

electron transition where the electron loses its energy not by emitting a photon but by creating pho-
nons, or in some processes involving light nuclei (surface diffusion, catalytic reactions). In most cases,
however, it is fulfilled to a high degree of accuracy and we may assume its validity throughout this
article. One of the important consequences for calculations of mechanical properties of crystals is lim-
iting their validity to absolute zero temperature.
If the relativistic effects are not essential, electron states in a solid may be described by the non-
relativistic many-electron Schrödinger equation
He;fRa g W ¼ EW; ð1Þ
with H being the electronic Hamiltonian with the nuclei at the positions Ra, W being the many-elec-
tron wavefunction and E being the energy of the system. A very successful approach to the many-elec-
tron problem is the DFT. In 1964, Hohenberg and Kohn [3] formulated two basic theorems establishing
formally the single-particle (single-electron) density q(r) as a variable sufficient for a description of
the ground state of a system of interacting electrons. According to their first theorem, the ground-state
single-particle density q(r) implicitly determines (to within a trivial constant) the external potential
Veff(r) acting on the electron system. The external potential then completes the Hamiltonian H (called
Kohn–Sham Hamiltonian). Once the Hamiltonian is known, all ground-state properties of the system
are implicitly determined. This is a great reduction of the many-electron problem as the single-parti-
cle density is a function of three variables only. All ground-state characteristics of the system in gen-
eral and the total ground-state energy in particular may, therefore, be considered as functionals of
only one function, the single-particle density q(r). According to the second theorem which has the
form of a variational principle, the total energy of the N-electron system, E[q], is minimized by the
ground-state electron density. Thus, the determination of the ground-state electron density and the
total energy becomes extremely simple compared to the problem of solving the 3N-dimensional
Schrödinger equation: we just vary the density q(r), a function of only three variables, regardless of
the number of particles involved, until we find the minimum of E[q]. The DFT has emerged as an extre-
mely powerful tool for analyzing a large variety of many-body systems as diverse as atoms, molecules,
bulk and surfaces of solids, liquids, dense plasmas, nuclear matter and heavy ion systems. It became a
basis of all modern electronic structure calculations and was awarded by the Nobel Prize in chemistry
in 1998.
The most serious problem in DFT is the fact that one part of the Kohn–Sham Hamiltonian
(exchange–correlation functional) is not known exactly and must be approximated. The simplest
approximation developed for this purpose is the local-density approximation (LDA) proposed already
by Kohn and Sham [4]. Here an inhomogeneous system is replaced by a locally homogeneous system
of density q for which the exchange–correlation energy per particle can be calculated. LDA has been
remarkably successful in describing the ground-state properties of a large range of physical systems.
Nevertheless, in systems with strong density gradients, LDA is not a very good approximation. For
example, fundamental band gaps in semiconductors and insulators are typically underestimated by
40%. LDA fails to reproduce the correct ground state in iron and cannot be applied in so-called
heavy-fermion systems. There have been several attempts to improve upon the LDA. In the general-
ized gradient approximation (GGA) the local function of electron density is replaced by a local function
of the density and the magnitude of its gradient. A nice overview of its performance and comparison
with LDA can be found in [5]. The weighted-density approximation (WDA) includes true non-local
information through Coulomb integrals of the density with model exchange–correlation holes. The
self-interaction corrected (SIC) methods try to remedy LDA by explicitly removing the self-interaction
terms in the electrostatic and exchange–correlation energies. The SIC–LDA approach better describes
inner core electrons and is successful in treating materials with f-electrons and transition metal oxi-
des. The so-called GW approximation is formulated by means of a perturbation expansion of the one-
particle Green function and approximates the electron self-energy by the product of an electron prop-
agator (G) and a screened Coulomb interaction (W). The conceptual simplicity of this method is an
advantage, but it is difficult to remove uncontrolled approximations. Finally, let us mention a very
effective scheme called LDA+U. It includes the on-site Coulomb interaction (U) for the description
of correlation effects in localized d- and f-bands and seems to be appropriate e.g. for Mott insulators.
A nice review of various approximations can be found, e.g., in [6,7]. Regrettably, it seems that none of
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 131

the proposed methods for going beyond the LDA leads to systematically improved results in a param-
eter-free manner.
Periodic boundary conditions are particularly useful for description of bulk systems representing
perfect crystals or a periodic repeating of some defects, such as grain boundaries, surfaces, vacancies
and their clusters and precipitates. Here a reasonably large unit cell (supercell) is periodically repeated
in the whole space. Corresponding translational symmetry can remarkably reduce computational
costs if the conditions for using this periodicity are applicable. Although such a model might seem
far from reality since real systems always contain some defects and surfaces, there are many cases
where it can be successfully used. From the qualitative point of view, these models can help us to
describe intrinsic properties of the crystal as well as of many types of defects and to contribute to
understanding the physical basis of deformation processes in solids. Nevertheless, they can also sup-
ply reliable quantitative data. For example, calculated values of equilibrium lattice constants usually
agree with experimental data within one or two percent and computed elastic moduli often fall within
intervals determined by scattered values coming from independent experiments. On the other hand,
periodic boundary conditions require large supercells to treat isolated crystal defects or surfaces in the
computational model of material. In such a case, the computational demands increase rapidly with the
size of the supercell. An example of a problem that is practically inaccessible to such computer sim-
ulations is the dislocation motion under applied loads. It is impossible to study single dislocations
since periodic boundary conditions require introduction of dislocation quadrupoles with dislocations
very closely separated from each other.
In periodic systems we may apply the Bloch theorem which enables us to calculate the electronic
wave functions and electron energies by effectively block-diagonalising the matrix of the Hamiltonian,
with each block (corresponding to a particular wave-vector k) having a manageable size. The size of
each block is the number of selected basis orbitals per atom, multiplied by the number of atoms in
the unit cell. The blocks are the smallest when there is only one atom per unit cell; in this case, some
methods yield a block size as small as 9  9, corresponding to one s- orbital, three p- orbitals and five
d- orbitals. Various methods used in electronic structure calculations may be distinguished according
to the choice of the basis functions. The better we choose them (according to the character of the prob-
lem), the smaller is their number needed for a description of the one-electron wave functions.
Historically, methods such as augmented (APW) or orthogonalized (OPW) plane waves, linear muf-
fin-tin orbitals (LMTO), linear combination of atomic (LCAO), Gaussian (LCGO) and augmented
Slater-type (LASTO) orbitals as well as augmented spherical waves (ASW) were applied. The
Korringa–Kohn–Rostoker (KKR) method proceeds by the use of the Green function of the
Kohn–Sham equation and is also called the Green function (GF) method. The pseudopotential
approach also became very popular. Nowadays, full-potential methods (without any shape approxi-
mation of the crystal potential) are often used to obtain as accurate characteristics of electronic
structure as possible and, for large systems, linear scaling methods, or O(N) methods with computa-
tional and memory requirements which scale linearly with the number of atoms in the system, are
widely used. A detailed description of these approaches may be found in many books and articles,
e.g. [8–11], to name but a few.
After choosing an appropriate basis, the system of equations for electronic structure is solved iter-
atively to self-consistency, i.e. the one-electron density q(r) must generate the effective one-electron
potential Veff(r) which, in turn, must provide the one-electron density q(r). The quality and speed of
the convergence of such calculations is related not only to the choice of a suitable basis, but also to
the sophistication of the iterative process, where as a plausible input atomic-like potentials are usually
employed and input and output potentials or densities are appropriately mixed before starting a new
iteration. Sometimes hundreds of iterations are needed, e.g. in metallic materials with a high peak in
the density of states alternating above and below the Fermi energy, or in most surface problems. If the
electronic structure calculations are well converged with respect of the size of the basis, number of
k-points in the Brillouin zone, etc., then, both in principle and in practice, there is a very good agreement
between the results obtained by methods using localized, non-localized and/or mixed-basis orbitals. This
is illustrated, for example, in the paper [12], which compares ab initio values of cohesive energies, equi-
librium volumes and elastic constants in ground-state crystal structures of 85 elements obtained by the
WIEN2k code (all-electron full-potential code based on augmented plane waves, i.e. employing localized
132 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

basis set) and by the VASP code (full-potential pseudopotential code using a plane-wave basis set).
Similar conclusion was reached also in a study of dispersions and energy positions of the calculated
bands in low-dimensional gold [13]. It turns out that, nowadays, well-converged ab initio electronic
structure calculations by full-potential codes provide very similar results, regardless of the type of
orbitals.

3. Mechanical response of crystals

3.1. Elastic coefficients and moduli

When a crystal is subjected to an external loading (with the stress state described by a stress tensor
rij), its shape and dimensions change. The change of a crystal geometry can be described using the
strain tensor eij. Relation between these two tensors (within the linear elastic limit) can be expressed
by means of the generalized form of Hooke’s law

rij ¼ C ijkl ekl ; ð2Þ

where the quantity Cijkl characterizes the stiffness of the crystal. For this reason, Cijkl are often called
elastic stiffness constants. As Nye [14] pointed out, there are more terms that are used in literature to
denote this quantity (elastic constants, elastic moduli, stress–strain coefficients, etc.) which can be
somewhat confusing because the term ‘‘elastic modulus’’ is sometimes used for an (in a sense) inverse
quantity Sijkl that expresses elastic compliance of the crystal. For the purpose of this review, we
decided to use the term elastic stiffness coefficients or, briefly, elastic coefficients. There is one more
reason for using the term elastic coefficient rather than elastic constant that arises if one evaluates Cijkl
for a stressed crystal. Experimental values of Cijkl are predominantly obtained from a response of orig-
inally unstressed (or unstrained) crystal to small strains. Such measurements conform to a model of
linear elasticity (for which the Hooke’s law was originally formulated). Beyond this small strain limit,
Eq. (2) can be used only with the assumption that values of Cijkl are not constant but depend on the
applied strain and stress. Due to experimental difficulty and due to rather limited interest, elastic coef-
ficients are not measured for finite strains. Ab initio methods, however, make calculations of their val-
ues possible in situations when they are needed.
There are numerous definitions of elastic coefficients as well as numerous ways how to calculate
them. The Hooke’s law (2) is one of them. Other (and more frequent) way is based on expanding
the energy U of the crystal as
X 1 XX
U ¼ U0 þ V ri ei þ V C ij ei ej þ . . . : ð3Þ
i
2 i j

From now on, we use the Voigt notation [14] reducing the number of indices. Definition of the
strain ei can also remarkably affect values of the calculated elastic coefficients for a crystal under
applied loading. Examples of differences caused by using different strain definitions were published
in several papers and reviews (e.g. [15,1]) along with analytical relations between them. Naturally,
all definitions must yield the same values of elastic stiffness coefficients for an unstressed crystal.
Since there is no unique generally accepted definition of elastic coefficients under loading (giving pref-
erence to one of the possible definitions), their differences might seem unimportant. This, however, is
not true in the case of elastic stability analysis when their values determine the onset of instability.
We illustrate their differences in the numerical experiment in Appendix A.
In engineering applications, elastic moduli (such as the Young’s modulus E, the bulk modulus B, or
shear moduli) are often used rather than the elastic coefficients Cij. These moduli can be expressed as a
linear combination of Cij and the relevant expressions can be found in literature (e.g., [14–18]). All
those expressions were, of course, formulated for an unstressed crystal. Results in Appendix A show
that they can be valid without a change also for a crystal under large deformations if the elastic coef-
ficients are calculated from the stress–strain response. Otherwise, modifications of these expressions
by appropriate components of the stress tensor are inevitable. Due to the nature of atomistic calcula-
tions, all the elastic moduli (except, perhaps, for the bulk modulus) are evaluated for single crystals
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 133

and, therefore, for particular crystallographic orientations. In order to also obtain their values for poly-
crystalline solids, several averaging schemes (as those of Voigt, Reuss, Hershey, etc.) have been pro-
posed in the past. A comprehensive discussion of their validity and applicability can be found, e.g.,
in the book [16]. Elastic moduli can be, of course, also calculated directly from the stress–strain
response under corresponding loading conditions as is, e.g., uniaxial loading (to obtain Young’s mod-
ulus) or pure shear.

3.2. Theoretical strength of perfect crystals

Ideal (theoretical) strength of a crystalline solid is the strength of its perfect crystal configuration
under particular loading conditions. Since the stress state is generally characterized by six stress ten-
sor components, then, consequently, theoretically infinite number of theoretical strengths exists for a
particular crystal. Therefore, the term theoretical strength (TS) must be always supplemented with
information not only on the chemical composition and crystal structure of the studied material but
also on the particular loading or deformation conditions. TS has been usually evaluated only for sev-
eral special cases of loading, each defined by a single parameter, e.g., a certain component of the stress
tensor, in particular for uniaxial tensile loading riut, hydrostatic (isotropic) tensile loading riht and
simple or pure shear sis. In these cases, TS can be simply defined as the maximum attainable stress
before reaching the mechanical instability of the crystal studied. Investigations of TS are important
not only to see the gap remaining between the real strength of advanced materials and their ideal
strength. The values of TS are also crucial for a fundamental theory of fracture. Indeed, the stress nec-
essary for the homogeneous nucleation of dislocations in the lattice (matrix) and mostly also from
grain boundaries approaches sis. There are also some serious indications [19–21] that the value of
ideal shear strength determines an extent of the dislocation core in a particular material. The local
stress for nucleation of a cleavage crack must overcome riut and the ratio sis/riht reflects well the ten-
dency of crystal matrix to become brittle or ductile. Moreover, a comparison of mechanical character-
istics of a perfect crystal with those of the related engineering material clearly distinguishes the roles
of perfect matrix and crystal defects in deformation and fracture processes.
The simplest approach to a determination of the theoretical strength assumes that, with increasing
strain, the stress increases up to its maximum which can be considered to be the ideal strength and
most early papers adopted this approach. The ideal strength was first studied using semi-empirical
approaches when describing atomic interactions. Within such schemes the parameters characterizing
interatomic forces are fitted to equilibrium properties of the material studied. However, their transfer-
ability to the state when this material is loaded close to its theoretical strength is not warranted. In
contrast, ab initio electronic structure (ES) calculations (that are subject of interest of this review)
can be performed reliably for variously strained structures and are thus capable to determine the ideal
strength of materials without the resort to doubtful extrapolations. The first paper dealing with the
ideal tensile strength from the first principles was probably that of Esposito et al. [22] who studied
tensile deformation in Cu. However, these authors did not perform relaxations of dimensions of the
loaded crystal in the directions perpendicular to the loading axis. Ideal shear strength was calculated
by Paxton et al. [23] for series of bcc and fcc metals again without any relaxation, similarly to few sub-
sequent studies [24,25]. Other ab initio calculations of properties of the systems far from equilibrium
have also been made, such as exploration of the structural stability, but the results were not employed
to evaluate the strength [26,27]. Probably the first ab initio simulation of a tensile test, including the
relaxation in perpendicular directions to the loading axis, was performed by Price et al. [28] for uni-
axial loading of TiC along the [0 0 1] axis. Results of some later systematic studies were compared with
experiments on whiskers. For example, calculations of the tensile strengths for [0 0 1] and [1 1 1] load-
ing axes in tungsten by Šob et al. [29] gave results already very close to experimental values measured
by Mihkailovskii et al. [30]. On the other hand, tensile strength values computed for Cu [31] were an
order of magnitude higher than available experimental data. Ab initio calculations of theoretical
strength under hydrostatic tension (i.e., negative hydrostatic pressure) appeared for the first time in
the paper of Šandera et al. [32] and several subsequent works [33,34]. Owing to preservation of the
crystal symmetry during this kind of deformation, simpler ab initio approaches might be applied. In
this case, however, comparison with experiment could not be made. The early development of
134 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

ideal-strength calculations has been reviewed in several works [1,35–38] where readers can find fur-
ther details.

