A Framework For Micro-Macro Transitions in Periodic Particle Aggregates of Granular Materials

Comput. Methods Appl. Mech. Engrg.
193 (2004) 225–256

www.elsevier.com/locate/cma
A framework for micro–macro transitions in periodic

particle aggregates of granular materials
Christian Miehe *, Joachim Dettmar
Institut f€
ur Mechanik (Bauwesen), Lehrstuhl I, Universit€at Stuttgart, Pfaffenwaldring 7, 70550 Stuttgart, Germany
Received 18 October 2002; received in revised form 25 September 2003; accepted 8 October 2003
Abstract
The paper outlines an approach to the modelling of the overall macroscopic response of periodic granular materials
based on a numerically evaluated micro-to-macro transition. We consider a homogenized macro-continuum with lo-
cally attached microstructure, representing an aggregate of discrete solid granules which possibly come into contact.
The micromechanical material response is governed by the interaction of rigid particles based on a Coulomb-type
friction law. Central results of the paper are new definitions of homogenized stresses and power-type microhetero-
geneity conditions for periodic particle aggregates undergoing deformations at finite strains. Here, the critical point is a
precise formulation of periodicity conditions for center-displacement fluctuations and rotations of particles associated
with a well-defined boundary of the unit cell, inducing the anti-periodicity of fictitious support forces and couples. With
these definitions at hand, it is shown that both static and power-type approaches yield the identical symmetric macro-
stress tensor. The overall expressions appear in an extremely compact format in terms of a periodicity frame and three
support force resultants associated with the three faces of the unit cell. We show that these new representations of
overall properties of granular aggregates are identical to those of microstructures for continuous heterogeneous
materials, and can formally be obtained in a straightforward manner by a limit of microstresses to microforces on the
boundary of the microstructure. On the computational side we outline details of an implicit incremental update
algorithm that yields the proposed overall stresses of the particle aggregate in a quasi-static, macro-deformation-driven
process. That includes fictitious dynamic relaxation processes based on artificial damping mechanisms in order to
overcome the highly complex local instability problems of particle clusters within the quasi-static approach. The
performance of the proposed framework is documented by means of a representative set of numerical examples.
Ó 2003 Elsevier B.V. All rights reserved.
Keywords: Homogenization; Granular materials; Particle aggregates; Discrete element formulations; Dynamic relaxation techniques
1. Introduction
A granular medium is considered to be an aggregate of solid granules as visualized in Figs. 1 and 2,

which transmits loads through discrete contact forces between the particles. A microstructure consists of a
*
Corresponding author. Tel.: +49-711-685-6379.
E-mail address: cm@mechbau.uni-stuttgart.de (C. Miehe).
0045-7825/$ - see front matter Ó 2003 Elsevier B.V. All rights reserved.
doi:10.1016/j.cma.2003.10.004
226 C. Miehe, J. Dettmar / Comput. Methods Appl. Mech. Engrg. 193 (2004) 225–256
Fig. 1. The notion of a continuum with granular microstructure. Associated with a typical point X 2 B of a homogenized macro-
continuum B R3 is a microstructure V R3 that characterizes a cell of an aggregate of solid particles which possibly come into
contact. A distinct averaging of the contact microstresses s in V determines the overall macro-stress s at X 2 B.
representative volume of contiguous particles and a remaining portion of unfilled voids. The evolution of
this microstructure in a macro-deformation-driven process is characterized by formations of new contacts
between particles and breakings of old contacts when the assembly moves to a new equilibrium state
associated with a given macroscopic deformation mode. The literature provides two fundamentally dif-
ferent approaches to the micromechanically-based description of the overall behavior of such granular
aggregates: analytical modelling techniques and discrete modelling techniques. The first approach defines
analytical macroscopic constitutive equations where the micromechanical information enters in terms of
structural or fabric tensors obtained from analytical homogenization (see for example [8–10,13,
16,17,23,24,26,42]). The second approach introduces microscopic constitutive equations for the discrete
particle interactions and defines the macroscopic response in terms of a computational exploitation of
averaging theorems. We refer to the works Cundall and Strack [14], Cundall [15], Bardet and Proubet [5],
Bardet [3], Borja and Wren [7,45] and references cited therein. See also Moreau [34] and Herrmann and
Luding [18] for computational treatments of granular matter. Analytical modelling techniques often apply
rigorous simplifications to the microstructural behavior. In contrast, the discrete modelling techniques
provide a straightforward insight into the local response of complex microstructures, for example with
regard to the analysis of material instabilities. Following the recent works Miehe [27,28], Miehe et al. [29–
31] on non-linear homogenization methods for heterogeneous materials at finite strains, this paper focusses
on the discrete modelling technique and develops a consistent framework for micro-to-macro-transitions in
periodic granular materials.
As shown in Fig. 1, we consider a homogenized macro-continuum with locally attached microstructure,
representing an aggregate of discrete solid granules which possibly come into contact. The length-scales
lmicro and lmacro are assumed to differ by an order of magnitude lmicro =lmacro 1 that ensures that boundary-
layer effects are of low order. The assumed large scale-difference also motivates to neglect inertial forces
with respect to contact forces on the boundary of the particle aggregate. The constitutive micromechanical
material response of the granular material is governed by the interaction of rigid particles based on a
Coulomb-type elastic–plastic friction law. A micro-to-macro transition of such a granular aggregate is
concerned with the definition of homogenized overall properties such as the overall strains, stresses and
powers. As pointed out in the review article Bardet [3], the literature contains essentially two approaches to
the definition of overall stresses for a granular aggregate under quasi-static deformations. The first ap-
proach, conceptually outlined in Christoffersen et al. [13], states that the virtual work of the homogenized
continuum should be identical with the work performed by the granular aggregate. The obtained stress
tensor becomes asymmetric if moments induced by contact forces on the boundary of the aggregate are
taken into account. The second approach as outlined for example in Bardet [3] is a purely static definition.
Therein, contact moments are not considered and a symmetric expression for the overall stress, starting
C. Miehe, J. Dettmar / Comput. Methods Appl. Mech. Engrg. 193 (2004) 225–256 227
from the definition of the volume-averaged stress in a single particle, is obtained. See also Kruyt and
Rothenburg [23] and Bagi [2] for conceptually similar definitions. In a recent paper, Bardet and Vardou-
lakis [6] showed that the stress tensor in granular materials has an asymmetry even when particle contacts
do not transmit moments. However, they pointed out that this stress asymmetry decreases with the ratio
between surface and volume of the particle aggregate. When this ratio becomes large, the stress asymmetry
disappears with or without external contact moments. Hence, we conclude that the appearance of the
asymmetry in virtual-work-type stress definitions may be considered as a second-order boundary effect of the
representative particle volume.
We believe that the source of incompatibilities in definitions of overall stresses in granular materials is
based on different choices of boundary conditions on the microscale. As a consequence, key goal of this
paper is to develop a framework for finite-strain homogenization of granular materials that contains overall
stress definitions consistent with both static- as well as work-type approaches. A central difference to the
above cited works is a distinct definition of the aggregate volume and its boundary conditions that governs
the representation of the macro-stresses. Here, we consider a periodic unit cell of particles that consists of a
well-defined boundary-frame which drives the interior particles. In a deformation-driven process with a
prescribed macroscopic deformation gradient, we assume periodicity of both the center-displacement fluc-
tuations as well as the rotations of the rigid particles associated with the boundary frame. Consistent with
these kinematic assumptions, we obtain anti-periodicity of fictitious support forces and support couples
associated with the centers of the particles on the boundary. This well-defined periodic cell then allows a
consistent transition of overall definitions for continuous microstructures of heterogeneous materials as
outlined for example in Hill [19], Krawietz [22], Nemat-Nasser and Hori [38] and Miehe [27,28] to discrete
microstructures of particle aggregates. The overall stress appears in terms of the fictitious support forces
defined at the boundary of the particle aggregate and turns out to be symmetric due to the anti-periodicity of
the support couples. The new representations of the overall stresses and the microheterogeneity condition,
summarized in the boxed Eqs. (40), (43) and (53), appear in an extremely compact format in terms of three
support force resultants and a periodicity triad that characterizes the deformed macro-mode of the periodic
cell. As a result of the well-defined relationship between driving macro-deformation and resulting macro-
stress, spatial (Kirchhoff) and material (Piola) overall stress definitions simply differ through the current
and the reference periodicity frames. Recall from the classical homogenization theory that periodicity
conditions reflect exact results for periodic structures, while deformation and stress boundary conditions
yield upper and lower bounds of the aggregateÕs stiffness. The stiffness obtained for periodic boundary
conditions lies between these bounds. See for example Miehe [27,28] and Miehe et al. [30] for studies of
these stiffness bounds in the context of the homogenization of continuous microheterogeneous materials.
The bounds fall together if the microstructure becomes infinitely large. As a consequence of these bound
properties, we consider the proposed periodicity conditions to be an optimal choice even for real particle
aggregates of arbitrary structure, because they are expected to yield the best results for a moderate size of
the microstructure. This characteristic is important for inhomogeneous large-scale micro–macro-analyses
where the size of the aggregate plays a critical role with respect to computing time.
On the computational side we outline details of an implicit incremental update algorithm that yields the
proposed overall stresses of the particle aggregate in a quasi-static, macro-deformation-driven process. This
development includes the formulation of an elastic–plastic contact law for the interaction of the particles in
an implicit time-discrete setting where we restrict our considerations to plane problems with disc-shaped
particles. In order to overcome the highly complex local stability problems of particle clusters within the
quasi-static approach under consideration, we consider a fictitious dynamic relaxation process as con-
ceptually outlined in Underwood [43], Bardet and Proubet [4], Park [41] and Papadrakakis [40]. The dy-
namic relaxation technique based on artificial global and local damping mechanisms is outlined in an
implicit setting in context of a Newmark algorithm with high damping characteristics, allowing the
application of relatively large time steps.
Fig. 2. Periodic structure of materials in the deformed configuration for (a) continuous and (b) granular particle microstructures. The
periodic cell structure is defined by the current periodicity frame fpa ga¼1...d . The mechanical characteristics of the material are invariant
P
with respect to a translation of the cell along the vector p ¼ da¼1 ma pa with ma 2 N.
The paper is organized as follows. In Section 2 we at first define new representations of micro-to-macro
transitions for periodic continuous microstructures in terms of a periodicity frame and discrete support force
resultants. Section 3 then derives the new micro-to-macro transitions for granular particle assemblies in a
straightforward manner from those for continuous microstructures by a limit of microstresses to micro-
forces on the boundary of the microstructure. Section 4 considers details of a micromechanical elastic–
plastic model for the interaction of the particles where we restrict our considerations to plane problems with
disc-shaped granules. The implicit quasi-static equilibrium response of the aggregate of particles is treated
in Section 5 including its consistent linearization based on a two-particle discrete element formulation.
Then in Section 6 we comment on the implicit global solution algorithm with dynamic relaxation that
provides an equilibrium state of the particle aggregate within a macro-deformation-driven incremental
process. The performance of the proposed computational procedure is documented in Section 7 by means
of representative numerical examples which treat regular and irregular granular aggregates for given
macroscopic deformation modes.
2. Micro–macro-transition for continuous microstructures
Consider the homogenized macro-continuum B R3 depicted in Fig. 1. Associated with a typical point
X 2 B is a microstructure V R3 which characterizes a representative assembly of granular particles. Key
goal of the paper is the derivation of new compact statements of micro–macro-transitions which define the
overall stresses and the power of periodic granular aggregates in a precise analogy to the homogenization of
heterogeneous continua. To this end, this section first outlines definitions for the overall properties of
continuous periodic microstructures in terms of surface data in a spatial geometric setting with respect to the
current configuration. The treatment follows the recent work Miehe [28] and is then generalized in the
subsequent Section 3 for discrete periodic particle assemblies.
2.1. Definition of a continuous periodic microstructure
Consider a periodic microstructure of a material as depicted in Fig. 2. The periodicity of the material is
defined in terms of a Lagrangian positively orientated periodicity frame fPa ga¼1...d with origin at X 0 2 R3 .
Greek indices loop over the space dimension, i.e. d ¼ 2 for two-dimensional plane and d ¼ 3 for three-
dimensional cubic microstructures. The periodicity in the Lagrangian reference configuration is expressed
by possible translations of the microstructure along the vector
X
d
P¼ ma Pa with ma 2 N ð1Þ
a¼1
with integers fma ga¼1...d . The assumed characteristic structure of the material is invariant with respect to this
translation. As a consequence, it is sufficient to define the material properties of the medium in a small
domain V R3 , called the cell of a periodic medium. This cell can be repeated by the translation vector P.
As pointed out by Anthoine [1], neither the periodicity frame nor the associated cell of a periodic medium
are uniquely defined. However, the most natural cell is the parallelogram spanned by the vectors Pa of the
periodicity frame as indicated in Fig. 3. For a given reference frame, all possible cells have the same volume
in the reference configuration given by
jVj ¼ P1 ðP2 P3 Þ > 0: ð2Þ
As shown in Fig. 3(a), the boundary oV of a parallelogram-shaped cell V can always be divided
[
d
oV ¼ ðoVþ
a þ oVa Þ ð3Þ
a¼1
into d pairs foVþ