3.2.1. Mechanical stability

The assumption that the theoretical tensile strength corresponds to the maximum of stress at the
stress–strain curve (related to so-called spinodal instability) was, however, disputed with the argu-
ment that crystals can fail via other mechanisms. An approach to crystal stability that became very
popular in physics and material science was introduced by Born and Furth [39,40]. Stability of a crystal
lattice was examined using a set of conditions that are expressed in terms of elastic coefficients. These
stability conditions (often called ‘‘elastic stability conditions’’) were derived from the general require-
ment of positive definiteness of the matrix of elastic stiffness coefficients. According to a definition
used for evaluating the elastic coefficients, the elastic stability conditions can differ by values that cor-
respond to certain stress tensor components (in dependence on the particular loading type). Thus, one
must always use a set of stability conditions that is compatible with the definition of computed elastic
coefficients. Several examples of derived stability conditions can be found, e.g., in works of Wallace
[18], Milstein et al. [41,15], Wang et al. [42,43], Zhou and Joos [44], Morris and Krenn [45], Pokluda
et al. [1], etc. One can also find few attempts to compare different sets of stability conditions
[15,1,46] in literature. Another (simplified) approach to the elastic stability of a crystal under uniaxial
loading will be described later. One of the important consequences of these studies was the recogni-
tion of the fact that the classical Polanyi–Orowan formula [47] constitutes a considerable overestima-
tion of the ideal uniaxial tensile strength [48]. On the other hand, the classical Frenkel’s formula [49]
still represents a reasonable rough estimation of the ideal shear strength.
While, for the simple case of a linear lattice (considering interactions between neighboring parti-
cles with central forces), Born [39] derived that its stability with respect to macroscopic deformations
(the elastic stability) implies also the stability with respect to any microscopic deformation, he could
not give a similar proof for a three-dimensional lattice. Moreover, later studies [50–55] confirmed (for
several examples) that the aforementioned implication cannot be transferred to three-dimensional
crystals. Existence of microscopic deformations that can lower the crystal energy is indicated in cal-
culations by so-called soft phonons. The concept of phonons was originally introduced to physics in
order to describe dynamics of the crystal lattice. For this reason, the instability related to soft phonons
is often called a dynamic instability. The term ‘‘soft phonon’’ comes from phase transitions (occurring,
e.g., due to changing temperature or applied pressure). Imaging a phonon as oscillations of certain
atoms in a crystal lattice, the softening means decreasing forces that act on the atoms displaced from
their equilibrium symmetry positions and, consequently, decreasing frequency of the oscillations. In
ab initio studies of crystal stability, the term ‘‘soft phonon’’ denotes such modes that, in fact, do not
correspond to oscillations since the forces acting on atoms displaced from their symmetry (no more
equilibrium) positions do not act toward these positions but in the opposite direction. Thus, the rel-
evant spatially periodic lattice distortion lowers the crystal energy and the energy attributed to the
soft phonon mode is negative. Since the phonon energy is associated with its frequency x, the stability
criterion can be expressed (in the harmonic approximation) by the requirement

½xðqsÞ2 > 0; ð4Þ

for all wavevectors q and all polarizations s (related to optical or acoustic, longitudinal or transverse
modes). In a sense, an elastic instability represents a long wavelength limit (i.e. for vanishing wavevec-
tors q ? 0) of an instability related to soft phonons. The squares of the phonon frequencies are
obtained as eigenvalues of the dynamical matrix [18,56]. According to the dynamical stability condi-
tion (4), the frequencies associated with soft phonons are imaginary numbers and displayed by con-
vention as negative values. Eigenvectors of the dynamical matrix describe directions and relative
magnitudes of the corresponding displacements of ions in the unit cell. Forces necessary for a con-
struction of the dynamical matrix are determined either directly in a supercell approach or perturba-
tively in a reciprocal space using linear response theory.
Let us mention several examples of ab initio predictions of phonon softening for q – 0 that can be
found in literature. Zhao and Harmon [50] predicted soft transverse acoustic phonons in their study of
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 135

dynamic stability of NixAl1x alloys. Clatterbuck et al. [53] reported instabilities related to soft pho-
nons with q close to the Brillouin zone boundary or intermediate q values in Al crystals under uniaxial
tension in the h1 0 0i, h1 1 0i and h1 1 1i directions and under h1 1 2if1 1 1g shear. Dubois et al. [54] stud-
ied mechanical stability of a silicon crystal in order to determine its TS under uniaxial tension, uniaxial
deformation and pure shear. Although most of the instabilities they reported were of an elastic nature,
in the case of h1 1 2if1 1 1g shear, finite soft phonons occurred before the crystal became elastically
unstable. Řehák et al. [55] reported an occurrence of soft phonons in the fcc crystal of Ir under isotro-
pic loading well before the onset of elastic instability. On the other hand, no remarkable reduction of
TS was found in fcc crystals of Al, Pt and Au [55].

3.2.2. Superposition of stresses

Since materials used in the majority of practical applications are usually subjected to a more com-
plex multiaxial loading, several attempts to evaluate TS under such loading conditions have also been
performed. In most of these cases, the loading was considered to be a superposition of two or more
simple loading modes. One example of this approach is a triaxial stress state comprising an axial stress
rax and transverse biaxial stresses rbi (see Fig. 1). The transverse stresses were then considered adjus-
table parameters and their effect on tensile [57–59] and compressive [60–62] strengths was studied
for several crystal symmetries and several orientations of the primary loading axis.
The very first results [57,58,63] suggested that the h1 0 0i tensile TS (determined by the maximum
rax value) increases almost linearly with increasing tensile transverse stresses rbi and decreases with
increasing compressive rbi for most cubic metals. Further studies [64,60] revealed just the opposite
effect of transverse stresses on the compressive TS. Similar trends were found also for the ultimate
strain (related to ideal strength). Such an intrinsic behavior can be easily understood by considering
the following reasoning: the tensile (compressive) biaxial stresses restrain (promote) the Poisson’s
contraction which inhibits (accelerates) displacive transformations that induce the spinodal instabil-
ity. It should be noted that the behavior of polycrystalline metallic materials containing defects is
quite opposite to that of perfect crystals, which is caused by accelerated (decelerated) nucleation,
growth and coalescence of microvoids due to superimposed tensile (compressive) biaxial stresses
[65]. This means that, in engineering metallic materials, the intrinsic properties are completely sup-
pressed by the presence of defects as dislocations and secondary-phase particles. Similar calculations
of tensile TS for other crystallographic directions [59] indicated that the effect of rbi depends on the
particular crystal and the loading direction. Very remarkable deviations from the previously observed
linear trend were found for covalent crystals [61]. The effect of transverse stresses is displayed in Fig. 2
for carbon and silicon in a diamond lattice. The data points plotted using solid symbols were taken
from the results of paper [61]. The open symbols represent the data obtained with a help of a different
procedure: they represent maximum values of rbi calculated as functions of biaxial strain at several
constant levels of rax. They complement the data computed previously and all together demarcate

ax

bi

bi

Fig. 1. An illustration of triaxial stress state that is a superposition of the axial rax and transverse biaxial rbi stresses. These
stress tensor components are represented by the arrows showing directions of forces acting normally to the faces.
136 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

Fig. 2. The envelope of theoretically attainable triaxial stresses in C and Si in diamond lattice.

a region of tensile stability of these crystals. The lines connecting data points for three low-index
directions must intersect or touch at least in one point that corresponds to the value of the isotropic
tensile TS.
Another example of multiaxial loading is a superposition of shear and normal stresses as illustrated
in Fig. 3. Very first studies for Lennard-Jones crystal [66] predicted almost linear increase of the shear
strength with increasing superimposed compressive stress rn (the effect of the transverse stresses rbi
was not considered and the shear planes were rigidly shifted to simulate a shear deformation). The
opposite impact was predicted for tensile stress. These semiempirical results were confirmed from
first principles for most of the studied crystals with the fcc [67,68] and bcc [69,70] structure. The paper
[67] describes results for shear of rigid planes with uncontrolled rbi values. In the subsequent calcu-
lations [68–70], the stress tensor was completely defined by the values of s, rn and rbi = 0. Different
trends for shear strength of fcc copper under normal compression [71,68] were examined (and com-
pared with behavior of other three fcc crystals) in a further study [72] that considered two individually
adjustable stress tensor components: the normal stress rn and in-plane stresses rbi (that stretch the
shear planes). Adjusting both the stresses to the same value forms an isotropic (hydrostatic) superim-
posed stress. In this case, the response of the shear strength of all four studied fcc metals (Ni, Cu, Pt
and Ir) was qualitatively same, exhibiting an increase under compression and a decrease under ten-
sion. Similar behavior under compression was formerly found for bcc Ta [25]. On the other hand,
the shear strength of Si, Ge and SiC crystals with the diamond lattice was found to be a decreasing
function of the compressive stress, while that of diamond (C) shows the opposite trend [73]. This

Fig. 3. Superposition of shear and normal stresses. Shear planes are sheared by the stress s, pulled apart (or pushed together) by
the stress rn and extended (or compressed) by the biaxial (in-plane) stresses rbi.
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 137

peculiar behavior of covalent crystals was explained by a balance between the pairwise repulsion and
the short-range, bond-order dependent attraction in interatomic bonding [74].
The above-described effect of the normal stress on the shear strength was included to a simplified
elastic stability analysis for crystals under uniaxial loading [75]. This analysis is based on an assump-
tion that an instability occurs in a crystal at the instant when the resolved shear stress (in a shear sys-
tem conveniently inclined to the loading axis) reaches the theoretical shear strength. Consequently,
the uniaxial strength riut is related to the moment of exceeding shear strength sis in one of possible
shear systems (ss) according to the relation

sis ðrn ; ssÞ

riut ¼ ;
cos / cos k

where the angles / and k are measured between the loading axis and the shear plane normal and the
shear direction, respectively. The approximate method is simple, time-saving and gives results well
comparable to data available from the standard elastic analysis as it can be seen from case studies
for fcc [75] and bcc [62] metals. Further advantage of the method is the fact that it can be used for
any crystallographic direction with just a very little computational effort.
In fcc crystals, the shear instability reduces the tensile strength in h1 0 0i and h1 1 1i loading direc-
tions for crystals of Al, Cu, Au, Ni, and Pt. Ir was the only example of a crystal that was stable against
shear up to the tensile stress maximum (the spinodal instability). On the other hand no (or small)
reduction of TS was predicted for all these crystals under h0 1 1i loading. On the contrary, results for
bcc crystals suggest that the cleavage stress for the loading along the h0 1 1i direction is greater than
in other directions and shear instabilities related to the f1 1 0g planes are to be expected. Similar shear
instabilities were predicted for the h1 1 1i loading direction in most crystals with exceptions of Mo and
W. In the h1 0 0i direction, shear instability on f1 1 2g planes reduces the strength of V, Nb and Ta,
whereas Fe, Mo and W are stable against shear. This explains the experimental fact that cleavage
on the f1 0 0g planes is a well-known fracture behavior of engineering materials based on the latter
elements. Crystals of C, Si, and Ge with the diamond structure were predicted to be entirely stable
against shear [61].
The existence of elastic instability has also an interesting consequence for the extent of dislocation
core regions in fcc, bcc and hcp metals. The core radius can be generally defined as that radius for
which the stress field of the dislocation that is predicted by elasticity theory equals the ideal shear
strength of the material [19,20]. Consequently, the core radius can be expressed as a function of elastic
constants and sis and, in particular, it depends on C11  C12 and C44. This means that the core radius
diverges as C11  C12 ? 0 and C44 ? 0, two of the elastic stability limits of crystals. Therefore, there
are three principal possibilities how to extend the region of dislocation cores in metals: (i) modifica-
tion of the elastic constants via alloying or other methods to let bcc, fcc or hcp structures approach the
elastic stability limits; (ii) to apply an external stress and (iii) to increase the internal stress by increas-
ing the dislocation density [21].

3.2.3. Comparison with experiments

Values of the ideal isotropic (hydrostatic) strength riht, the ideal tensile strength under uniaxial
loading riut and the ideal shear strength sis are collected in Table 1. Since there are plenty of calculated
strength values in literature, we restrict our list only to the values calculated using ab initio methods
within models with a well-defined stress tensor (i.e., optimization procedures enabled, e.g., relaxation
of transverse stresses during simulated tensile loading or relaxation of normal and plane stresses dur-
ing a simulation of pure shear) for crystals with the fcc, bcc and diamond structure. Even after such
restrictions, the list is by no means complete. All the listed values of sis for fcc crystals and for crystals
with diamond lattice were computed for a shear on the f1 1 1g planes. Therefore, the table contains
only information about the shear direction. In the case of bcc crystals, the calculations were performed
for shear along the h1 1 1i direction and Table 1 contains the complementary information on the shear
plane.
It is of a particular interest to compare the results of available experimental tensile tests performed
on whiskers with those simulated theoretically. A long-term experience with testing of whiskers
138 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

Table 1
Calculated values (in GPa) of the ideal tensile strength under isotropic (hydrostatic) riht and uniaxial riut loading and the ideal
shear strength sis. Boldface font marks strength values that correspond to elastic instabilities occurring prior to reaching the
maximum stress at the stress–strain curve and the star added denotes instabilities related to soft phonons with finite wave vectors.

riht riut sis

C dia 88.5a,b,c h1 0 0i 225b,d, 226c h1 1 0i 113.3a
h1 1 0i 130d, 127c h1 1 2i 94.0e, 95f
h1 1 1i 90d, 95f, 93.1c
Si dia 26.4g, 15.4a, 15.5b,c h1 0 0i 26.3b, 27.5c, 26.4h h1 1 0i 9.62a, 8.0h
h1 1 0i 17.9c, 17.0h h1 1 2i 8.1e, 6.5f
h1 1 1i 24.8i, 23f, 21.0h, 21.3c
Ge dia 11.1b,c h1 0 0i 16.8b, 16.4c h1 1 2i 5.2e, 4.5f
h1 1 0i 11.8c
h1 1 1i 14f, 14.6c
Al fcc 11.2a, 11.0j, 12.2o, 12.0⁄o h1 0 0i 13.1k, 12.6l, 12.1l, 12.9p, 9.20⁄p, h1 1 2i 1.85m, 3.12n, 2.84a, 3.33p,
11.6q, 9.0n 3.16⁄p
h1 1 0i 4.92p, 4.89⁄p, 4.5n
h1 1 1i 11r, 11.1l, 11.3p, 8.8n, 8.95⁄p
Ni fcc 29.2a, 28.9b, 27.4t h1 0 0i 35.2b, 36.1s, 18.3n h1 1 2i 5.05a, 5.64n, 5.1s
h1 1 0i 9.5u, 10.5s, 9.1n h1 1 0i 6.4s
h1 1 1i 33.1u, 34.1s, 15.4n
Cu fcc 26.0g, 29.1v, 20.4a, 19.8b, 20.2j h1 0 0i 33w, 24.1b, 9.4x, 22.5q, 9.3n h1 1 2i 2.65m, 2.16a, 3.01n
h1 1 0i 31w, 5.2u, 4.6n
h1 1 1i 29w, 20.3u, 7.5n
Ag fcc 20.4v, 17.6a, 11.4j h1 0 0i 12.3q h1 1 2i 1.65a
g b o b q
Ir fcc 57.7 , 40.1 , 47.5 , 43 ⁄o
h1 0 0i 44.5 , 48.3 h1 1 2i 17.1n
h1 1 0i 26.3u
h1 1 1i 43.4u
Pt fcc 43.2v, 39.6b, 40.1o, 36⁄o h1 0 0i 34.1b, 32.1q, 7.0n h1 1 2i 2.75n
h1 1 0i 3.5u, 3.5n
h1 1 1i 29.9u, 6.1n
Au fcc 26.2v, 23.5a, 23.2b, 23.1o, 19.2z, h1 0 0i 18.6b, 3.6n, 4.2y h1 1 2i 0.85a, 1.66n
18.8⁄o,
h1 1 0i 2.0u, 1.8n
h1 1 1i 13.8u, 3.2n
V bcc 40.2v, 38.3aa, 32.7b h1 0 0i 19.8b, 17.8ab, 17.4ac, 16.1ae, f1 1 0g 6.54ab, 6.3ad,
11.5ab,ac, 11.6ae
h1 1 0i 33.3ad, 37.6ae, 15.0ac f1 1 2g 5.6af, 5.5ab
h1 1 1i 31.4ac, 26.7ae, 22.8ac
Cr bcc 47.4aa, 21.0t
Fe bcc 28.5a, 27.7b, 26.7t h1 0 0i14.2ag, 12.7ah, 12.4b, 12.6ai, 10.6ac f1 1 0g 8.14a, 7.8ai, 7.5ad
h1 1 0i 33.0u, 32.3ad, 16.7ac f1 1 2g 7.51a, 7.2ai, 6.0af
h1 1 1i 27.3ah, 27.7u, 27.5ac, 19.3ac f1 2 3g 7.57a
Nb bcc 37.6v, 34.1aa, 31.6b h1 0 0i19.0b, 13.1aj, 17.2ab, 12.5ab, 17.1ac, f1 1 0g 7.6aj, 7.8ab, 7.1ad
12.7ac
h1 1 0i 27.6ad, 17.1ac f1 1 2g 6.4aj, 6.1af, 6.0ab
h1 1 1i 27.3ac, 19.8ac
Mo bcc 50.4v, 42.2aa, 43.2a, 42.9b h1 0 0i 28.3b, 28.8aj, 27.9ac, 26.7v f1 1 0g 15.2a, 16.1aj, 15.1ad
h1 1 0i 40.7u,ad, 32.6ac f1 1 2g14.8a, 15.0af, 15.8aj
h1 1 1i 28.4u, 29.9ac f1 2 3g14.9a
Ta bcc 42.3v, 41.3aa h1 0 0i 13.4ab, 10.3ac f1 1 0g7.1ab, 6.5ad
h1 1 0i 33.9ad, 13.4ac f1 1 2g 4.9af, 6.5ab
h1 1 1i 25.6ac, 17.1ac
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 139

Table 1 (continued)

riht riut sis

g v aa a b ak b,al ac
W bcc 57.4 , 57.4 , 53.1 , 50.2 , 50.7 h1 0 0i 29.5 , 28.9 , 27.7 f1 1 0g 18.2ak, 17.5a, 17.1ad
h1 1 0i 49.4u, 54.3al, 49.2ad, 35.9ac f1 1 2g18.1ak, 17.4a, 16.8af
h1 1 1i 37.5u, 40.1al, 39.0ac f1 2 3g 17.6ak, 17.3a
a
Ref. [76].
b
Ref. [58].
c
Ref. [61].
d
Ref. [77].
e
Ref. [73].
f
Ref. [78].
g
Ref. [32].
h
Ref. [54].
i
Ref. [79].
j
Ref. [80].
k
Ref. [81].
l
Ref. [82].
m
Ref. [83].
n
Ref. [75].
o
Ref. [55].
p
Ref. [53].
q
Ref. [64].
r
Ref. [84].
s
Ref. [85].
t
Ref. [86].
u
Ref. [59].
v
Ref. [33].
w
Ref. [31].
x
Ref. [87].
y
Ref. [88].
z
Ref. [89].
aa
Ref. [34].
ab
Ref. [90].
ac
Ref. [62].
ad
Ref. [70].
ae
Ref. [91].
af
Ref. [69].
ag
Ref. [92].
ah
Ref. [93].
ai
Ref. [94].
aj
Ref. [95].
ak
Ref. [96].
al
Ref. [29].

reveals that, in accord with the statistical weakest link concept, a reduction of whisker volume leads to
a higher strength (e.g. [97–99]). Therefore, only experimental data obtained with micro and especially
nanowhiskers (or nanoneedles) will be mentioned here. Since there is, naturally, always a rather high
scatter of experimental data, only the highest achievable strengths should be taken into account. For
example, Brenner [97,100] measured the highest values of tensile strength of copper microwhiskers in
the range of 2–3 GPa for loading along the h1 1 0i direction. This is about one half of the ideal strength
of 4.6 GPa calculated including the elastic stability analysis (see Table 1). For Cu nanowhiskers, how-
ever, Richter et al. [99] reported the highest values in between 5 and 7 GPa which is even higher than
the value of 5.2 GPa evaluated as the maximum of the computed stress–strain dependence. Shpak
et al. [101] and Kotrechko et al. [102] reported values of 20 GPa for molybdenum and 23 GPa for tung-
sten nanoneedles loaded in the h1 1 0i direction. These values represent about 2/3 of theoretically pre-
dicted ones (see Table 1). For tungsten nanowhiskers, Mikhailovskij et al. [103] received even the
value of 31 GPa which is very close to the theoretical one. There is no surprise that the smallest values
of about 1 GPa of theoretically predicted strengths were found for alkali metals (isotropic tension, [32–
34]) and the highest ones were reported for carbon structures as diamond (3D), graphene (2D sheets
140 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

and nanotubes) and carbyne (1D string). However, it is interesting to note that the absolutely highest
strength is exhibited by the carbyne string followed by graphene and diamond only. Indeed, the
strength of carbyne strings related to a critical pondermotive force measured in the field-ion micro-
scope was as high as 270 GPa [104] though still significantly lower than the predicted theoretical
strength of 410 GPa. Strength values in the range of 90–130 GPa were measured for graphene by
nanoindentation in the atomic force microscope [105,102] which is well comparable with the theoret-
ical value of 110 GPa for zigzag and 121 GPa for armchair nanotubes, respectively [106]. There are two
main reasons why the transition 3D ? 2D ? 1D corresponds to increasing tensile strength [107,102]:
(i) the decreasing coordination number of atoms (increasing portion of free-surface area) causes a
transfer of electrons from the surface to the bulk which strengthens the inner bonds; (ii) increasing
part of purely stretched directional covalent bonds makes the carbyne structure to be most stiff (many
of diamond bonds and some of graphene ones are bended).
When comparing the theoretical results obtained for zero Kelvin with experimental values deter-
mined mostly at room temperature, a legitimate question arises on how important is this temperature
difference. Unfortunately, there is no direct first-principles tool for such estimations yet: the most
physically justified concept based on ab initio electron–ion dynamics is still under development
[108]. Thus, to the best of our knowledge, the only ab initio result reported by Shang et al. [109]
was achieved in an indirect way. The authors estimated the thermally-induced normal stress rn from
phonon calculations of thermal expansion to check a corresponding reduction of the value of ideal
shear strength sis according to the above mentioned function sis(rn). This indicated only a nearly
10% decrease of sis within the temperature range h0; 1000i K. Nearly the same result was previously
obtained when considering the phonon fluctuations of Einstein harmonic oscillators [48]. However,
assessments based on other semi-empirical approaches mostly show a much steeper reduction of
sis or riut. For example, the classical formulas [49,47] indicate that sis and riut should follow the tem-
perature change of the shear and the Young’s moduli, respectively. This means an approximate 10%
change within the range h0; 300i which is nearly in agreement with an analysis of Tschopp and
McDowell [110] based on molecular dynamics. Therefore, it seems that the temperature raise from
0 K to 300 K can cause a 10% decrease in sis, at the most.
In general, one can say that the experimental data on riut obtained in defect-free materials already
approach the theoretical ones provided that a proper stability analysis is taken into account.
Experimental values of ideal shear strength can be obtained from nanoindentation tests of single crys-
tals and even polycrystals by analyzing so called pop-in effects as discussed below in more detail. The
experimental values obtained in this way are also in a good agreement with the calculated ones.
All values in Table 1 and the related discussion correspond to single crystals of pure elements.
Strength of alloys and compounds was, of course, also studied from first principles. In most cases,
however, simple systems with stoichiometric arrangements of atoms were subject of interest owing
to the possibility of their modelling using relatively small computational cells within periodic bound-
ary conditions. As relevant examples one can mention, e.g., studies of borides [111–113], carbides
[114,115,28] and nitrides [114,116] of metals or other intermetallics [35,76]. Major disadvantage of
this approach is then, for example, a limited number of particular concentrations of individual alloying
elements that can be simulated using small supercells. If one requires to study the effects of small con-
centrations of some element on the alloy strength (or other properties), increasing computational
demands due to increasing size of necessary supercells reveal to be a limiting factor of this approach.
The problem can be surmounted by using some of few alternative approaches such as the coherent
potential approximation (CPA) or the concept of special quasi-random structures (SQSs).
Performance of these two methods in predicting elastic properties of disordered alloys is comparably
good [117]. Applications of such methods for obtaining the strength are rare and we can only mention
calculations of strength of Mo–V random solid solutions using the CPA in connection with the exact
muffin-tin orbitals method [91].

3.3. Properties of interfaces

First-principles studies of interfaces of various types are relatively computationally demanding

owing to the necessity of using large supercells. Size of the supercells must guarantee that the
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 141