a ; oVa ga¼1...d of identical sides corresponding to each other through translations along the
periodicity vectors Pa . Two corresponding sides oVþ
a and oVa are said to be opposite. The distance of
þ þ
corresponding points X a 2 oVa and X a 2 oVa is given by the vectors of the periodicity frame, i.e.
Xþ
a X a ¼ Pa : ð4Þ
Outward normals at corresponding points X þ þ

a and X a are opposite, i.e. N a ¼ N a .
2.2. Periodic fluctuations and anti-periodic tractions
Let ut : V ! R3 denote the non-linear deformation map of the microstructure V R3 at time t 2 R. ut

maps material points X 2 V onto spatial points x ¼ ut ðXÞ of the current configuration ut ðVÞ as indicated
in Fig. 3. F :¼ rX ut ðXÞ is the associated deformation gradient with Jacobian J :¼ det½F > 0. The local
deformation of the microstructure is linked to the current macroscopic deformation gradient FðtÞ via the
ansatz
x ¼ ut ðXÞ ¼ FðtÞX þ wðX; tÞ in V ð5Þ
containing the linear part FX governed by the given macroscopic deformation gradient F and the super-
imposed fluctuation field w. The non-trivial periodicity of the deformation of the microstructure is de-
scribed by periodic fluctuations
w ¼ wþ on oVþ
a [ oVa ð6Þ
at opposite points X þ
a and X a (see Fig. 3(b) for a visualization). Furthermore, the fluctuation at the origin
X 0 of the periodicity frame and the d positions X a :¼ X 0 þ Pa defined by the periodicity vectors are set to
zero
w0 ¼ wa ¼ 0 at X 0 ; fX a ga¼1...d ; ð7Þ
which excludes rigid body motions of the microstructure. As a result of the conditions (6) and (7), all
corner points of the parallelogram-shaped cell are driven by the affine macroscopic mode FX as shown in
þ
Fig. 3. The distance of deformed corresponding points xþ
a 2 ut ðoVa Þ and xa 2 ut ðoVa Þ may then be
expressed by
xþ
a x a ¼ pa ð8Þ
in terms of the deformed Eulerian periodicity frame
pa ðtÞ ¼ FðtÞPa ; ð9Þ
(see also Fig. 2). These vectors define the volume of the deformed cell via
jvj :¼ det½FðtÞjVj ¼ p1 ðp2 p3 Þ > 0: ð10Þ
Now let r denote the Cauchy stress tensor and t :¼ rn the traction vector on the deformed boundary
ut ðoVÞ of the microstructure. Consistent with the periodicity of the fluctuation field is the anti-periodicity
of the traction field
t ¼ tþ on ut ðoVþ
a Þ [ ut ðoVa Þ ð11Þ
at opposite points xþ þ
a and xa with opposite outward normals na ¼ na .
2.3. Equilibrium conditions for the microstructure
Assume an equilibrium state of the continuous microstructure expressed by the Cauchy stresses
div½r ¼ 0 and rT r ¼ 0 in ut ðVÞ; ð12Þ
where r is assumed to be related to the metric g of the current configuration by some solid material model,
for example the elastic constitutive equation r ¼ r^ðg; F; XÞ. Integrating the conditions (12) over the domain
ut ðVÞ of the current configuration and applying the Gauss theorem yields the global conditions of equi-
librium
Z Z
t da ¼ 0 and ðt x x tÞ da ¼ 0: ð13Þ
ut ðoVÞ ut ðoVÞ
Taking into account the split of the surface of the periodic microstructure into opposite parts, we recast
(13) by using (11) and (8) into the form
X
d X
d
fa þf0 ¼ 0 and ðf a pa pa f a Þ ¼ 0 ð14Þ
a¼1 a¼1
in terms of the stress resultants ff a ga¼1...d defined by
Z
f a :¼ t da ð15Þ
ut ðoVþ
aÞ
Fig. 3. Reference and current configurations of the cells of a periodic material for (a,b) continuous and (c,d) granular particle
microstructures. The Lagrangian periodicity frame fPa ga¼1...d is mapped by the macroscopic deformation map F onto the Eulerian
periodicity frame fpa ga¼1...d . A fictitious equilibrium
P group of stress resultants ff a ga¼1...d of the faces oVþ
a located at fxa ga¼1...d defines
the macroscopic Kirchhoff macro-stress s ¼ da¼1 sym½f a pa =jVj.
associated with the d deformed surfaces ut ðoVþ a Þ. (14)1 defines the force f 0 . Using the identity
a ðb cÞ ¼ a ðb c c bÞ for the double cross product of vectors, we observe that (14)2 may be recast
Pd
into the form a¼1 pa f a ¼ 0. This is an equilibrium condition for couples with respect to the deformed
corner point x0 of the microstructure. Thus Eq. (14) characterize the stress resultants ff a ga¼1...d and f 0 as a
fictitious equilibrium group of forces located at the deformed corner points fxa ga¼1...d and x0 of the
microstructure as depicted in Fig. 3(b).
2.4. Definition of the macro-stresses
The overall Kirchhoff macro-stress associated with the microstructure under consideration is defined by
the volume average
Z
1
s :¼ s dV ; ð16Þ
jVj V
which for the equilibrium state (12) can be recast into the form
Z
1
s ¼ ðt x þ x tÞ da: ð17Þ
2jVj uðoVÞ
Taking into account the split of the surface of the periodic microstructure into opposite parts, we recast
(17) by using (11) and (8) into the form
1 X d
s ¼ ðf pa þ pa f a Þ ð18Þ
2jVj a¼1 a
in terms of the stress resultants of the d surfaces defined in (15). This representation in terms of the d
fictitious equilibrium group of forces ff a ga¼1...d located at the deformed corner points fxa ga¼1...d visualized
in Fig. 3(b) is considered to be the most compact representation of the overall macro-stress s of periodic
microstructures. 1
2.5. Hill–Mandel-type microheterogeneity condition
The so-called Hill–Mandel microheterogeneity condition demands the average of the microscopic work
to be equal to the macroscopic work (see for example [19,22,27,28]). For the representation with respect to
the Eulerian current configuration, we define this statement in the absolute form
Z
1
s : g ¼ s : g dV ð21Þ
jVj V
with s : g :¼ sab gab , which can be recast for the equilibrium state (12) into
Z
1
s : g ¼ t x da ð22Þ
jVj ut ðoVÞ
with t x :¼ ta xb gab . Insertion of ansatz (5) into the averaging theorem (22) gives the constraint
Z
1
t w da ¼ 0 ð23Þ
jVj ut ðoVÞ
on the fluctuation field w, that is satisfied by the periodicity (6) of the fluctuation field and the anti-periodicity
(11) of the traction field on the boundary of the deformed microstructure. Taking into account the split of the
surface of the microstructure into opposite parts, we recast (22) by using (11) and (8) into the form
1 X d
s : g ¼ f p : ð24Þ
jVj a¼1 a a
This representation of the Hill–Mandel-type microheterogeneity condition in terms of the stress resultants
ff a ga¼1...d and the deformed periodicity frame fpa ga¼1...d is consistent with the definition (18) of the macro-
stress. The Eulerian rate form of the microheterogeneity constraint (21) has the representation
Z
1 1 1
s : Lv g ¼ s : Lv g dV ð25Þ
2 jVj V 2
1
In what follows we prefer the Kirchhoff stress s as a stress measure of large-strain applications. Other overall macro-stresses are