interfaces that appear repeatedly (due to periodic boundary conditions) in the model do not interact
with each other. On the other hand, such studies are motivated by increasing interest of scientists in
understanding physical phenomena occurring at interfaces or in thin films and by potential industrial
applications of various layered systems as, e.g., coatings of cutting tools, recording media, electronic
devices, etc. Since this review is focused primarily on mechanical properties, we do not intend to men-
tion studies focused on electronic or optical properties and we only limit our attention to investiga-
tions of elasticity and cleavage of interfaces and grain boundaries. First ab initio studies on
interfaces started to appear in early 1990s. Freeman and Wu [118] studied clean surfaces and inter-
faces and observed a magnetic enhancement at interfaces with inert substrates and magnetic suppres-
sion caused by strong interaction with nonmagnetic transition metals. We mention these results here
since magnetism can play an important role in stabilizing a crystal structure and, therefore, it can also
be a decisive factor for stability and mechanical response of the interface. Surface and interface ener-
gies were computed, e.g., by Schonberger et al. [119] for Ti/MgO and Ag/MgO, Hartford [120] for Fe/VN
or Siegel et al. [116] for Al/VC and Al/VN interfaces. Nevertheless, an important quantity closely
related to mechanical properties is the cleavage energy and the cleavage stress. Such calculations
can be found, e.g., in the early papers as that by Price et al. [28] or in later works of Lazar et al.
[121,122]. Mechanical stability and strength of various interfaces were also calculated. For example,
Zhang et al. [123,124] calculated stress–strain curves of interfaces in the TiN–SiNx nanocomposites
under tension and shear. These systems attracted wide attention of scientists due to their industrial
applications. Their dynamical stability was investigated by Marten et al. [125] with the use of com-
puted phonon spectra. More recently, Ivashchenko et al. [126] performed quantum molecular dynam-
ics studies of TiN/SiNx/TiN interfaces at the temperatures of 300 K and 1400 K.
Grain boundary (GB) is an extended planar defect that represents an interface between two grains
of the same chemical composition. Since the majority of engineering materials are polycrystalline,
grain boundaries significantly affect their macroscopic behavior. Mechanical properties of GB as,
e.g., hardness or fracture toughness can be dramatically changed by a presence of defects or impuri-
ties. For this reason, GBs attracted attention of many materials scientists which resulted in a new con-
cept in materials science and engineering, called Grain Boundary Engineering. The concept was first
proposed by Watanabe in the early 1980s [127] and is considered to be a viable tool to improve
the mechanical properties of engineering materials such as strength, ductility, creep, and corrosion
resistance through a systematic modification of grain boundary distribution. Ab initio studies of GB
that have been performed to date were mostly devoted to a phenomenon of GB embrittlement due
to the presence of segregated impurities. The earliest studies were based on modelling of small clus-
ters that substituted the GB region in the vicinity of the impurity atom. This approach was adopted,
e.g., by Painter and Averill [128] for estimating the effect of boron and sulfur on binding in Ni and,
thus, on GB cohesion. Ab initio modelling of some specific GBs by using supercells containing the grain
boundary including several atomic planes (representing the bulk structure inside grains) became fea-
sible owing to further development of computational resources. It is not surprising that the systems
that attracted most attention usually were GBs in materials with a wide industrial exploitation such
as Fe, Ni and Al. Since steels belong to the most technologically important materials, GBs in Fe became
one of the first subjects of research (despite increased computational demands due to its magnetic
ordering). For example, Wu et al. [129–131] studied an effect of segregated phosphorus and boron
on R3ð1 1 1Þ GB in Fe. They reported a B–Fe hybridization that permits covalent bonding normal to
the boundary (contributing to cohesion enhancement) and nonhybridized interaction of Fe and P with
the opposite effect on GB cohesion [130]. Their conclusions were based on calculated energies of
ground-state configurations of GB with and without impurities. Similar approach was used in several
later studies as, e.g., in their further study on GB in Fe with segregated Mo and Pd [132] or in the paper
of Lu et al. [133] on R5ð2 1 0Þ tilt GB in Ni3Al with hydrogen impurity. Bismuth-induced embrittlement
of GB in Ni and Cu was studied by Kang et al. [134] who found the Bi bilayer to be the most stable
impurity phase, in agreement with former study by Luo et al. [135].
Let us note that theoretical studies predict enhancement of magnetic moments at clean GBs and
surfaces in ferromagnetic metals (see, e.g., Refs. [118,136,132,137,138] and references therein). On
the other hand, a strong reduction of magnetic moments at R5ð2 1 0Þ GB and (2 1 0) free surface (FS)
in fcc ferromagnetic nickel decorated by phosphorus or boron has been reported in calculations by
142 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

Geng and co-workers [137]. Complete killing of magnetization in the GB layer and at the FS by segre-
gated phosphorus was regarded as an exception among the sp- impurities studied, but it was con-
firmed nearly as a rule by Všianská and Šob [139,140] who found that interstitially segregated Si, P,
S, Ge, As, and Se produce magnetically dead layers at the R5ð2 1 0Þ GB in fcc ferromagnetic nickel
and all 12 sp- impurities studied (i.e., Al, Si, P, S, Ga, Ge, As, Se, In, Sn, Sb, and Te) nearly kill or substan-
tially reduce the magnetic moments at the (2 1 0) FS. Further, it was found that interstitially segregated
Si is a GB cohesion enhancer, substitutionally segregated Al and interstitially segregated P exhibit
none or minimum strengthening effect and interstitially segregated S, Ge, As, Se and substitutionally
segregated Ga, In, Sn, Sb and Te are GB embrittlers in nickel. Most experimental results on GB segre-
gation in metals may be found in the book [141]. When analyzing available theoretical values and
experimental data on GB segregation, it was found that for impurities with a low solubility (lower than
1 at%) the calculated energies of grain boundary segregation are unreliable [142]. In this case, the cal-
culated energy of a solute in the bulk does not correspond to equilibrium state and cannot be used
when evaluating segregation energies. On the other hand, embrittling or strengthening effect of
solutes can still be deduced quite reliably, because the energy of the solute atom located in the bulk
far away from the interface is cancelled out. A more detailed comparison of calculated and experimen-
tal values of segregation energy of impurities at GBs in iron may be found in [143]. Recently, we can
see quite a lot of scientific activity in the field of impurity segregation at GBs and its effects on prop-
erties of materials [144–148]. Because of lack of experimental data in many cases, theoretical findings
often have a character of a theoretical prediction, as it is, e.g., in the case of systematic studies
[139,140,149,148]. As the segregation phenomena are important for technological applications of
the nickel-based materials, we hope that the calculated results may motivate experimentalists to per-
form more measurements on impurity segregation in technologically important materials.
Computational tensile tests became the next step in the effort to explain elementary mechanisms
playing role in the process of intergranular fracture, because they are related not only to the initial
(ground-state) electronic structure but also to its changes with increasing strain. Such tensile tests were
reported, e.g., for clean R5 [81] and R9 [150] GBs in fcc Al. Tensile tests of R5ð2 1 0Þ GB in fcc Ni with sev-
eral sulfur concentrations were performed in order to explain its S induced embrittlement [151,152]. The
model of R3 GB in Fe became a subject of further ab initio investigation of bond mobility mechanism in
GB embrittlement [153]. Several models have been employed for such computational tensile tests. Most
of the models conformed to the uniaxial deformation, i.e., to the model neglecting the Poisson’s contrac-
tion during incremental elongation of the computational supercell (in a direction perpendicular to the
interface). The simplest model is based on linear scaling of the atomic positions at each strain increment.
This model was adopted, e.g., by Kitamura and Umeno [81] in calculations of R5 tilt GB in Al that verified
validity of the effective medium theory for Al. For calculations of the cleavage energy and the cleavage
stress, a different model considering rigid grain displacement was employed: the computational cell
is elongated by introducing an increasing gap at the interface (between two determined fracture sur-
faces). Examples of applications of this model can be found in works by Lu et al. [150], Yamaguchi
et al. [151] or Sanyal et al. [154] that were devoted to study of GBs or in works by Price et al. [28] or
Lazar et al. [121,122] focused on cleavage of pristine crystals under loading mode I and on VN/TiN mul-
tilayers. For first-principles simulations of tensile test on crystals with GBs, physically more relevant
models considered also optimizations of atomic positions at each strain increment. This approach can
be found already in the early papers on this topic by Kohyama [155] and Ochs et al. [156] and in many
later studies [157–160,153,152]. Some of these studies employed the homogeneous extension of the
supercell (with linear scaling of atomic positions) [155,157–159,153], some others preferred the rigid
grain displacement [156] as the deformation step preceding the subsequent atomic optimizations.
Some studies also compared these two approaches (e.g., [159,160]) and their results revealed that the
optimization process unifies the computed energy-strain dependences for strains almost up to the onset
of instability. Near the critical strain, however, results of the optimizing process start to differ which is
very likely a consequence of the fact that the localized strain introduced by the rigid grain displacement
disrupts the atomic bonds more seriously than the linear scaling of atomic positions (optimized positions
from the preceding deformation step). For the sake of keeping computational demands reasonably low,
effect of Poisson’s contraction on results of the computational tensile test has been considered very
rarely. There are, however, a few good reasons why to take it into account. For example, Tian et al.
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 143

[159] reported a remarkable reduction of the strain energy and consequent reduction of tensile strength
when the Poisson’s contraction was included in their computational tensile test. Yuasa and Mabuchi
[153] calculated tensile strength of Fe R3ð1 1 1Þ tilt GB with and without segregated phosphorus. Their
results of 25 GPa for a clean GB and 16 GPa for P-segregated GB seem to be very high since the tensile
strength of a perfect Fe crystal in the h1 1 1i direction was calculated to be about 19 GPa (see Table 1).
Similarly, values obtained by Yamaguchi et al. [151] for S-segregated Ni R5ð2 1 0Þ GB of 30 GPa for the
bulk fcc Ni, 26 GPa for a clean GB and 22 GPa for GB with small amount of sulfur are very likely much
higher than h2 1 0i tensile strength of a bulk Ni (the value for h2 1 0i loading is not included in Table 1
but one can expect it to be slightly higher than that for the h1 1 0i direction).
As mentioned at the beginning of this section, periodic boundary conditions impose severe limita-
tions to the choice of a particular interface that can be studied from first principles. There are, how-
ever, coherent interfaces that can be investigated with the help of such methods without significant
difficulties such as twin boundaries [161–166]. These interfaces are worth mentioning since twinning
is an important deformation mechanism (that can take part in deformation processes particularly at
low temperatures) and, therefore, studies of their structure and energy yield information useful for
prediction of mechanical properties.

4. Applications

4.1. Intrinsic hardness

Ab initio calculated values of elastic moduli also seem to provide some insight into the intrinsic
hardness (IH) of perfect covalent crystals. In metallic and ionic materials, the bonding is delocalized
and hardness is determined by extrinsic factors as precipitates, impurities, grain boundaries, etc. In
covalent materials, however, the bonding is localized in electron spin pairs, thus hardness is an intrin-
sic property. Therefore, in order to design new superhard materials, the materials scientists have
aimed at finding materials with high values of bulk and shear moduli which reflect the high strength
of atomic bonds (e.g. [167,168]). Some empirical and semi-empirical correlations were proposed as,
m
for example, the formula HV ¼ Ck Gn developed by Chen et al. [169] where HV is the Vickers hardness,
C, m and n are empirical coefficients and k is the Pugh’s modulus ratio G/B. However, such depen-
dences exhibit a big scatter and, as a rule, they are even not monotonic and unequivocal. A review
of semi-empirical models can be, for example, found in [170]. Some authors employed ab initio calcu-
lations to determine elastic or cleavage characteristics considered in semi-empirical correlations with
IH. One of such concepts [171] deals with the cleavage energy 2cs appearing in the classical formulas
rit ¼ ðEcs =aÞ1=2 of Griffith [172] and Orowan and Polanyi [47,173] for the tensile stress necessary to
create an unstable crack in a perfect crystal. Here cs is the surface energy, E is the Young’s modulus
and a is the inter-planar spacing of the planes perpendicular to the tensile axis in the unstressed state.
First-principles calculations of the surface energies for a number of covalent crystals indicate a power
law relationship between cs and the measured hardness HV.
Probably the first work that established a direct link between IH and ab initio calculations was that
of Gao et al. [174] who employed the Gilman’s theory of plastic deformation in covalent crystals
[168,175]. In these crystals, the dislocation energy strongly depends on its position and, therefore,
their plastic shearing demands a break (and remake) of electron pairs. The break means that two elec-
trons become excited from the valence- to the conduction band so the activation energy for plastic
glide is twice the band gap, Eg. Thus, the bond strength is determined by Eg and the hardness of overly
covalent crystals could be expressed as
2:5
IH ðGPaÞ ¼ AN a Eg ¼ 350½ðNe Þ2=3 expð1:191f i Þ=db ; ð5Þ

where A is a coefficient of proportionality, Na is the covalent bond number per unit area, Ne is the elec-
tron density expressed as the number of valence electrons per cubic angstrom, db is the bond length in
angstroms and fi is the ionicity of the chemical bond. Eq. (5) shows that, for very hard materials, the
following conditions should be met: (i) a high density of valence electrons (bond density), (ii) a short
bond length and (iii) high portion of covalent bonding. Using this equation, the trend of IH could be
144 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

satisfactorily predicted for about 30 covalent and polar covalent crystals. Several years later, Gao [176]
and Šimůnek [177,178] independently published first-principles calculations of the bond electron
densities that reflect the bond strength. In the first work, the overlap population analysis according
to Mulliken [179] was employed and aided by ab initio calculations based on Vanderbilt pseudopoten-
tials and LDA. This analysis utilizes a technique for the projection of plane-wave states onto a localized
basic set to calculate atomic charges and bond populations. In the latter works, the bond strength was
determined by the valence electron numbers of the atoms in the bond, the crystal valence electron
density, the number of bonds and the bond length between the atoms. The LDA plane-wave all-elec-
tron pseudopotential method was utilized to calculate these quantities [177]. Both approaches
resulted in a good agreement between theory and experiment for IH values in the range of 2 orders
of magnitude. Moreover, they showed that IH is controlled by the weakest bonds transversely oriented
to the direction of indentation. Another ab initio aided approach was proposed by Lazar et al. [111].
The authors assumed that the IH of an intrinsically brittle material at the onset of indentation is
related to the minimum critical stress for breaking the bonds between planes of orientation [h k l]min,
for which the critical stress has the lowest value. Following this concept, IH is equal to the lowest of all
cleavage stresses considered to be a function of the orientation [h k l] and possible different cleavage
planes for the same direction. Such calculated hardness values for some diboride crystals were found
to agree well with experimental data obtained for low applied indentation loads.
It should be emphasized, however, that the validity of all these models of IH is naturally
restricted to overly covalent, brittle materials of a very high strength and hardness. Moreover, for
many ultra-hard materials, these models do not correctly reproduce the values of load-invariant
indentation hardness (e.g. HV) defined as an average pressure beneath the indenter under conditions
of fully developed plasticity – see, e.g., the monograph of Tabor [180]. Thus, the basic problem with
the mentioned models of IH is their applicability in the range of rather low indentation loads only,
i.e., out of the load-invariant range related to higher indentation loads. In the latter range, the non-
linear elastic stress–strain response associated with a perfect crystal no more exists and the plastic
resistance, controlled by the dislocation activity or other plasticity mechanisms, determines the
average pressure beneath the indenter. Relatively good correlation of the shear modulus G with
‘‘plastic’’ hardness HV for most of the super-hard materials is mainly due to the fact that processes
of dislocation multiplication and movement are controlled by the Peierls-Nabarro (PN) stress
directly proportional to G (e.g., [48,181]). This correlation, however, breaks down for materials
exhibiting instabilities of electronic structure or transformations to metastable/instable phases upon
larger shear deformations and indentation loads. It is worthwhile to illustrate this phenomenon
more specifically. For example, C3N4 crystal has a high elastic moduli as reported by Cohen [182]
and Zhang et al. [183] but a low hardness of less than 30 GPa [184]. This is because, upon a shear
strain of about 0.24, the non-binding electron pairs on nitrogen interact with the atomic orbitals of
carbon and the system transforms into a soft graphitic-like phase. Borides of 5d metals which have
very high elastic moduli (low compressibility) and covalent bonds have been claimed to be super-
hard. However, their correctly measured hardness at higher loads, where it is load-invariant, is
below 30 GPa [180,185]. For the ReB2 crystal, for example, the frequently quoted hardness value
of 48 GPa is incorrect since it was measured at a low applied load; the load-invariant value is below
30 GPa [186]. Note that the model of Lazar et al. [111] also predicts an incorrectly high value of the
hardness of ReB2. The reason is that, upon shear, the ReB2 crystal undergoes a number of electronic
instabilities and transformations into metastable or instable phases with lower shear resistance than
that of equilibrium structure at the zero strain [112]. OsB2 and IrB2 have similar structures and high
elastic moduli but very different load-invariant hardness of about 30 and 18 GPa, respectively. This
is because the slightly different electronic structure (one valence electron more in IrB2) causes com-
pletely different plastic deformation paths [113]. There are also other examples which show that the
models of IH of perfect crystals describe their stiffness in the elastic indentation range rather than
their load-invariant resistance to the plasticity assisted indentation.
Some recent results reveal that the load-invariant indentation hardness can be correlated with the
ideal (theoretical) shear strength sis or, even better, with the unstable stacking fault energy (USFE).
Indeed, the resistance to dislocation motion can be expressed in terms of the PN stress spn as
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 145

spn ¼ G=ð1  mÞ expðG=2sis Þ;

 0 slip system
where m is Poisson’s ratio [181]. Jhi et al. [114] calculated the values of sis along ð1 1 0Þ½1 1
of nitrides and carbides of transition metals. The lower hardness of nitrides (in comparison with car-
bides) was attributed to their lower ideal shear strength (shear resistance) as compared to that of car-
bides. Since a strong hydrostatic stress is present beneath the indenter, however, the values of sis
should generally be determined under conditions of coupled hydrostatic pressure (see Sections 3.2
and 4.3). The exponential part of the PN stress correlates the dislocation core structure with an exact
description of the force–displacement relationship (e.g. [112]). In order to better simulate the local-
ized plastic strain associated with the dislocation motion, one can mutually shift two rigid blocks of
atoms to constrain the deformation to the shear plane. For a series of such shear displacements, the
related c-surface according to Vítek [187] can be determined from first principles, along with the val-
ues of USFE. In this way, the lower plastic resistance of TiN (despite its higher elastic moduli) as com-
pared to that of TiC could be explained by higher values of both sis and USFE obtained for the TiC
crystal [115]. The underlying mechanism for such a mechanical response is a directional distribution
of the valence charge around the carbon atoms developing at large strains. Ab initio calculated c-sur-
faces have, of course, wider applications in materials science, e.g., they can also be used for studies of
planar dislocation cores and for evaluations of the Peierls energy and the Peierls stress [188,189], the
quantities important for mechanical behavior of materials. Very recently, the c-surfaces have also
been employed for analysis of behavior of planar defects and dislocations in transition-metal disili-
cides [190].
Note that Zhou et al. [171] found a power-law relationship between ab initio computed values of
USFE and those of the measured hardness. In spite of these valuable findings, however, the state-
of-the-art of the computational ab initio approaches seems to be still rather far from a capability of
constructing a model that would enable a physically justified prediction of such a complex material
characteristic as the load-invariant hardness.

4.2. Intrinsic brittleness and ductility

Brittleness and ductility belong to the most important mechanical properties of solids which are
closely related to their fracture modes. The simplest empirical measures indicating the brittle/ductile
response are the Pugh’s criterion (G/B ratio) [191] and the Cauchy pressure CP [192]. It was believed
that the ductile behavior is associated with G/B < 0.5 and, vice versa, G/B > 0.5 means brittleness.
The Cauchy pressure (for example, for cubic crystals CP = C12  C44) yields negative values for covalent
(brittle) solids such as diamond or Si and positive ones for ductile metals such as Ni or Al [193,170].
However, in the paper [194] it was found that the Cauchy pressure is negative not because of covalent
bonding as it was originally suggested by Pettifor in 1992 [195] but because of long-range electrostatic
potential contributions. That is why, for example, bond-order potentials have a problem in reproduc-
ing negative Cauchy pressure from angular-dependent term and one has to modify them by adding
environmentally repulsive contributions.
More fracture-related and physically based criteria of intrinsic brittleness/ductility are associated
with stability conditions of cracked solids (usually perfect crystals). Indeed, when a cracked crystal
is subjected to a tensile loading, there are basically two extreme possibilities for its behavior – either
an unstable cleavage fracture is observed (brittleness) or a local shear associated with the dislocation
emission stabilizes the crack tip by plastic blunting (ductility). Solutions of that problem are usually
constructed by means of two classical approaches introduced by Kelly et al. [66] (KTC criterion) and
Rice and Thompson [196] (RT criterion). The former criterion deals with the ratio of tensile to shear
ideal strengths riut/sis of the investigated crystal whereas the latter one analyses a related mechanism
of dislocation emission. Both these concepts are physically equivalent: when considering the simplest
estimates of riut [47] and sis [49] and an averaged angle between the crack plane and possible shear
(slip) systems, they lead to an equal relationship [197]. In terms of these simple estimates, moreover,
pffiffiffi
the proportionality relationship sis =riut / G= E shows an approximate connection of criteria pro-
posed by Kelly et al. and Pugh. It should be noted that there are also many metallic and semiconductor
materials in which the crack propagation is assisted by dislocation emission, i.e., the formation of a
146 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

new surface and the dislocation emission at the crack tip occur concurrently. A theory of the fracture
behavior of such materials was proposed by McMahon et al. [198–200]. Such materials lie somewhere
in between the above mentioned extreme cases and represent a transient brittle/ductile behavior.
Nevertheless, the calibration according to both KTC and RT criteria purveys a numerical ranking of
materials that shows whether the behavior of a particular material is closer to intrinsic ductility or
brittleness.
According to the KTC criterion, the crystal is considered to be intrinsically ductile when sis is
exceeded earlier than the riut at the crack tip during its loading. In this case, indeed, the dislocations
are created in a convenient slip system and blunt the tip before an unstable cleavage fracture can
occur. This corresponds to the following condition of intrinsic ductility:
r1 sis =riut sr < 1; ð6Þ
where r1 is the maximum principal stress and sr is the resolved shear stress on the active slip plane
adjacent to the crack tip. In a special case of mode I loading (stress intensity factors KI – 0, KII = KIII = 0),
Eq. (6) can be rewritten as
riht =sis > 2½1 þ sinðH=2Þ= sin H cos U ¼ NKTC ; ð7Þ
where H is the angle between the crack plane and the intersecting slip plane and U is the angle
between the Burgers vector and the direction perpendicular to the crack front [65]. The left-hand side
of Eq. (7) is a ratio of material characteristics where the uniaxial ideal strength riut was replaced by the
hydrostatic strength riht since this fracture characteristic is more relevant to the stress state at the
crack front. Indeed, an almost isotropic tension r1 = r2 = 1.6r3 acts in the crack plane ahead of the
crack front under plane strain conditions [201]. Using ab initio approaches, the ratio riht/sis for a par-
ticular crystal can be computed in a sufficiently precise way since both values of riht and sis calculated
with respect to stability conditions do not significantly differ from those received without taking sta-
bility analyses into account – see Table 1. The right-hand side (NKTC) depends only on crystallography,
i.e., on a particular crack plane and the related shear (slip) system (crack/shear configuration). Let us
note that only slip in crystallographic planes containing the crack front can effectively produce ductile
blunting of the crack tip (so called blunting configurations). Therefore, values NKTC were calculated for
all possible combinations of crack planes {h k l} and crack fronts u v w (for low Miller and direction
indices 0 6 h, k, l, u, v, w 6 4) that are related to blunting slip systems. The mean value NKTC, averaged
over all investigated configurations in a particular crystal (solid), can then be accepted as a proper
parameter for a particular crystal structure of the solid [202].
In the frame of the RT criterion, a balance between the repulsive force on the emitted dislocation
(induced by external loading) and the attractive mirror force (induced by a presence of free surfaces at
crack flanks) gives a basic equation for an assessment of the ductile/brittle behavior. When the crack

Table 2
Intrinsic brittleness and ductility of crystals according to KTC (given in bold) and RT criteria.

Crystal Structure sis (GPa) riht (GPa) CKTC CRT

C diam 95 88.5 6.7 4.5
Mo bcc 15.1 43 3.0 5.2
Ge diam 4.9 11.1 2.8 3.4
W bcc 17.5 57.4 2.6 4.7
Si diam 7.3 19 2.4 2.7
Fe bcc 7.2 27.7 2.2 2.4
Ir fcc 17.1 48 1.8 3.6
Nb bcc 6.1 34.1 1.5 1.7
Al fcc 3.16 11.7 1.4 1.2
V bcc 5.1 37 1.3 1.9
Ta bcc 5.7 41.8 1.2 2.0
Ni fcc 5.3 28.5 0.96 1.7
Cu fcc 2.65 22.5 0.61 1.2
Ag fcc 1.65 16.9 0.50 1.1
Pt fcc 2.75 41 0.35 1.3
Au fcc 1.25 24 0.27 0.9
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 147

driving force corresponding to such a balance is lower than that related to the Griffiths condition of
unstable cleavage fracture, the crystal is intrinsically ductile. This leads, in an analogy to Eq. (7), to
the following condition of intrinsic ductility:

cs =cb > 4½1 þ ð1  mÞ tan2 U=½ð1 þ cos HÞ sin2 H ¼ NRT ; ð8Þ

where cs is the surface energy (2cs is the fracture energy) and cb is the energy barrier for the disloca-
tion nucleation [65].
According to Eq. (7) or (8), the condition of intrinsic crystal ductility reads CKTC  NKTC sis/riht < 1 or
CRT  NRT cb/cs < 1, while CKTC > 1 or CRT > 1 indicates intrinsic brittleness. Mean values of both CKTC
and CRT are displayed for 15 crystals with cubic lattices in Table 2.
The ranking of crystals is made according to the values of CKTC (given in bold) starting from the
most brittle ones (C-diam, Mo, Ge, W and Si) and ending with the most ductile ones (Ni, Cu, Ag, Pt
and Au). Note that the ranking on the basis of CRT gives a similar result. Both rankings are roughly
in a qualitative agreement with a general experience with ductile/brittle behavior of engineering
materials based on a particular element. From the quantitative point of view, however, all the values
of CKTC and CRT (calculated for T = 0 K) are somewhat shifted toward the intrinsic brittleness when
compared with our experience associated with higher temperatures. This is easily understandable
for the ranking based on the CRT values: the thermal activation significantly reduces the local energy
barrier cb for the dislocation nucleation already in the temperature range (0; 300) K while, in this
range, its influence on the value of surface energy cs is relatively negligible. At higher temperatures,
therefore, the values of CRT will be smaller than those at 0 K and the ranking will be shifted more
to the ductile side. In the case of CKTC-based ranking, however, the temperature dependences of both
sis and riht can be considered to be comparable which means that a negligible change of CKTC values
with increasing temperature may be expected. Here, the main physical reason for such a shift is dif-
ferent: a decrease in the ideal shear strength with increasing tensile normal stress rn in the crack-tip
shear systems during the loading. The ab initio calculated functions sis(rn) are available for all crystals
under evaluation [63,68,69,72] and, therefore, one can make a correction of the CKTC-based ranking
with respect to this phenomenon. For blunting crack/slip configurations, a simple formula
sr =rn ¼ ð1 þ cos HÞ=ðsin H cos UÞ stands for the ratio of shear to normal stresses acting at the crack
tip [65]. This formula gives a high mean normal stress rn,m  3.4 sr in the shear systems close to
the mean angles H = U = p/4. During the crack-tip loading, the shear stress remains in the range sr
2 h0; sis(0)i, where sis(0) is the ideal shear strength calculated for rn = 0. This results in an averaged
applied value r  n  rn;m =2 ¼ 1:7 sis(0) that is related to sr = sis(0)/2. The values of r
 n and the resulting
CKTC(rn)-values are displayed in Table 3 for all investigated crystals. One can clearly see that the crys-
tals with diamond lattice remain definitively intrinsically brittle (CKTC  1), the bcc crystals get closer
to the brittle/ductile transition boundary (CKTC P 1) and the fcc crystals come close to the intrinsic

Table 3
Ranking of crystals according to modified KTC criterion.

Crystal Structure r n (GPa) CKTC(rn)

Ge diam 8.57 3.1
Si diam 12.8 2.6
C diam 166.0 2.1
Fe bcc 12.6 1.5
Nb bcc 10.7 1.5
V bcc 9.80 1.4
Mo bcc 26.4 1.3
W bcc 30.6 1.3
Ir fcc 29.9 1.1
Ta bcc 9.98 1.0
Ni fcc 9.28 0.73
Al fcc 5.53 0.53
Cu fcc 4.64 0.46
Ag fcc 2.89 0.15
Pt fcc 4.81 0.15
Au fcc 2.19 0.11
148 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

ductility (CKTC  1). The fcc crystal of Ir represents the only one exception to that rule. Indeed, this
crystal contains a high portion of covalent bonds [72] and, therefore, it exhibits a transient ductile/
brittle response close to bcc crystals. The extremely high value of r  n  166 GPa at the crack tip in
the crystal of C (diamond) requests a large extrapolation of the calculated sis(rn) data which could
influence the mutual ranking of the diamond-based crystals. Nevertheless, the intrinsic brittleness
of the diamond crystal remained valid without any doubt. Thus, the ranking in Table 3 is already in
a very good agreement with a common experience about the ductile/brittle behavior of polycrystalline
engineering materials. This means that the tendency of a particular engineering material (compound)
to brittle or ductile behavior is, at least partially, predetermined by intrinsic properties of the crystal
lattice of its basic element.
Another ab initio based study as an attempt to systematically classify the intrinsic ductility and
brittleness of various crystals was made by Ogata et al. [76], where ‘‘shearability’’ and ‘‘tensibility’’
were defined as sm  arg max sðsÞ and t m  arg max rh ðð1 þ eÞV 0 Þ, respectively. Here s and e are the
engineering shear and isotropic tensile strains, rh is the hydrostatic stress and V 0 is the equilibrium
volume. That is, the shearability and the tensibility designates the strain at which the corresponding
stress reaches its maximum (ideal strength). In this study, which covered 22 various metals, alloys and
ceramics, it was demonstrated that there is an approximately linear relation between the ideal shear
strength sis and the shearability ðsis ’ 2Gsm =pÞ, while the variation in the tensibility among the crys-
tals is relatively small. Moreover, it was proposed that the two material properties suggested can be
used for materials design; for example, the brittleness parameter of Rice [203] may be crudely esti-
mated as b / ðG=BÞðs2m Þ=ðt 2m Þ. This parameter is the inverse to the cs/cb ratio in Eq. (8) that expresses
the crystal ductility. Thus, this study qualitatively confirms the brittleness/ductility ranking of crystals
in Table 3 and, moreover, it indicates the position of investigated hcp crystals as follows: Mg in
between bcc and fcc metals while Ti and Zn among the fcc crystals.