then defined by standard geometric relationships, for example the Cauchy stress r the second Piola stress S or the first Piola (nominal)
stress defined by T
1 T T
:¼ det½F1 s; S :¼ F sF ; T :¼ gsF
r ð19Þ
with the Jacobian det½F :¼ jvj=jVj of the macroscopic deformation gradient FðtÞ. However, only r and T can be defined by averaging
expressions similar to (16). With the above introduced definitions for periodic microstructures, we obtain the representations
1 X d
1 X d
1 1 X d
¼
r f a pa ; S ¼ ðF f a Þ Pa ; T ¼ ðgf a Þ Pa ð20Þ
jvj a¼1 jVj a¼1 jVj a¼1
for homogenized stresses alternatively to (18).

and links the macro- and microscopic stress powers with respect to the volume of the microstructure in the
reference configuration. Here, the micro- and macroscopic rates of deformation tensors d :¼
Lv g ¼ sym½g F_ F 1 and d :¼ 12 Lv g ¼ sym½gF_ F are defined as Lie derivatives of the Eulerian metric
1 1
2
g ¼ gab with respect to the micro- and macroscopic deformations, respectively. In analogy to (22) we get the
alternative rate form
Z
1 1
s : Lv g ¼ t x_ da ð26Þ
2 jVj ut ðoVÞ
in terms of the power of the tractions t on the surface. Similar to expression (24), this can be recast due to
(11) and (8) into the expression
1 1 X d
s : Lv g ¼ f p_ ð27Þ
2 jVj a¼1 a a
It states that the macroscopic stress power multiplied by the volume jVj of the microstructure is identical to
the power of the fictitious stress resultants ff a ga¼1...d located at the corners of the microstructure. Com-
paring (25) and (27), this statement can be considered to be a balance of internal and external powers of the
microstructure. It can also be used as a starting point for the derivation of expression (18) for the macro-
stress alternative to the equilibrium considerations outlined above. 2
3. Micro–macro-transition for granular particle aggregates
Now we propose a consistent transition of the above outlined continuous definitions of overall prop-
erties to the case of discrete periodic particles aggregates as visualized in Figs. 1 and 2(b), obtained by a limit
of microstresses to microforces on the boundary of the microstructure. We show that the equilibrium state,
the overall macroscopic stresses and the microheterogeneity conditions of particle aggregates can be ex-
pressed by the boxed formulas (14), (18) and (27) in a form identical to the continuous formulation.
3.1. Definition of a periodic aggregate of particles
Consider the aggregate of particles depicted in Fig. 3(c). With the interior domain of the representative
volume V we associate the N particles Pp , p ¼ 1 . . . N . The center points X p 2 Pp of these particles are
assumed to be elements of the domain, i.e.
Xp 2 V for p ¼ 1 . . . N : ð29Þ
The boundary oV is associated with M particles Qq , q ¼ 1 . . . M, shaded in Fig. 3(c). The center points
X q 2 Qp of these particles are assumed to be elements of the surface, i.e.
X q 2 oV for q ¼ 1 . . . M: ð30Þ
The particles Qq , q ¼ 1 . . . M provide a frame for the macroscopic driving of the particle aggregate in the
interior of the representative volume. The periodicity of the aggregate is defined with respect to this frame
2
The first invariant and the macroscopic stress power with respect to the reference volume jVj of the periodic microstructure can be
expressed in terms of the alternative macro-stresses (19) by
1 1
I :¼ s : g ¼ S : C ¼ T : F and P :¼ s : Lv g ¼ S : C_ ¼ T : F_ ; ð28Þ
T 2 2
where C :¼ F gF is the macroscopic right Cauchy–Green tensor. These power expressions correspond with the stress expressions (20).
in complete analogy to the continuous formulation outlined in Section 2.1. It is defined in terms of a
Lagrangian positively orientated periodicity frame fPa ga¼1...d with origin at X 0 2 R3 . Translations by the
vector P defined in (1) leave the characteristics of the granular material invariant as shown in Fig. 2(b). The
periodicity frame defines a canonical cell of granular material with volume jVj defined in (2). According to
(3) the boundary oV of the parallelogram-shaped cell V is divided into d pairs foVþ
a ; oVa ga¼1...d of
identical sides corresponding to each other through translations along the periodicity vectors Pa . This
induces the decomposition of the M particles on the boundary into d pairs of equal opposite particles Qþ q and
Qq as shown in Fig. 3(c). The distance of corresponding center points of particles on the boundaries
Xþ þ
a 2 oVa and X a 2 oVa is given by the vectors of the periodicity frame as defined in (4).
3.2. Periodic fluctuations/rotations and anti-periodic forces/couples
The position of the center points of the particles on the boundary of the granular aggregate is assumed to
be linked to the current macroscopic deformation gradient FðtÞ according to the ansatz (5), i.e.
xq ðtÞ ¼ FðtÞX q þ wq ðtÞ for q ¼ 1 . . . M: ð31Þ
Thus the current position of each particle center is decomposed into a part affine to the given macro-
scopic deformation gradient F and a superimposed fluctuation. Assuming rigid particles, the deformation
map for particles on the boundary reads
xðX; tÞ ¼ FðtÞX q þ wq ðtÞ þ Qq ðtÞ½X X q for q ¼ 1 . . . M; ð32Þ
where the proper orthogonal tensor Qq ðtÞ 2 SOð3Þ describes the particle rotation. Then a non-trivial peri-
odicity of the deformation of the granular aggregate may be described by periodic fluctuations and rotations
w þ
q ¼ wq and Q þ
q ¼ Qq ð33Þ
of opposite particles with reference center positions X þ
q and X q , respectively. Furthermore, the fluctuation
at the origin X 0 of the periodicity frame and the d positions X a :¼ X 0 þ Pa defined by the periodicity
vectors are set to zero,
w0 ¼ wa ¼ 0 at X 0 ; fX a ga¼1...d ; ð34Þ
see Fig. 3(d), which excludes rigid body motions of the microstructure. As a result of conditions (33) and
(34), all corner points of the parallelogram-shaped cell are driven by the affine macroscopic mode FX. The
distance of deformed corresponding particles xþ
q and xq may then be expressed according to (8) in terms of
the deformed Eulerian periodicity frame fpa ga¼1...d defined in (9). Consistent with the periodicity (33) of the
fluctuations and rotations is the anti-periodicity of the particle support forces and couples on the boundary. As
indicated in Fig. 4(a), these fictitious support forces and couples act at the centers of the particles. We have
a þ
q ¼ aq and m þ
q ¼ mq ð35Þ
at opposite deformed center points xþ
q and xq of the particles on the boundary.
3.3. Equilibrium conditions for the microstructure
3.3.1. Equilibrium for particles in the interior domain

The particles Pp in the interior of the representative domain interact with the others by discrete contact
forces f cp at discrete contact points xcp on the deformed surfaces of the particles as shown in Fig. 4(c). We
consider each particle Pp to be a continuum itself for which the microequilibrium conditions (12) and (13)
hold. In order to get an equilibrium condition in terms of the discrete contact forces, we consider the limit
tda ! f cp at xcp and transform the continuous integrals (13) into the discrete sums
Fig. 4. Equilibrium states of a particle aggregate, (a) for particles on the boundary with
P anti-periodic support forces aq and support
couples mq , (b) for the whole particle aggregate with support force resultants f a :¼ q2oVþa aq and (c) for particles in the interior
domain. The fictitious equilibrium group of support force resultants ff a ga¼1...d located at fxa ga¼1...d define the Kirchhoff macro-stress
P
s ¼ da¼1 sym½f a ðxa x0 Þ=jVj.
X X
f cp ¼ 0 and ðf cp xcp xcp f cp Þ ¼ 0 for p ¼ 1 . . . N : ð36Þ
c2Acp c2Acp
c
P Ap isc the active cset of contact points associated with the particle Pp . (36)2 can be recast into the
Here,
form c2Acp ðxp xp Þ f p ¼ 0, where xp is the actual center point of the particle. Thus the above equations
represent the local equilibrium conditions for the forces and the couples for each particle Pp with respect to
the actual center points xp .
3.3.2. Equilibrium for particles on the boundary

The particles Qq on the boundary of the representative volume element interact with particles in the
interior of the domain by discrete contact forces f cq at contact points xcq on the deformed surfaces of the
particles. Furthermore, the driving of the particles is performed by the fictitious support forces aq and
couples mq at the deformed center points xq of the particles as visualized in Fig. 4(a). The discrete equi-
librium conditions read
X X
f cq ¼ aq and ðxcq xq Þ f cq ¼ mq for q ¼ 1 . . . M ð37Þ
c2Acq c2Acq
for the forces and the couples of each particle Qq with respect to the deformed center points xq . Acq is the
active set of contact points associated with the particle Qq .
3.3.3. Equilibrium of the particle aggregate

Note carefully that in (37) the contact forces f cq are considered to be ‘‘real forces’’ while the support
forces aq and support couples mq are ‘‘fictitious quantities’’. They provide the fictitious support of the finite-
sized periodic unit cell containing the aggregate of particles. These support forces can be considered as
Lagrangian multipliers associated with the periodicity constraints (33). Taking into account f cq ¼ f cp for
the contact point c of two particles Qq and Pp , we may consider (37) as statements which define the fictitious
support forces aq and couples mq . Due to definition (30), we consider the support forces acting on the
deformed surface of the representative volume. Thus, applying equilibrium conditions for the whole par-
ticle aggregate, we have
X
M X
M
aq ¼ 0 and ðxq aq þ mq Þ ¼ 0 ð38Þ
q¼1 q¼1
for the support forces and the couples depicted in Fig. 4(b). Taking into account the anti-periodicity (35)2
of the support couples, we may recast these equations into the form
X
M X
M
aq ¼ 0 and ðaq xq xq aq Þ ¼ 0: ð39Þ
q¼1 q¼1
Thus the support couples drop out and the above equations can formally be understood to be discrete
forms of the continuous equilibrium conditions (13) for the limit tda ! aq at xq . Furthermore, when taking
into account the split of the surface of the periodic microstructure into opposite parts, by using the anti-
periodicity (35)1 of the support forces we recast (39) into
X
d X
d
fa þf0 ¼ 0 and ðf a pa pa f a Þ ¼ 0 ð40Þ
a¼1 a¼1
in terms of resultant support forces ff a ga¼1...d defined by

X
f a :¼ aq ð41Þ
q2oVþ
a
for the d faces of the periodic unit cell. 3 Clearly, (41) is the discrete counterpart to (15). Thus the global
equilibrium of the particle aggregate can formally be expressed in a form identical to the continuous setting
(14) in terms of a fictitious equilibrium group of resultant support forces ff a ga¼1...d located at the deformed
corner points fxa ga¼1...d of the aggregate as shown in Figs. 3(d) and 4(b).
3.4. Definition of the macro-stresses
With the above observations at hand, the overall Kirchhoff macro-stress may be defined in a straight-
forward manner from (17) by considering the limit tda ! aq for the fictitious support forces on the
boundary of the particle aggregate, yielding the expression
1 X M
s ¼ ðaq xq þ xq aq Þ: ð42Þ
2jVj q¼1
Observe that the support couples do not have an influence on the macro-stress due to anti-periodicity
condition (35)2 and its consequence (39)2 . Taking into account the split of the surface of the periodic
microstructure into opposite parts, by using the anti-periodicity (35)1 of the support forces we recast (42)
into the form
1 X d
s ¼ ðf pa þ pa f a Þ ð43Þ
2jVj a¼1 a
identical to the continuous formulation (18) in terms of the resultant support forces of the d faces defined in
(41). This representation in terms of the d fictitious forces resultants ff a ga¼1...d located at the deformed
3
The resultant support forces ff a ga¼1...d of the d faces of the periodic unit cell are simply the sums of all support forces (Lagrange
multipliers) aq acting on that individual face. In this context, note carefully that each particle in the corner point of the aggregate is
assumed to be loaded with d separate Lagrangian link forces related to the d faces of the unit cell.
points fxa ga¼1...d visualized in Figs. 3(d) and 4(b) is considered to be the most compact representation of the
overall macro-stress s of periodic particle aggregates. The overall Cauchy and first Piola (nominal) stresses
and T of the granular aggregate appear in the form (20) identical to the continuous theory. Due to the
r
well-defined relationship between the driving deformation F and the resulting overall stress, the first Piola
stress T differs from the Kirchhoff stress s only through the material and spatial periodicity frames Pa and
pa , respectively.
An expression for the averaged stress similar to (42) in the small-strain context was derived by Bagi [2].
However, this author does not take into account couples at the boundary of the aggregate which we pointed
out to be critical for the modelling of periodicity. The obtained symmetry of the new Kirchhoff stress
definition (43) in the presence of fictitious couples on the boundary of precisely defined periodic aggregates
is considered to be an important result of this paper. It seems to clarify observations made in Bardet and
Vardoulakis [6] who for certain non-periodic boundary conditions characterized the stress asymmetry in
granular materials as a boundary effect. Again, let us emphasize that in our treatment the couples depicted
in Fig. 4(a) are fictitious quantities that enforce the rotational periodicity of the granules on the boundary of
the unit cell. Hence, their presence does not motivate micropolar macro-theories where couple stresses
appear as independent fields for which additional constitutive assumptions are needed as considered for
example in Kanatani [21], M€ uhlhaus and Vardoulakis [35], Chang and Liao [11] or Chang and Ma [12].
This is clearly underlined by the fact that these couples do not influence the obtained overall stress defi-
nitions (42) and (43).
With the equilibrium conditions (37)1 and (36)1 for the forces acting on the particles on the boundary
and the interior domain, we may recast the obtained macro-stress representation (43) into
2 3
1 4 XM X XN X
s ¼ f c xq þ f cp xp 5 : ð44Þ
jVj q¼1 c2Ac q p¼1 c2A c
q p
sym
This formulation can be expressed in terms of a sum over all contacts Ac in the granular aggregate
1 X c
s ¼ ½f xS;c þ f cM xM;c sym ; ð45Þ
jVj c2Ac S
where xS;c and xM;c are the center positions of two contacting particles S and M as visualized in Fig. 5.
Introducing the branch or fabric vector l c and the associated branch contact force f c via
l c :¼ xS;c xM;c and f c :¼ f cM ¼ f cS ; ð46Þ
we obtain the representation