4.3. Nanoindentation

The nanoindentation is considered to be a very promising experimental approach to studying the

incipient plasticity or deformation-induced phase transformations in single crystals and polycrystals
containing defects (e.g. [204–208]). Since the stressed volume beneath the sharp indenter may be
locally dislocation-free, nanoindentation can also be used for an experimental verification of calcu-
lated values of ideal shear strength sis (e.g. [209,210]). According to the Hertzian contact model
[211], local shear component of the stress field reaches its maximum value in a certain depth under
the surface of the indented crystal (under the indenter tip). When this maximum stress approaches
sis, it can be high enough to nucleate and emit dislocations. This process is detected as a characteristic
discontinuity (so-called pop-in) on the load–displacement curve and the related local plasticity could
be experimentally identified in the form of dislocation rosettes [212].
During the last 20 years, a number of theoretical models simulating the nanoindentation test have
been published (e.g. [213–218]). These models utilized either the simple Hertzian solution of stress
distribution under a sphere indenter or more sophisticated multiscale approaches. The multiscale
models include atomistic simulations based on either molecular dynamics or ab initio methods and
can be build up using different integration concepts: simultaneous and sequential. The models based
on ab initio calculations mostly utilize sequential integration of approaches on three different levels.
In these models, the atomistic level consists of two independent computations to determine: (i) the
nonlinear anisotropic stress–strain characteristics of a perfect crystal relevant to the loading (inden-
tation) direction and (ii) the dependence of sis on the hydrostatic stress rh, i.e., the function sis (rh)
up to the pop-in event. The mesoscopic level includes calculations of both the resolved shear stress
sr and the hydrostatic stress rh in all possible crystallographic shear planes up to the pop-in event.
A finite element analysis of the stress field beneath the nanoindenter during the indentation process
represents the third, macroscopic level of the multiscale model. Such models can than predict both the
penetration depth and the indentation force related to first reaching sis in one of the possible shear
systems. Hereafter, we will briefly discuss the results of some of these models simulating nanoinden-
tation in fcc and bcc metallic crystals.
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 149

Fig. 4. Influence of hydrostatic rh and normal rn stresses on the values of ideal shear strength sis for crystals of nickel and
tungsten. All the stress values are normalized by the value of strength s0is under pure shear.

4.3.1. FCC metals

Multiscale models of nanoindentation tests in Ni and Cu single crystals were reported by Horníková
et al. [219,218]. A three-dimensional isotropic FEM model of the nanoindentation tests was used with
the help of the finite element ANSYS code. A spheroconical diamond indenter with the radius of
0.2 lm (0.5 lm) in agreement with related experiments for Ni (Cu) was pressed into a 5 lm thick sub-
strate disc with the radius of 10 lm. A Coulomb model was applied in the analysis involving friction
(see [218] for details). Tension/compression stress–strain curves for [0 0 1] indentation direction in Cu
and Ni crystals were calculated from first principles. Multilinear approximations of these curves were
utilized in the ANSYS code procedure as an equivalent stress–strain dependence. The functions sis(rn)
for the set of {1 1 1} slip planes were also computed using the ab initio approach. In fcc metals, the
normal stress rn could be employed instead of the more general hydrostatic stress rh since the depen-
dences sis(rn) and sis(rh) are well comparable in the whole range of applied loading up to the pop-in
event (see Fig. 4).
A global calculation procedure simulated, step by step, the penetration of the indenter into the
crystal (substrate) in order to identify the critical site within some of the crystallographic planes

h1 1 2i{1 1 1} where the condition for the dislocation emission was firstly reached. This particularly
means that levels of the resolved shear stress sr, the normal stress rn and the related sid (rn) in all
the shear systems inside the whole crystal were gradually tracked during the indentation.
Eventually, the critical condition sr,max = sis (rn) was reached in one of the mesh elements close to
the loading axis and thus the predicted pop-in effect corresponded to the penetration depth
hNi = 5.5 nm (indentation force FNi = 35 lN) in the Ni crystal and hCu = 12 nm (FCu = 100 lN) in the
Cu crystal. The related experimentally determined depths were h⁄Ni = 4.5 nm (F⁄Ni = 40 lN) and
h⁄Cu = 10 nm (F⁄Cu = 150 lN). In spite of some differences in experimental and theoretical values, the devi-
ations within the range of tens of percent are still well acceptable.

4.3.2. BCC metals

First estimates of the ideal shear strength for bcc crystals from nanoindentation experiments
yielded surprisingly high values. For example, Bahr et al. [213] estimated the value of sis for the bcc
tungsten above 28 GPa, compared to sis values in Table 1 which range from 16.8 to 18.2 GPa. These
overestimations were partially attributed to a nonlinearity of the stress–strain response (that is not
included in the Hertzian theory) [220] or to the effect of triaxiality of stress state at the point of crystal
instability [215]. The determination of the function sis(rh) for bcc crystals is absolutely necessary
since, unlike for rather weak fcc crystals, the dependences sis(rn) and sis(rh) for strong bcc crystals
150 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

are significantly different [221], particularly for very high applied normal and hydrostatic stresses
close to the onset of pop-in event – see Fig. 4.
Hereafter, the results recently reported by Šandera et al. [221] for the single crystal of tungsten will
be mentioned since, unlike in other papers, the multiscale model was coupled with special and thor-
ough nanoindentation experiments. These experiments were performed on a Hysitron Triboscope,
where the indentation device has been mounted on the scanner head of an atomic force microscope
instead of the original standard cantilever holder. A spheroconical indentation tip of the radius
R = 850 nm and the angle a = 70° was used in experiments. A careful statistics of load–displacement
curves was performed based on a set of 100 nanoindentations applied in the [1 1 1] crystallographic
direction. The related penetration depth of the nanoindenter was h⁄W = 87.7 nm and the loading force
F⁄W = 8050 lN. The theoretically predicted pop-in effect, associated with the critical condition sr,max =
sis(rh), corresponded to the indentation depth hW = 85.9 nm in a very good agreement with the
experimental value. Although the related loading force FW = 9760 lN was found to be higher than
the experimental one, the multiscale model could be considered to give a reasonable prediction of the
pop-in event.
The results of multiscale models reveal that the nanoindentation may serve as a very good tool for
verification of ab initio calculated values of ideal shear strength at least for cubic crystals.

5. Conclusions

During the last two decades, ab initio methods started to be frequently used for theoretical predic-
tions of mechanical properties of solids due to their capability to provide sufficiently precise data on
mechanical states far away from the stress-free equilibrium conditions. Of course, there are still some
serious limitations of their application as, for example, restrictions to zero absolute temperature or to
rather small supercells without extended lattice defects. Nevertheless, basic intrinsic mechanical char-
acteristics of crystals of many elements and alloys could already be calculated, as also documented in
this review article. Values of elastic constants can be predicted in a satisfactory agreement with exper-
iments not only near the stress-free state but also during the whole range of attainable deformations.
This enables us to perform analyses according to Born’s elastic stability conditions important for cor-
rect predictions of ideal strength values under various kinds of tensile or compressive loadings.
However, one must be careful with an appropriate modification of stability conditions with respect
to both the adopted definition of elastic coefficients and the kind of applied loading (deformation).
This is illustrated by a numerical simulation of isotropic tensile loading of a molybdenum crystal as
presented in one of the appendices of this article. Since the elastic stability analysis is a rather cum-
bersome procedure, a simplified analysis for crystals under uniaxial loading was proposed and veri-
fied. This approximate method is simple, time-saving and gives results well comparable to data
available from the standard elastic analysis. Moreover, it can be used for any crystallographic direction
with just a very little computational effort. However, the elastic instability represents just a long
wavelength limit of an instability related to soft phonons. Therefore, the ab initio aided dynamic sta-
bility analysis based on phonon spectra represents the most sophisticated and precise tool for deter-
mination of theoretical strength of solids. As demonstrated in the second Appendix, models of
microscopic deformation corresponding to soft phonons can serve as very useful supplements to
the dynamic stability analysis. Thanks to such analyses, the predicted values of ideal strength start
to closely approach the experimental data obtained for nanowhiskers or from multi-scale models of
nanoindentation tests. On the other hand, advanced ab initio approaches are, in general, still not able
to yield adequate theoretical predictions of intrinsic hardness. The ab initio calculated ranking of
intrinsic brittleness/ductility response of crystals shows that the tendency of a particular engineering
material (compound) to brittle or ductile behavior is, at least partially, predetermined by intrinsic
properties of the crystal lattice of its basic element. On the other hand, the intrinsic fracture behavior
of perfect crystals subjected to triaxial tensile loading (the higher triaxiality, the higher fracture strain)
is opposite to that of engineering materials. In the latter materials, the intrinsic properties are com-
pletely suppressed by the presence of defects as dislocations and secondary-phase particles.
Although the ab initio calculations of strength of interfaces and segregated grain boundaries are only
in the very first stage, it can be expected that their importance will rapidly increase.
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 151

Finally, let us make some remarks to future development of ab initio calculations and methods. In
many cases of loading modes and materials, the stability of a solid revealed to be controlled by the
occurrence of soft phonons. Therefore, the dynamic stability analyses of various kinds of materials
subjected to different loading modes certainly belong to most challenging tasks. Ab initio calculations
of cohesive strength of interfaces, segregated grain boundaries, surface layers and strength of compos-
ites are also expected to play an important role in the near future. Ab-initio aided computations of
changes in the geometry of dislocation cores will be performed for various advanced materials and
loading conditions. Semi-empirical potentials adjusted by ab initio methods will be preferentially uti-
lized in molecular-dynamic models of deformation and fracture of solids and, last but not least, the
first-principles calculations of mechanical properties of virtual alloys and materials have bright pro-
spects. As concerns the future of ab initio methods, a considerable effort must be spent in a develop-
ment of temperature-sensitive concepts. One of the main challenges is to find suitable internal
approximations to solve some problems appearing with a correct identification of ground-state struc-
tures of more complex materials as, for example, the NiTi martensite [166]. There is also a big exper-
imental challenge closely related to ab initio calculations. Indeed, there are still too few reliable
experimental data on the theoretical strength and interface properties.

Acknowledgments

This research was supported by the Czech Science Foundation [Projects No. GA P108/12/0144 (JP),
GA P108/12/0311 (MČ) and GA 14-22490S (MŠ)], by the Project CEITEC – Central European Institute of
Technology (CZ.1.05/1.1.00/02.0068) from the European Regional Development Fund (JP, MČ, MŠ) and
by the Academy of Sciences of the Czech Republic [Institutional Project RVO:68081723 (MŠ)].

Appendix A

A.1. Numerical experiment for elastic constants and moduli

In order to illustrate differences between elastic coefficients (and elastic moduli) calculated according
different definitions for a nonzero stress, we focus on changes of the bulk modulus during isotropic defor-
mation of a cubic crystal. The magnitude of isotropic deformation can be expressed by a relative volume v
that is defined as a ratio of the current (reference) volume V to the equilibrium volume V0 (of the
unstressed crystal). Since the cubic symmetry is preserved during the loading, the crystal elasticity
can be characterized by three elastic coefficients along the entire deformation path. The bulk modulus
B is a linear combination of two of them. To calculate a dependence of its value on the crystal volume,
B is evaluated for several (reference) volumes imposing the following perturbation: lattice parameter
ar corresponding to each reference volume is multiplied by ð1 þ dÞ with d being a small deformation from
the reference state. The related change of the relative volume can be then calculated as v r ¼ ð1 þ dÞ3 . The
bulk modulus is defined for materials under hydrostatic pressure as the ratio of the applied pressure to
the negative dilatation [16]. Since we focus here on a region of tensile stresses, the first definition of the
bulk modulus we use is based on the change of isotropic stress r with changing vr as
dr
B1 ¼ ;
dv r
Another possibility is to evaluate the elastic coefficients C11 and C12 and to express the bulk modulus
as B ¼ 13 ðC 11 þ 2C 12 Þ. The elastic coefficients can be calculated, e.g., from the energy expansion (3). If
the strain is defined as a small strain, then the aforementioned distortion yields e1 ¼ e2 ¼ e3 ¼ e ¼ d
and the bulk modulus can be calculated
2
1 d U
B2 ¼ :
9V r de2
2
Following the Lagrangian (finite) strain definition, one gets g1 ¼ g2 ¼ g3 ¼ g ¼ d þ d2 and for the
bulk modulus
152 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

2
1 d U
B3 ¼ :
9V r dg2
The last definition considered in this appendix is also based on the elastic coefficients. In this case,
however, the coefficients are calculated according to the Hooke’s law (2) and the bulk modulus is
expressed using the relation

 
1 dr1 dr2
B4 ¼ þ2 :
3 de1 de1
In this case, the lattice is distorted from the reference cubic state applying the uniaxial strain e1 ¼ d
with e2 ¼ e3 ¼ 0.
All the four definitions were applied to a bcc crystal of Mo and the energy and stresses were com-
puted by the ab initio code VASP [222]. In order to minimize the numerical errors, values of d were
changed in the range of 0 ± 0.010 with the increment of 0.005 and the computational parameters were
relatively tight. The Projector Augmented Waves method implemented in the VASP [223] was used
with the energy cutoff of 500 eV and the Brillouin zone was sampled using 29  29  29 k-points.
The energy was converged to change less than 0.1 leV in two subsequent iterations.
The results are displayed in Fig. 5. As can be seen, values of B1 and B4 are practically equal since the
data points cover. Although the two approaches are very different, they are consistent in one impor-
tant aspect: They both are calculated using a stress–strain response. The lines connecting the related
data points intersect the zero line approximately at v = 1.52. At this relative volume, the isotropic
stress reaches its maximum of 43.4 GPa. In the case of isotropic loading, the volume corresponding
to the stress maximum also corresponds to the point of inflexion on the energy-volume curve.
Since the bulk modulus must vanish at the stress maximum, definitions of B1 and B4 are consistent
with original Born’s criterion for stability that yields for isotropic loading of a cubic lattice set of con-
ditions C 11 þ 2C 12 > 0; C 11  C 12 > 0; C 44 > 0.
The other two definitions of bulk modulus that are based on changes of energy gave systematically
greater values of B2 and B3. Their differences in comparison with B1 and B4 increase with increasing v.
From the work of Wallace [18] or from the elastic stability conditions derived later [43,44] it follows
that the linear combination C 11 þ 2C 12 calculated from the energy (3) as a function of Lagrangian strain
differs from that calculated from the stress–strain response by a value of the isotropic stress r. If the
energy is expanded as a function of a small strain or if the elastic constants are defined according to
Milstein [41,15] using a set of natural variables (lengths of cell edges and the angles between them),