1 X c
s ¼ ðf l c þ l c f c Þ: ð47Þ
2jVj c2Ac
This classical macro-stress definition in terms of the branch contact forces f c between the particles was
used for example in Christoffersen et al. [13], Kruyt and Rothenburg [23], Borja and Wren [7] and Bardet
[3]. It can be motivated by a weighted sum of the average stresses in a each particle of the aggregate
Z
1 X N
1 1 X c
s ¼ sp jVp j with sp :¼ s dV ¼ ðf xcp Þsym ; ð48Þ
jVj p¼1 jVp j Vp jVp j c2Ac p
p
(see for example [3,25]). This shows that our symmetric stress definition (43) for periodic aggregates is
consistent with the classical static definition.
3.5. Hill–Mandel-type microheterogeneity condition
Clearly, the anti-periodic couples associated with the periodic rotation on the boundary do not con-
tribute to the power of the aggregate. Hence, by the limit tda ! aq we obtain from (22) an analogy to HillÕs
averaging theorem in the absolute form
1 X M
s : g ¼ aq xq : ð49Þ
jVj q¼1
Insertion of ansatz (32) into the averaging theorem (49) gives the constraint
1 X M
aq w q ¼ 0 ð50Þ
jVj q¼1
on the fluctuations wq of the center points of the particles on the boundary. This is satisfied by the peri-
odicity (33)1 of the particle fluctuations and the anti-periodicity (35)1 of the particle support forces on the
boundary of the aggregate. Taking into account the split of the surface of the microstructure into opposite
parts, we recast (49) by using (35)1 into the form
1 X d
s : g ¼ f p ð51Þ
jVj a¼1 a a
identical to the continuous formulation (24) in terms of the resultant support forces f a of the d surfaces
defined in (41).
An Eulerian rate form of the microheterogeneity constraint for the particle assembly is formally obtained
from (26) by the limit tda ! aq for the fictitious support forces on the boundary of the microstructure
1 1 X M
s : Lv g ¼ aq x_ q ð52Þ
2 jVj q¼1
in terms of the power of the support forces aq on the boundary. Similar to expression (51), due to the anti-
periodicity condition (35)1 this can be recast into the form
1 1 X d
s : Lv g ¼ f p_ ð53Þ
2 jVj a¼1 a a
identical to the continuous formulation (27) in terms of the power of the resultant support forces f a defined
in (41).
A comparison of (25) and (53) states a balance of macroscopic and microscopic powers of the granular
aggregate. When inserting p_ a ¼ F_ F pa , this condition can also be used as a starting point for the deri-
1
vation of expression (43) for the macro-stress alternative to its derivation by equilibrium considerations
outlined above. We consider this compatibility between the static and the work-type definition of the
overall stress as an important result of this paper. It clarifies for periodic aggregates the incompatibilities
outlined in Bardet [3] and Bardet and Vardoulakis [6]. The result is obtained only for the above defined
fictitious support forces on the boundary of a geometrically well-defined periodic unit cell. Recall that the
definition of overall stresses in terms of boundary data is also consistent with the fundamental work Hill
[19] on non-linear homogenization of continuous media. Our developments show that the equilibrium state,
the overall macroscopic stresses and the microheterogeneity conditions can be expressed in a unified format
for periodic continuous and granular particle microstructures by the boxed formulas (40), (43) and (53). Basis
for these unified formulations is the above outlined precise definition of the geometry of a periodic
microstructure in terms of two sets of particles associated with the interior domain and the boundary of the
aggregate.
4. Micromechanical model for quasi-static particle interaction
In what follows we investigate a simple micromechanical model for a quasi-static interaction of two
particles. We successively point out contact–normal-force and frictional–tangential-force mechanisms by
restricting our investigations to plane problems of disc-shaped rigid particles. The deformation map of a
typical rigid particle may be written in the form
xðX; tÞ ¼ xp ðtÞ þ QðtÞ½X X p ð54Þ
consisting of the translation xp ðtÞ of the center of the particle and a superimposed rotation governed by the
proper orthogonal tensor QðtÞ 2 SOð3Þ. For plane problems, the rotation can be expressed in the form

cos #p ðtÞ sin #p ðtÞ
QðtÞ ¼ : ð55Þ
sin #p ðtÞ cos #p ðtÞ
Thus the particle kinematics (54) are described by the generalized position vector
T
d p ðtÞ :¼ xp ðtÞ #p ðtÞ 2 R3 ð56Þ
containing the center position vector and the particle rotation angle.
4.1. Elastic normal contact-force mechanism
A micromechanical model for the particle contact interaction considers two particles PM and PS referred
to as master and slave particles, respectively. The center positions of these particles are X M and X S in the
reference configuration and xM and xS in the current configuration, respectively (see Fig. 5). Associated
with the two particles M and S we define a penetration measure
ec :¼ ðRS þ RM Þ jl c j with c :¼ ðM; SÞ; ð57Þ
where RM and RS are the radii of the master and slave particles, respectively. The penetration measure serves
firstly as a local contact check, i.e. we define the current active set of particle contacts via
AðtÞ :¼ fc :¼ ðM; SÞjec ðtÞ > 0g: ð58Þ
Secondly, as a kinematic variable it enters a penalty-type constitutive equation for the computation of the
contact normal force P c between contacting particles M and S in the form
P c ¼ w0p ðec Þ for c 2 A: ð59Þ
wp is a constitutive potential function for the contact pressure which is assumed to be convex. Its derivative
w0p is monotonously increasing with w0p ð0Þ ¼ 0 and can be considered to be a (possibly non-linear) penalty
function, which approximatively enforces the constraint ec ¼ 0 in the case of particle contact. Specific
functions are discussed for example in Mindlin [32] and Mindlin and Deresiewicz [33].
4.2. Frictional–cohesive tangential contact-force mechanism
4.2.1. Continuous formulation

Assume the master and slave particles PM and PS to be in contact. In an infinitesimal time interval dt with
rotations d#M and d#S of the two particles, we observe the infinitesimal slip dcc ¼ RM ðd#M d#c Þ þ
Fig. 5. Deformation of disc-shaped rigid particles which come into contact. The contact problem is defined in terms of two particles, a
master p ¼ M and a slave particle p ¼ S. X M 2 VM and X S 2 VS denote the particle centers in the reference configuration and
xM 2 SM and xS 2 SS the particle centers in the current configuration. l c :¼ xS xM is the so-called branch vector.
RS ðd#S d#c Þ as depicted in Fig. 5. d#M and d#S are the independent rotations of the two particles and d#c
the rotation due to the displacements of the particle centroids. The differences d#M d#c and d#S d#c
then represent the incremental relative rotations of the particles. Based on this consideration, as a basic
kinematic variable we introduce the slip rate
c_ c ¼ RM ð#_ M #_ c Þ þ RS ð#_ S #_ c Þ for c 2 A: ð60Þ

2
The contribution due to the displacements of the particle centroids is given by #_ c ¼ e3 ðl c l_ c Þ=jl c j in
terms of the branch vector l and its rate l_ (see Fig. 5). Setting approximatively jl j RM þ RS constant for
c c c
the case of contact, from (60) we obtain the result
c_ c ¼ RM #_ M þ RS #_ S e3 ðl c l_ c Þ=ðRM þ RS Þ for c 2 A: ð61Þ
The current slip may then be defined by the integration of the slip rate
Z t
c
c ¼ c_ c dt for c 2 A ð62Þ
t0c
in the time interval ½t0c ; t where the two particles are in contact. We decompose the total slip into elastic and
plastic parts with
cce :¼ cc ccp for c 2 A: ð63Þ
The contact tangential force is then assumed to be determined by the constitutive expression
T c ¼ w0t ðcce Þ for c 2 A ð64Þ
in terms of a (convex) constitutive potential wt with w0t ð0Þ ¼ 0. This force is assumed to be bounded by a slip
criterion function of the Coulomb-type
/ðT c ; P c ; bÞ ¼ jT c j P c tan½q b 6 0 with b ¼ w0h ðaÞ: ð65Þ
Here, q is the friction angle and b a cohesion of the particle-to-particle contact. The latter is assumed to be
governed by a potential function wh in terms of the hardening variable a. The evolution of the tangential slip
in (63) and the cohesive hardening variable in (65) are assumed to be governed by
Tc
c_ cp ¼ k and a_ ¼ k; ð66Þ
jT c j
where k is determined by the Kuhn–Tucker-type loading–unloading conditions

k P 0; / 6 0; k/ ¼ 0: ð67Þ
This rounds off our constitutive model for the description of the frictional–cohesive tangential contact
mechanism. Observe that the micromechanical model is described by the three basic microstress potential
functions wp , wt , wh and the slip criterion function /. Specific forms of these functions are commented on in
Section 7.
4.2.2. Algorithmic formulation

In order to obtain time-discrete values of the tangential force T c in the case of active contact, we con-
struct an implicit integration algorithm for the elastic–plastic Coulomb-type model outlined above. To this
end, consider a time interval ½tn ; tnþ1 and assume all variables at time tn as known. Subsequently, all
variables without subscript are evaluated at time tnþ1 . An algorithm for the integration of the slip rate (61)
may yield the update of the total slip
cc ¼ ccn þ RM ð#M #Mn Þ þ RS ð#S #Sn Þ e3 l cn ðl c l cn Þ=ðRM þ RS Þ ð68Þ
for c 2 A. The incremental step is elastic for
/ðT c ; P c ; b Þ < 0 with T c :¼ w0t ðcc ccpn Þ and b :¼ w0h ðan Þ: ð69Þ
We then set T c ¼ T c for the tangential force and perform the updates ccp ¼ ccpn and a ¼ an . If (69) is vio-
lated, a fully implicit integration of the evolution equation (66) yields the updates
Tc
ccp ¼ ccpn þ c and a ¼ an þ c ð70Þ
jT c j
for the plastic slip and the hardening variable with increment c P 0. For given c, we evaluate the Coulomb
function / in (65) with tangential force from (64) and determine its linearization
o/
G :¼ ¼ w00t ðcc ccp Þ þ w00h ðaÞ ð71Þ
oc
that governs a Newton update of the incremental parameter in the form
c ( c þ G1 /: ð72Þ
The iteration cycle (70)–(72) is repeated until convergence is obtained in the sense j/j < tol. The algo-
rithmic tangent moduli govern the sensitivity of the tangential force T c with respect to a change of the total
slip cc . Starting with (64) and inserting the result occ c ¼ G1 w00t T t =jT c j obtained from the consistency con-
dition / ¼ 0 by the implicit function theorem, we derive the closed-form expression for the algorithmic
tangent modulus
oT c
Ctep :¼ ¼ w00t G1 w002
t : ð73Þ
occ
The underlined last term occurs only in the case of plastic loading with non-zero incremental parameter
c > 0.
Note that in the formulation the plastic slip ccp and the internal variable a appear as internal variables
which describes the history dependent material model of the particle contact. With regard to the updates
(68) and (70) we store the data base
Hcn :¼ fccpn ; an ; ccn ; xMn ; #Mn ; xSn ; #Sn g ð74Þ

Fig. 6. Plane two-particle contact element. (a) Branch mesh with two-noded finite elements representing a master and a slave particle
in contact. Definition of (b) discrete nodal positions and rotations and (c) discrete nodal forces, nodal couples and contact forces.
associated with the contact point c at time tn . Clearly, the current tangential slip is set to zero when the two
particles separate, i.e.
ccp ¼ cc ¼ 0 for c 62 A: ð75Þ
5. Implicit discrete element formulation for rigid particles
In this section we set up an implicit finite element formulation for the two-body contact problem of disc-
shaped rigid particles and derive a closed-form expression for the associated algorithmic tangent operator.
The contact element governs a branch of active contact of a currently deformed particle aggregate as
visualized in Fig. 6(a). The set of all current particle contacts is obtained by contact search algorithms which
check the penetration condition (58). The literature contains several possible search algorithms for discrete
element simulations which try to reduce the contact detection time, for example grid cell, adaptive grid
subdivision and binary tree methods. For an overview we refer to Williams and OÕConnor [44]. The effi-
ciency of several search algorithms was investigated in Munjiza et al. [36]. A no binary search algorithm for
bodies of similar size, with total detection time proportional to N was proposed by Munjiza and Andrews
[37].
5.1. Two-particle contact element for plane problems
Based on the constitutive expressions of the micromechanical model outlined above we introduce a
displacement-type contact element associated with the master and slave particles introduced in Fig. 5. This
element exists only in the case of contact. It then connects the centers of the master and slave particles by
the current branch vector l c as shown in Fig. 6(b). The discrete nodal positions and rotations, forces and
couples are introduced in Fig. 6(b)–(c). We consider the contact element position vector
T
d c :¼ ½ xM #M xS #S ; ð76Þ
which contains the center positions and rotations of the two particles. The initial positions at the beginning
of the process are d c0 ¼ ½ X M #M0 X S #S0 T . The associated element force vector
T
pc :¼ ½ f M mM fS mS ð77Þ
contains the contact element forces and couples at the centers of the two particles. The relationship between
the element force vector (78) and the displacement vector (76) is governed by the contact model described in
the section above. With the micromechanical forces P c and T c defined in (59) and (64) we obtain the
representation
2 c 3 2 c 3
P n T t
6 0 7 6 T c RM 7
c c c c c c
p ðd ; Hn Þ ¼ pp þ pt with pp :¼ 4 6 7 c
and pt :¼ 4 6 7 ð78Þ
P c n 5 T c t 5
0 T c RS
in terms of the current contact normal and tangential vectors
n :¼ l c =jl c j and t :¼ e3 n ð79Þ
at the contact point. Hcn
contains the history variables defined in (74) at time tn of a typical time interval
½tn ; tnþ1 . The equilibrium conditions (13) for the forces and couples acting on the particles yield the global
system of non-linear algebraic equations
pðd; staten Þ :¼ A pc ðd c ; Hcn Þ ¼ 0 with d :¼ A d c ; ð80Þ

c2A c2A
which are subject to the boundary conditions (33) and (34) of particle fluctuations and rotations. 4 Ac2A
denotes the element assembly operator for the branch mesh with currently active contacts. p is the global
force vector of the particle aggregate and in the deformation-driven process is considered to be a function of
the global position vector d of the particles and all
[
staten :¼ Hcn ð81Þ
c2A
history variables at time tn at the beginning of the time increment ½tn ; tnþ1 under consideration.
5.2. Algorithmic tangent stiffness operators
In view to a solution of the non-linear equations with Newton-type iterative solvers, we consider the
linearization of (80)
Lin p :¼ p þ kDd with k :¼ od pðd; staten Þ: ð82Þ
The global tangent matrix of the particle aggregate is assembled by
k :¼ A kc with kc :¼ od c pc ðd c ; Hcn Þ ð83Þ
c2A
from the contact element stiffness matrix kc which represents the derivative of the element force vector (78)
with respect to the element position vector (76). We denote kc as the consistent or algorithmic tangent
stiffness matrix associated with the contact element. It can be split up
kc ¼ kcp þ kct ð84Þ
4
These boundary conditions can be implemented either by a Lagrange multiplier method as outlined in Miehe [28] or directly by
prescribing periodic fluctuations/rotations as link constraints of a fluctuation/rotation-type finite element implementation. The latter
formulation considers the fluctuations of the particle centers as the primary displacement variables as outlined in detail for the
homogenization of continuous microstructures in Miehe et al. [29].
into a part associated with the normal contact force mechanism of Section 4.1 and a part associated with
the tangential force mechanism of Section 4.2.
5.2.1. Tangent of contact normal force mechanism