Fig. 5. The bulk modulus calculated using four different definitions for several values of relative volume v. The data points for B1
and B4 cover.
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 153

the resulting value of C 11 þ 2C 12 differs even by 2r [1]. Then, for the values of bulk modulus, it follows
B2 = B1 + r/3 and B3 = B1 + 2r/3. When we modify the values of B2 and B3 by subtracting r/3 and 2r/3,
all dependences in Fig. 5 unify.
Some authors [224,225] use volume conserving distortions to compute elastic moduli from the
energy as a function of strain. Since there is no distortion conserving crystal volume and giving B
directly from the energy expansion, we add here also a comparison of values of tetragonal shear mod-
ulus C 0 ¼ ðC 11  C 12 Þ=2. When we apply the volume-conserving orthorhombic strain e1 ¼ d, e2 ¼ d
2
and e3 ¼ 1d
d
2 to the crystal, the shear modulus can be obtained from the relation

2
1 d U
C 01 ¼ :
4V r dd2
Using the same strain as for calculation of B4, one can get the tetragonal shear modulus from the
stress
 
1 dr1 dr2
C 02 ¼  :
2 de1 de1
Fig. 6 displays values of C0 obtained from both definitions for several relative volumes. It can be
clearly seen that the values are almost equal. Thus, this numerical experiment also illustrates the
equivalence of elastic moduli obtained from stress–strain response and those calculated from the
energy expansion using volume conserving deformations.

A.2. Model of microscopic deformations related to soft phonons

Here we present an example of a dynamic stability analysis. Let us study phonon spectra of the fcc
crystal of Pt under the same loading as in Appendix A.1, i.e., under the isotropic tension. This system
was previously studied in paper [55].
Fig. 7 displays dispersion curves (based on the data from Ref. [55]) calculated for three levels of iso-
tropic strain e (from 0.07 to 0.09). Values of the corresponding relative volume v ¼ ð1 þ eÞ3 are 1.225,
1.260 and 1.295. Only dispersions that predict soft phonon modes are included in the figure. According
to the calculated phonon spectra, the fcc crystal of Pt becomes elastically unstable at the isotropic
strain of 0.09 but the first instability occurs in the lattice already at a strain below 0.08. Soft modes
occur in U ? K and U ? X directions in the Brillouin zone. According to our analysis of dynamical
matrix, phonons in both directions are transverse. Since the instability associated with a transverse
phonon in U ? K (i.e., [1 1 0]) direction (polarized in [0 0 1]) was studied in [226] on the example of

Fig. 6. Values of the tetragonal shear modulus C 0 calculated using two different definitions as functions of the relative crystal
volume v.
154 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

Fig. 7. Phonon dispersion lines for fcc crystal of Pt at three isotropic strains e. As usual, the negative vertical axis is used for
plotting imaginary frequencies. The vertical dashed line corresponds to a soft phonon wave vector qs.

qs ¼ 13 ½1 0 0 2ap
Fig. 8. The model of a microscopic deformation of an fcc lattice corresponding to a soft phonon of the wavevector ~
with the polarization in [0 1 0]. Displacements of atoms in (1 0 0) lattice planes are indicated by the arrows.

Fig. 9. Crystal energy Etot as a function of the (1 0 0) lattice planes displacement d in the [0 1 0] (transverse mode) direction for
three values of the isotropic strain e. E0 is energy of the undistorted cubic crystal.
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 155

fcc Ir, we pay our attention here to the U ? X (i.e., [1 0 0]) direction only. In this direction, the two
transverse modes are degenerated (polarizations [0 1 0] and [0 0 1] are equivalent). The instability
develops approximately for a soft phonon of the wavevector ~ qs ¼ 13 ½1 0 0 2ap, where a is the lattice
parameter. Fig. 8 shows a model of microscopic deformation corresponding to this soft phonon.
The wavelength of the periodic displacements of (1 0 0) lattice planes is 3a. Instabilities predicted
by phonon spectra were then verified by building the computational model according to Fig. 8 and
computing the total energy Etot as a function of the displacement amplitude d. The results are pre-
sented in Fig. 9. In agreement with the phonon spectra in Fig. 7, displacing the (1 0 0) planes in the
[0 0 1] direction lowers the crystal energy already at the strain of 0.08. Decrease of the energy is, how-
ever, very small since the instability corresponding to the considered wavevector has its onset just
about this strain. At the isotropic strain of 0.09, the decrease of the energy is clearly visible.
Using the model of microscopic deformation we also verified that the displacements of (1 0 0) lat-
tice planes along the [1 0 0] direction (the longitudinal mode) do not lower the crystal energy for any of
the isotropic strains.

References

[1] Pokluda J, Černý M, Šandera P, Šob M. J Comput Aided Mater Des 2004;11:1–28.
[2] Ziman JM. Principles of the theory of solids. Cambridge University Press; 1972.
[3] Hohenberg P, Kohn W. Phys Rev 1964;136:B864–71.
[4] Kohn W, Sham LJ. Phys Rev 1965;140:A1133–8.
[5] Ziesche P, Kurth S, Perdew JP. Comput Mater Sci 1998;11:122–7.
[6] Janesko BG. Int J Quantum Chem 2013;113:83–8.
[7] Perdew JP, Ruzsinszky A. Int J Quantum Chem 2010;110:2801–7.
[8] Bowler DR, Miyazaki T. Rep Prog Phys 2012;75:036503.
[9] Saad Y, Chelikowsky J, Shontz S. SIAM Rev 2010;52:3–54.
[10] Hafner J. J Comput Chem 2008;29:2044–78.
[11] Martin RM. Electronic structure: basic theory and practical methods. Cambridge University Press; 2004.
[12] Lejaeghere K, Van Speybroeck V, Van Oost G, Cottenier S. Crit Rev Solid State Mater Sci 2014;39:1–24.
[13] Hüger E, Zelený M, Káňa T, Šob M. Phys Stat Sol (RRL) 2008;2:117–9.
[14] Nye JF. Physical properties of crystals: their representation by tensors and matrices. Oxford University Press; 1985.
[15] Hill R, Milstein F. Phys Rev B 1977;15:3087–96.
[16] Huntington HB. The elastic constants of crystals. Academic Press; 1958.
[17] Ledbetter HM, Reed RP. J Phys Chem Ref Data 1973;2:531–618.
[18] Wallace DC. Thermodynamics of crystals. New York: John Wiley & Sons; 1972.
[19] Krenn CR, Roundy D, Morris Jr JW, Cohen ML. Mater Sci Eng A 2001;319–321:111–4.
[20] Chrzan DC, Sherburne MP, Hanlumyuang Y, Li T, Morris Jr JW. Phys Rev B 2010;82:184202.
[21] Sawyer CA, Morris Jr JW, Chrzan DC. Phys Rev B 2013;87:134106.
[22] Esposito E, Carlsson AE, Ling DD, Ehrenreich H, Gelattt CD. Philos Mag A 1980;41:251–9.
[23] Paxton AT, Gumbsch P, Methfessel M. Philos Mag Lett 1991;63:267–74.
[24] Xu W, Moriarty JA. Phys Rev B 1996;54:6941–51.
[25] Soderlind P, Moriarty JA. Phys Rev B 1998;57:10340.
[26] Craievich PJ, Weinert M, Sanchez JM, Watson RE. Phys Rev Lett 1994;72:3076.
[27] Alippi P, Marcus PM, Scheffler M. Phys Rev Lett 1997;78:3892–5.
[28] Price DL, Cooper BR, Wills JM. Phys Rev B 1992;46:11368–75.
[29] Šob M, Wang LG, Vitek V. Mater Sci Eng A 1997;234–236:1075–8.
[30] Mihkailovskii I, Poltinin P, Fedorova L. Fizika Tverdogo Tela 1981;23:1291–5.
[31] Šob M, Wang LG, Vitek V. Kovove Mater 1998;36:145–52.
[32] Šandera P, Pokluda J, Wang LG, Šob M. Mater Sci Eng 1997;234–236:370.
[33] Černý M, Šandera P, Pokluda J. Czech J Phys 1999;49:1495–501.
[34] Song Y, Yang R, Li D, Wu WT, Guo ZX. Phys Rev B 1999;59:14220–5.
[35] Šob M, Pokluda J, Černý M, Šandera P, Vitek V. Mater Sci Forum 2005;482:33–8.
[36] Ogata S, Umeno Y, Kohyama M. Modell Simul Mater Sci Eng 2009;17:013001.
[37] Grimvall G, Magyari-Kope B, Ozolins V, Persson KA. Rev Mod Phys 2012;84:945–86.
[38] Morris Jr JW, Krenn CR, Roundy D, Cohen ML. Elastic stability and the limits of strength. In: Turchi PE, Gonis A, editors.
Phase transformations and evolution in materials. Warrendale: TMS; 2000.
[39] Born M. Proc Cambridge Philos Soc 1940;36:160–72.
[40] Born M, Furth R. Proc Cambridge Philos Soc 1940;36:454.
[41] Milstein F. Phys Rev B 1971;3:1130–41.
[42] Wang J, Li J, Yip S, Phillpot S, Wolf D. Phys Rev B 1995;52:12627–35.
[43] Wang J, Yip S, Phillpot SR, Wolf D. Phys Rev Lett 1993;71:4182–5.
[44] Zhou Z, Joos B. Phys Rev B 1996;54:3841–50.
[45] Morris Jr JW, Krenn CR. Philos Mag A 2000;80:2827–40.
[46] Wang H, Li M. J Phys: Condens Matter 2012;24:245402.
[47] Orowan E. Rep Prog Phys 1949;12:185.
156 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

[48] Kelly A, Macmillan MH. Strong solids. Oxford: Clarendon Press; 1986.
[49] Frenkel J. Z Phys 1926;37:572.
[50] Zhao GL, Harmon BN. Phys Rev B 1992;45:2818–24.
[51] Binggeli N, Keskar NR, Chelikowsky JR. Phys Rev B 1994;49:3075–81.
[52] Parlinski K, Li ZQ, Kawazoe Y. Phys Rev Lett 1997;78:4063–6.
[53] Clatterbuck DM, Krenn CR, Cohen ML, Morris Jr JW. Phys Rev Lett 2003;91:135501.
[54] Dubois SM, Rignanese GM, Pardoen T, Charlier JC. Phys Rev B 2006;74:235203.
[55] Řehák P, Černý M, Pokluda J. J Phys: Condens Matter 2012;24:215403.
[56] Gonze X, Lee C. Phys Rev B 1997;55:10355–68.
[57] Clatterbuck DM, Chrzan DC, Morris Jr JW. Scripta Mater 2003;49:1007–11.
[58] Černý M, Pokluda J. Phys Rev B 2007;76:024115.
[59] Černý M, Pokluda J. Phys Rev B 2010;82:174106.
[60] Černý M, Pokluda J. Comput Mater Sci 2011;50:2257–61.
[61] Černý M, Řehák P, Umeno Y, Pokluda J. J Phys: Condens Matter 2013;25:035401.
[62] Černý M, Šesták P, Pokluda J, Šob M. Phys Rev B 2013;87:014117.
[63] Černý M, Pokluda J. Mater Sci Forum 2008;567–568:73–6.
[64] Řehák P, Černý M, Pokluda J. Key Eng Mater 2011;465:183–6.
[65] Pokluda J, Šandera P. Micromechanisms of fracture and fatigue: in a multiscale context. London: Springer; 2010.
[66] Kelly A, Tyson WR, Cottrell AH. Philos Mag 1967;15:567–81.
[67] Černý M, Pokluda J. Mater Sci Eng A 2008;483–484:692–4.
[68] Černý M, Pokluda J. Comput Mater Sci 2008;44:127–30.
[69] Černý M, Šesták P, Pokluda J. Comput Mater Sci 2010;47:907–10.
[70] Černý M, Šesták P, Pokluda J. Comput Mater Sci 2012;55:337–43.
[71] Ogata S, Li J, Yip S. Science 2002;298:807.
[72] Černý M, Pokluda J. Acta Mater 2010;58:3117–23.
[73] Umeno Y, Černý M. Phys Rev B 2008;77:100101.
[74] Umeno Y, Shiihara Y, Yoshikawa N. J Phys: Condens Matter 2011;23:385401.
[75] Černý M, Pokluda J. J Phys: Condens Matter 2009;21:145406.
[76] Ogata S, Li J, Hirosaki N, Shibutani Y, Yip S. Phys Rev B 2004;70:104104.
[77] Telling RH, Pickard CJ, Payne MC, Field JE. Phys Rev Lett 2000;84:5160.
[78] Roundy D, Cohen ML. Phys Rev B 2001;64:212103.
[79] Huang YM, Spence JCH, Sankey OF. Philos Mag A 1994;70:53–62.
[80] Černý M, Pokluda J. J Alloys Compd 2004;378:159–62.
[81] Kitamura T, Umeno Y. Modell Simul Mater Sci Eng 2003;11:127.
[82] Li WX, Wang TC. J Phys: Condens Matter 1998;10:9889–904.
[83] Roundy D, Krenn CR, Cohen ML, Morris Jr JW. Phys Rev Lett 1999;82:2713–6.
[84] Ogata S, Kitagawa H. Comput Mater Sci 1999;15:435–40.
[85] Liu YL, Zhang Y, Zhou HB, Lu GH, Kohyama M. J Phys: Condens Matter 2008;20:335216.
[86] Černý M, Pokluda J, Šob M, Friák M, Šandera P. Phys. Rev. B 2003;67:035116.
[87] Černý M, Šob M, Pokluda J, Šandera P. J Phys: Condens Matter 2004;16:1045–52.
[88] Wang H, Li M. J Phys: Condens Matter 2010;22:295405.
[89] Wang H, Li M. J Phys: Condens Matter 2009;21:455401.
[90] Nagasako N, Jahnátek M, Asahi R, Hafner J. Phys Rev B 2010;81:094108.
[91] Li X, Schoenecker S, Zhao J, Johansson B, Vitos L. Phys Rev B 2013;87:214203.
[92] Clatterbuck DM, Chrzan DC, Morris Jr JW. Philos Mag Lett 2002;82:141.
[93] Friák M, Šob M, Vitek V. Philos Mag 2003;83:3529.
[94] Clatterbuck DM, Chrzan DC, Morris Jr JW. Acta Mater 2003;51:2271.
[95] Luo W, Roundy D, Cohen ML, Morris Jr JW. Phys Rev B 2002;66:094110.
[96] Roundy D, Krenn CR, Cohen ML, Morris Jr JW. Philos Mag A 2001;81:1725–47.
[97] Brenner SS. J Appl Phys 1956;27:1484–91.
[98] Kiener D, Grosinger W, Dehm G, Pippan R. Acta Mater 2008;56:580–92.
[99] Richter G, Hillerich K, Gianola DS, Mönig R, Kraft O, Volkert CA. Nano Lett 2009;9:3048–52.
[100] Brenner SS. J Appl Phys 1957;28:1023–6.
[101] Shpak AP, Kotrechko SO, Mazilova TI, Mikhailovskij IM. Sci Technol Adv Mater 2009;10:045004.
[102] Kotrechko S, Mikhailovskij I, Mazilova T, Ovsjannikov O. Key Eng Mater 2014;592–593:301–6.
[103] Mikhailovskij IM, Mazilova TI, Voyevodin VN, Mazilov AA. Phys Rev B 2011;83:134115.
[104] Mikhailovskij IM, Sadanov EV, Kotrechko S, Ksenofontov VA, Mazilova TI. Phys Rev B 2013;87:045410.
[105] Lee C, Wei X, Kysar JW, Hone J. Science 2008;321:385–8.
[106] Liu F, Ming P, Li J. Phys Rev B 2007;76:064120.
[107] Brenner DW. Phys Rev B 1990;42:9458–71.
[108] Horsfield A, Finnis M, Foulkes M, et al. Comput Mater Sci 2008;44:16–20.
[109] Shang SL, Wang WY, Wang Y, et al. J Phys: Condens Matter 2012;24:155402.
[110] Tschopp MA, McDowell DL. Int J Plast 2008;24:191–217.
[111] Lazar P, Chen XQ, Podloucky R. Phys Rev B 2009;80:012103.
[112] Zhang R, Legut D, Niewa R, Argon A, Veprek S. Phys Rev B 2010;82:104104.
[113] Zhang RF, Legut D, Wen XD, et al. Phys Rev B 2014;90:094115.
[114] Jhi SH, Louie SG, Cohen ML, Morris Jr JW. Phys Rev Lett 2001;87:075503.
[115] Zhang RF, Sheng SH, Veprek S. Scripta Mater 2013;68:913–6.
[116] Siegel DJ, Hector Jr LG, Adams JB. Acta Mater 2002;50:619–31.
[117] Friák M, Hickel T, Körmann F, et al. Steel Res Int 2011;82:86–100.
 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158 157