Taking the derivatives of (59) and (79)1 with respect to the position of the center of the master particle
yields
oxM P c ¼ w00p n and oxM n ¼ ð1 n nÞ=jl c j: ð85Þ
With these results and the relationship oxS P c ¼ oxM P c at hand, from (78) we derive the closed-form
representation
2 3
Ap 0 Ap 0
6 0 0 0 07
kcp ¼ 6
4 Ap 0 Ap 0 5
7 ð86Þ
0 0 0 0
of the tangent matrix associated with the normal force mechanism. Here, we introduced the abbreviation
!
w0p w0p
Ap :¼ w00p þ c n n c 1: ð87Þ
jl j jl j
5.2.2. Tangent of contact tangential force mechanism

Based on (68), taking the derivatives of (64) and (79)2 with respect to the position of the center of the
master particle yields
oxM T c ¼ Ctep d and oxM t ¼ ð~

1 þ t nÞ=jl c j ð88Þ
with the definitions

~ 0 1
d :¼ ðe3 l cn Þ=ðRM þ RS Þ and 1 :¼ : ð89Þ
1 0
The derivatives of the tangential force with respect to the rotations of the particle become
o#M T c ¼ Ctep RM and o#S T c ¼ Ctep RS : ð90Þ
Ctep
is the algorithmic tangent modulus of the elastic–plastic slip model defined in Eq. (73) of Section 4.2.
With these results and the relationship oxS T c ¼ oxM T c at hand, from (78) we derive the closed-form
representation
2 3
At Ctep RM t At Ctep RS t
6 Ctep RM d Ctep R2M Ctep RM d Ctep RM RS 7
kct ¼ 6
4 At
7 ð91Þ
ep
Ct RM t At Ctep RS t 5
Ctep RS d Ctep RM RS Ctep RS d Ctep R2S
of the tangent matrix associated with the tangential force mechanism. Here, we introduced the abbreviation
w0t ~
At :¼ Ctep d þ ð1 þ t nÞ: ð92Þ
jl c j
6. Dynamic regularization of the quasi-static response
An inherent problem of the above outlined quasi-static approach is the possible singularity of the global
tangent stiffness matrix (83) due to multiple stability problems in particle clusters and the loss of contacts of
individual particles or clusters of particles. An artificial dynamic relaxation analysis of the above described
quasi-static approach circumvents these problems (see for example [4,40,41,43]). In this section we consider
an implicit numerical model for the dynamic regularization of the quasi-static analysis of granular
assemblies, where both global and local artificial damping mechanisms are introduced.
6.1. Global damping mechanism and equations of motion
A dynamic generalization of the non-linear static equilibrium equation (80) assumes the form
md€ þ cd_ þ pðd; staten Þ ¼ 0 ð93Þ
in terms of the fictitious global mass matrix m and the fictitious global damping matrix c of the particle
aggregate. ð_
Þ :¼ dð
Þ=dT denotes the derivative with respect to a fictitious time associated with the
incremental driving step ½tn ; tnþ1 of the non-linear static equilibrium response. These two matrices are as-
sumed to be of simple diagonal form. The contributions of a typical particle appear in the form
mp :¼ diag½mp ; mp ; hp and cp :¼ ag mp ; ð94Þ
where mp represents a fictitious mass and hp a fictitious mass moment of inertia of the pth particle. For the
disc-shaped particles these are computed with a fititious density q . The damping matrix is assumed to be
proportional to the mass matrix by the parameter ag . The fictitious discrete initial-boundary value problem for
the generalized positions (56) of the particles of the aggregate is defined for the fictitious time T associated
with the incremental driving step ½tn ; tnþ1 . It consists of the dynamic system of Eq. (93), the initial conditions
dðT ¼ 0Þ ¼ d n and d_ ðT ¼ 0Þ ¼ d_ n ð95Þ
for the free particle deformation degrees and the boundary conditions (33) and (34) for particle fluctuations
and rotations. The discretization of the dynamic system (93) in a time step DT :¼ Tkþ1 Tk is performed by
a Newmark algorithm. Here, the acceleration and velocity at current time Tkþ1 are assumed to be
_
d€ ¼ ðd d~ Þ=ðbDT 2 Þ and d_ ¼ d~ þ cDT d€ ð96Þ
with predictors for velocity and position vectors
2
_ DT
d~ :¼ d_ k þ DT ð1 cÞd€ k d~ :¼ d k þ DT d_ k þ
and ð1 2bÞd€ k : ð97Þ
2
Insertion of (96) and (97) into the system (93) evaluated at time Tkþ1 yields the update formula for the
global generalized position vector
d ( d þ k1
eff reff ð98Þ
of the particle aggregate, obtained by some solver of linear equations, in terms of the effective residual and
tangent matrices
1 c
reff :¼ md€ cd_ pðd; staten Þ and keff ¼ mþ c þ k; ð99Þ
bDT 2 bDT
where k is the tangent matrix of the static equilibrium response defined in (83). The parameters b and c
determine the stability and accuracy of the Newmark-algorithm (see [20]). The algorithm is unconditionally
stable for 2b P c P 1=2. A widely used method for structural dynamic applications is the so-called average
acceleration method with b ¼ 1=4 and c ¼ 1=2. The discretization process of the equation of motion leads
to high frequency modes which may be representative for the behavior of the particle differential equations.
Thus it is necessary to have some form of algorithmic damping to remove the high frequency behavior of
the particle structure considered. High-frequency dissipation is obtained by c > 1=2 and an optimized
parameter b can be selected to maximize the high-frequency dissipation. The strongest possible high-fre-
quency dissipation is obtained by b ¼ 1 and c ¼ 3=2. Clearly, in the dynamic relaxation procedure under
consideration, we apply this strong global damping characteristics.
6.2. Local contact damping mechanism
A second kind of damping is referred to as contact damping. It essentially defines additional viscous
forces Pvc and Tvc at the contact points, which are superimposed onto the equilibrium forces defined in (59)
and (64) via
P c ( P c þ Pvc and T c ( T c þ Tvc : ð100Þ
The viscous forces are defined by the constitutive functions
Pvc ¼ w0pv ð_ec Þ and Tvc ¼ w0tv ð_cc Þ for c 2 A ð101Þ
in terms of the rate e_ c of the penetration defined in (57) and the slip rate c_ c defined in (61). The algorithmic
counterpart of (101) in a typical time step DT :¼ Tkþ1 Tk of the relaxation analysis uses the approximations
e_ c ¼ ðec eck Þ=DT and c_ c ¼ ðcc cck Þ=DT : ð102Þ
Taking these superimposed viscous response into account, we get the extended structures of the tangent
stiffness operators defined in Section 5.2. These extensions are obtained by the replacements
w0p ( w0p þ w0pv and w00p ( w00p þ w00pv =DT ð103Þ
in (87) of the contact normal stiffness operator and
w0t ( w0t þ w0tv and Ctep ( Ctep þ w00tv =DT ð104Þ
in (91) and (92) of the contact tangent stiffness operator.
7. Representative numerical examples
In this section we consider some characteristic examples of two-dimensional simulations of regular and
irregular particle aggregates. The first example visualizes the free motion of a heap of particles in order to
illustrate the performance of the proposed quasi-static particle interaction model. Then the next example
treats micro-to-macro transitions by illustrating the macroscopic material response of a regular package of
circular discs under a prescribed macroscopic deformation mode. Finally, the last example is concerned
with the micro-to-macro transition of an irregular assembly of particles. Here, the orientations of the fabric
are also discussed. For all subsequent model computations, all constitutive force-potential functions
introduced above are assumed to be quadratic. We set
cp c2 ct c2 h
wp ¼ e ; wt ¼ c ; w h ¼ a2 ð105Þ
2 2 e 2
in terms of the three material parameters cp , ct and h for the quasi-static response outlined in Section 4. In
connection with the angle of internal friction . in the Coulomb-type slip criterion function (65)
/ðT c ; P c ; bÞ ¼ jT c j P c tan½q b 6 0 ð106Þ
they govern the simple micromechanical response under consideration. Furthermore, for the potentials of
the artificial local damping mechanism outlined in Section 6 we assume
cp ct
wpv ¼ ac e_ c2 ; wtv ¼ ac c_ c2 ð107Þ
2 2 e
in terms of the local parameter ac . This parameter, in connection with the global parameters . and ag in
(94), governs the dynamic regularization of the quasi-static response. In all examples, the Newmark
algorithm is applied with high numerical frequency dissipation, characterized by the parameters c ¼ 3=2
and b ¼ 1.
7.1. Free motion of a heap of particles
We consider an irregular package of 1604 randomly generated circular discs. The radii of the discs are
chosen between 31.5 and 57 units. The initial geometry is defined by width/height 3808/3783 units. The
normal and tangential penalty parameters are cp ¼ ct ¼ 108 , the contact damping coefficient is ac ¼ 0:01,
the global damping parameter is ag ¼ 4:0, the internal friction angle q ¼ 30°, the hardening modulus
h ¼ 106 and the density is set to q ¼ 100:0 (Table 1). The initial configuration of the assembly is docu-
mented in Fig. 7(a). In this setup the displacements, velocities and accelerations of all particles are zero.
From this initial configuration the particles move downwardly due to body forces governed by the density
q ¼ 100 and acceleration g ¼ 10 in vertical direction. The assembly is bounded by a wall on the left side
and a bottom plate. This example demonstrates the performance of the implicit algorithm proposed for a
complex evolution of the particle interaction. It is characterized by a lot of contacts which form and break
during the deformation process. The simulation is carried out with a time step Dt ¼ 0:01. A sequence of
deformation states is depicted in Fig. 7(a)–(f) at times t ¼ 0, t ¼ 125, t ¼ 312:5, t ¼ 625, t ¼ 937:5 and
t ¼ 1250, respectively. Fig. 8 depicts the current configurations of the particles at time t ¼ 1250 with the
distributions of the contact normal forces P c , the so-called branch mesh. The width of the lines connecting
the centers of the particles reflect the magnitudes of the contact forces present in the respective contact
elements.
7.2. Combined compression–shear mode of a regular package
In this example the particle assembly is composed of 64 uniform circular particles with radii 50.1 units
giving a slight initial compression overlap of 0.2 units. The particles are initially arranged in a simple
rectangular package. Every second row of particles is shifted by 5°. The geometry of the assembly is defined
Table 1
Free motion test––model parameters
Normal stiffness cp ¼ 108
Tangential stiffness ct ¼ 108
Linear hardening h ¼ 106
Angle of internal friction . ¼ 30°
Contact damping ac ¼ 0:01
Global damping ag ¼ 4:0
Density of discs . ¼ 100:0
Time step length Dt ¼ 0:01
Fig. 7. Irregular package of 1604 circular disks under dead weight: (a) Initial configuration t ¼ 0. Deformed configurations at (b)
t ¼ 125, (c) t ¼ 312:5, (d) t ¼ 625. (e) t ¼ 937:5 and (f) final configuration t ¼ 1250. Clearly, the angle of internal friction of 30° in the
final state can be observed.
Fig. 8. Irregular package of 1604 circular discs under dead weight: Normal contact forces P c and at deformed configuration t ¼ 1250.
by width/height 700/689.36 units. Thus the reference configuration has a volume of 482552.0 units. The
initial configuration of the assembly is shown in Fig. 9(a). The normal and tangential stiffness parameters
Fig. 9. Regular package of 64 circular disks under combined compression–shear deformation mode at (a) F 12 ¼ 0:0, (b) F 12 ¼ 0:054,
(c) F 12 ¼ 0:064, (d) F 12 ¼ 0:200. (e) F 12 ¼ 0:232, (f) F 12 ¼ 0:3. In (b) and (c) the microstructures before and after the first sharp
decline of the r11 -stresses are shown (see Fig. 10(a)). The decline is caused by an upward movement of the fifth and sixth columns of
particles. In (d) and (e) the package before and after the sharp increase in stresses at F 12 0:21 is depicted. Here the two middle rows of
discs slide to the left. The final increase in stresses is denoted by the upper three rows of particles moving to the left until the densest
package in (f) is reached.
are cp ¼ ct ¼ 108 , the viscous damping coefficients for contact and global damping ac ¼ 0:0001 and
ag ¼ 40:0. The angle of internal friction is . ¼ 30°, the fictitious density of all discs set to . ¼ 100:0. Here
the hardening modulus is set to h ¼ 104 . All parameters used are assorted in Table 2. The combined
compression–shear mode imposed on the regular package of discs is characterized by the macroscopic
deformation gradient