[118] Freeman AJ, Wu R. J Magn Magn Mater 1991;100:497–514.

[119] Schönberger U, Andersen OK, Methfessel M. Acta Metall Mater 1992;40:S1–S10.
[120] Hartford J. Phys Rev B 2000;61:2221–9.
[121] Lazar P, Redinger J, Podloucky R. Phys Rev B 2007;76:174112.
[122] Lazar P, Podloucky R. Phys Rev B 2008;78:104114.
[123] Zhang RF, Argon AS, Veprek S. Phys Rev Lett 2009;102:015503.
[124] Zhang RF, Argon AS, Veprek S. Phys Rev B 2009;79:245426.
[125] Marten T, Isaev E, Alling B, Hultman L, Abrikosov I. Phys Rev B 2010;81:212102.
[126] Ivashchenko VI, Veprek S, Turchi PEA, Shevchenko VI. Phys Rev B 2012;85:195403.
[127] Watanabe T, Tsurekawa S, Zhao X, Zuo L. The coming of grain boundary engineering in the 21st century. In: Haldar A,
Suwas S, Bhattacharjee D, editors. Microstructure and texture in steels. London: Springer; 2009.
[128] Painter GS, Averill FW. Phys Rev Lett 1987;58:234–7.
[129] Wu R, Freeman AJ, Olson GB. J Mater Res 1992;7:2403–11.
[130] Wu R, Freeman AJ, Olson GB. Science 1994;265:376380.
[131] Wu R, Freeman AJ, Olson GB. Phys Rev B 1994;50:75.
[132] Geng WT, Freeman AJ, Wu R, Olson GB. Phys Rev B 2000;62:6208.
[133] Lu G, Kioussis N, Wu R, Ciftan M. Phys Rev B 1999;59:891–8.
[134] Kang J, Glatzmaier G, Wei SH. Phys Rev Lett 2013;111:055502.
[135] Luo J, Cheng H, Asl KM, Kiely CJ, Harmer MP. Science 2011;333:1730–3.
[136] Turek I, Drchal V, Kudrnovský J, Šob M, Weinberger P. Electronic structure of disordered alloys, surfaces and
interfaces. Springer; 1997.
[137] Geng WT, Freeman AJ, Wu RQ. Phys Rev B 2001;63:064427.
[138] Čák M, Šob M, Hafner J. Phys Rev B 2008;78:054418.
[139] Všianská M, Šob M. Phys Rev B 2011;84:014418.
[140] Všianská M, Šob M. Prog Mater Sci 2011;56:817–40.
[141] Lejček P. Grain boundary segregation in metals. Heidelberg Dordrecht London New York: Springer; 2010.
[142] Lejček P, Šob M, Paidar V, Vitek V. Scripta Mater 2013;68:547–50.
[143] Lejček P, Zheng L, Hofmann S, Šob M. Entropy 2014;16:1462–83.
[144] Bhattacharya S, Kohyama M, Tanaka S, Shiihara Y. J Phys: Condens Matter 2014;26:355005.
[145] Jin H, Elfimov I, Militzer M. J Appl Phys 2014;115:093506.
[146] Zhang L, Zhang Y, Lu GH. J Phys: Condens Matter 2013;25:095001.
[147] Rajagopalan M, Tschopp MA, Solanki KN. J Mater 2014;66:129–38.
[148] Liu W, Han H, Ren C, et al. Comput Mater Sci 2015;96(Part A):374–8.
[149] Yamaguchi M, Shiga M, Kaburaki H. J Phys: Condens Matter 2004;16:3933.
[150] Lu GH, Deng S, Wang T, Kohyama M, Yamamoto R. Phys Rev B 2004;69:134106.
[151] Yamaguchi M, Shiga M, Kaburaki H. Science 2005;307:393–7.
[152] Schusteritsch G, Kaxiras E. Modell Simul Mater Sci Eng 2012;20:065007.
[153] Yuasa M, Mabuchi M. Phys Rev B 2010;82:094108.
[154] Sanyal S, Waghmare UV, Subramanian PR, Gigliotti MFX. Appl Phys Lett 2008;93:223113.
[155] Kohyama M. Philos Mag Lett 1999;79:659–72.
[156] Ochs T, Beck O, Elsässer C, Meyer B. Philos Mag A 2000;80:351–72.
[157] Zhang Y, Lu GH, Deng S, et al. Phys Rev B 2007;75:174101.
[158] Kart HH, Cagin T. Jamme 2008;30:117.
[159] Tian ZX, Yan JX, Xiao W, Geng WT. Phys Rev B 2009;79:144114.
[160] Janisch R, Ahmed N, Hartmaier A. Phys Rev B 2010;81:184108.
[161] Ogata S, Li J, Yip S. Phys Rev B 2005;71:224102.
[162] Kibey S, Liu JB, Johnson DD, Sehitoglu H. Appl Phys Lett 2007;91:181916.
[163] Lane NJ, Simak SI, Mikhaylushkin AS, Abrikosov IA, Hultman L, Barsoum MW. Phys Rev B 2011;84:184101.
[164] Šesták P, Černý M, Pokluda J. Intermetallics 2011;19:1567–72.
[165] Šesták P, Černý M, Pokluda J. Comput Mater Sci 2014;87:107–11.
[166] Huang X, Ackland GJ, Rabe KM. Nat Mater 2003;2:307–11.
[167] Liu AY, Cohen ML. Science 1989;245:841–2.
[168] Gilman JJ. Science 1993;261:1436–9.
[169] Chen XQ, Niu H, Li D, Li Y. Intermetallics 2011;19:1275–81.
[170] Ivanovskii AL. Prog Mater Sci 2012;57:184.
[171] Zhou S, Zhou X, Zhao Y. J Appl Phys 2008;104:053508.
[172] Griffith AA. Philos Trans Roy Soc A 1921;221:163–98.
[173] Polanyi M. Z Phys 1921;7:323.
[174] Gao F, He J, Wu E, et al. Phys Rev Lett 2003;91:015502.
[175] Gilman JJ. J Appl Phys 1975;46:5110–3.
[176] Gao F. Phys Rev B 2006;73:132104.
[177] Šimůnek A, Vackář J. Phys Rev Lett 2006;96:085501.
[178] Šimůnek A. Phys Rev B 2009;80:060103.
[179] Mulliken RS. J Chem Phys 1955;23. 1833, 1841, 2338, 2343.
[180] Tabor D. The hardness of metals. Oxford University Press; 2000.
[181] Argon A. Strengthening mechanisms in crystal plasticity. Oxford University Press; 2007.
[182] Cohen M. Phys Rev B 1985;32:7988–91.
[183] Zhang Y, Sun H, Chen C. Phys Rev B 2006;73:064109.
[184] Veprek S, Weidmann J, Glatz F. J Vac Sci Technol A 1995;13:2914–9.
[185] Brazhkin V, Dubrovinskaia N, Nicol M, et al. Nat Mater 2004;3:576–7.
158 J. Pokluda et al. / Progress in Materials Science 73 (2015) 127–158

[186] Chung HY, Weinberger MB, Levine JB, et al. Science 2007;316:436–9.
[187] Vítek V. Philos Mag 1968;18:773–86.
[188] Schoeck G, Ehmann J, Fahnle M. Philos Mag Lett 1998;78:289–95.
[189] Schoeck G. Mater Sci Eng A 2012;558:162–9.
[190] Paidar V, Čák M, Šob M, Inui H. Intermetallics 2015;58:43–9.
[191] Pugh SF. Philos Mag 1953;45:823.
[192] Johnson R. Phys Rev B 1988;37:3924.
[193] Kamran S. Phys Rev B 2009;79:024106.
[194] Nguyen-Manh D, Vitek V, Horsfield AP. Prog Mater Sci 2007;52:255–98.
[195] Pettifor DG. Mater Sci Technol 1992;8:345.
[196] Rice JR, Thomson R. Philos Mag 1974;29:567.
[197] Lin I. Mater Sci Lett 1983;2:295.
[198] McMahon Jr CJ, Vitek V. Acta Metall 1979;27:507–13.
[199] Jokl ML, Vitek V, McMahon Jr CJ. Acta Metall 1980;28:1479–88.
[200] Jokl ML, Vitek V, McMahon Jr CJ, Burgers P. Acta Metall 1989;37:87–97.
[201] Broek D. Elementary engineering fracture mechanics. Springer; 1982.
[202] Pokluda J, Šandera P. Kovove Mater 1993;31:23.
[203] Rice JR. J Mech Phys Solids 1992;40:239–71.
[204] Goken M, Kempf M. Z Metall 2001;92:1061–7.
[205] Chiu YL, Ngan AHW. Acta Mater 2002;50:1599–611.
[206] Vodenitcharova T, Zhang LC. Int J Solids Struct 2003;40:2989–98.
[207] Minor AM, Asif SS, Shan Z, et al. Nat Mater 2006;5:697–702.
[208] Nili H, Zadeh KK, Bhaskaran M, Sriram S. Prog Mater Sci 2013;58:1–29.
[209] Lilleodden ET, Zimmerman JA, Foiles SM, Nix WD. J Mech Phys Solids 2003;51:901–20.
[210] Schuh CA. Mater Today 2006;9:32–40.
[211] Hertz H. J Reine Angew Math 1881;92:156–71.
[212] Armstrong RW, Elban WL, Walley SM. Int J Mod Phys B 2013;27:1330004.
[213] Bahr DF, Kramer DE, Gerberich WW. Acta Mater 1998;46:3605–17.
[214] Zhang LC, Tanaka H. JSME Int J A 1999;31:546–59.
[215] Krenn CR, Roundy D, Cohen ML, Chrzan DC, Morris Jr JW. Phys Rev B 2002;65:134111.
[216] Lorenz D, Zeckzer A, Hilpert U, Grau P, Johansen H, Leipner HS. Phys Rev B 2003;67:172101.
[217] Zhu T, Li J, Vliet KJV, Ogata S, Yip S, Suresh S. J Mech Phys Solids 2004;52:691–724.
[218] Horníková J, Šandera P, Černý M, Pokluda J. Eng Fract Mech 2008;76:3755–62.
[219] Horníková J, Buršíková V, Šandera P, Černý M, Pokluda J. Acta Metall Slovaca 2007;13:65–9.
[220] Morris Jr JW, Krenn CR, Roundy D, Cohen ML. Mater Sci Eng A 2001;309–310:121–4.
[221] Šandera P, Pokluda J, Schöberl T, Horníková J, Černý M. Proc Mater Sci 2014;3:1111–6.
[222] Kresse G, Furthmüller J. Phys Rev B 1996;54:11169–86.
[223] Kresse G, Joubert D. Phys Rev B 1999;59:1758–75.
[224] Mehl MJ, Osburn JE, Papaconstantopoulos DA, Klein BM. Phys Rev B 1990;41:10311–23.
[225] Mehl MJ. Phys Rev B 1993;47:2493–500.
[226] Řehák P, Černý M, Šob M. Modell Simul Mater Sci Eng 2015 [submitted for publication].