0:001 0:01
F ¼1þk : ð108Þ
0:01 0:001
This deformation is applied in steps of Dk ¼ 0:1. In order to reach static equilibrium states after each
load-increasing time step, 49 relaxation steps Dt ¼ 0:1 are executed. Fig. 9 gives a sequence of deformed
configurations at macroscopic shear deformations (a) F 12 ¼ 0:0, (b) F 12 ¼ 0:054, (c) F 12 ¼ 0:064,
Table 2
Combined compression–shear test––model parameters
Linear hardening h ¼ 10000:0
Fig. 10. Regular package of 64 circular disks under combined compression–shear deformation mode: (a) Cauchy-stresses versus F 12 ,
(b) Cauchy-stresses versus t in the time interval [45, 70 s], with r11 : solid line, r22 : dashed line, r12 : short-dashed line. The letters a–f
refer to the deformation states depicted in Fig. 9.
(d) F 12 ¼ 0:20, (e) F 12 ¼ 0:232 and (f) F 12 ¼ 0:30. Fig. 9(b) and (c) show the deformation of the
microstructure before and after the sharp decline in r 11 -stresses at F 12 0:06 as shown in Fig. 10(a). Here
the fifth and sixth columns of particles form a cluster that suddenly slides upwardly giving a closer package.
In Fig. 9(d) and (e) the microstructures before and after the sharp increase in stresses at F 12 0:21 are
depicted. Clearly, the cluster formed by the two middle rows abruptly slides to the left again giving the
microstructure a closer package of discs. Finally, the last increase in stresses can be explained by a further
movement of the particle cluster composed of the upper three rows to the left until the densest package as
depicted in Fig. 9(f) is reached. From that point on the assembly of discs deformes homogeneously, the
discs begin to overlap equally. Evidently, the penalty parameters then cause the contact forces to strongly
increase. This increase finally results in a steady rise in stresses. Note that because the assembly is composed
of mono-disperse discs, the particle clusters as shown in Fig. 9(d) and (e) form. This phenomenon is ob-
served in experiments with particles of same size and shape. It does not happen in assemblies of particles
with different sizes and shapes. In Fig. 10(a) the normalized Cauchy stresses r =ct are plotted against the
shear deformation component F 12 . Fig. 10(b) then shows the components of the Cauchy stresses in the time
interval 45, 70 s. In this interval five load-increasing steps are executed. Clearly, within the relaxation after
each
P load-increasing step the stresses tend to stationary values. The sum of incremental plastic slips
c
c c against simulation time T is depicted in Fig. 11. Fig. 11(a) displays the sum of incremental plastic
c2A
slips for the whole simulation time interval. In addition, Fig. 11(b) shows the sum of incremental plastic
slips of the assembly for a particular time interval of 25 s. Comparing Fig. 11(a) and (b) reveals that in a
wide range of the whole processing time the sum of accumulated incremental plastic slips is nearly zero after
approximately half the relaxation time.
7.3. Compression–shear test of an irregular package
Now we analyze an irregular package that consists of 400 randomly arranged circular discs. The radii of
the particles lie between 35 and 55 units. The initial geometry of the package is defined by width/height
1717/1703 units giving a reference volume of the microstructure of 2924051.0 units. Note that 76 particles
build the boundary frame of the assembly. Normal and tangential stiffnesses are given as cp ¼ ct ¼ 108 , the
contact damping coefficient is ac ¼ 0:01, global damping is governed by ag ¼ 40:0 and the angle of internal
friction chosen as . ¼ 30°. The hardening modulus used is h ¼ 10 and the fictitious density of the discs set
Fig. P
11. Regular package of 64 circular disks under a combined compression–shear
P deformation mode. (a) Sum of incremental plastic
slips c2Ac cc versus process time, (b) sum of incremental plastic slips c2Ac cc versus process time in the time interval [45, 70 s].
Table 3
Shear test––model parameters
Linear hardening h ¼ 10
to . ¼ 50:0. Table 3 gives a summary of the parameters used in this example. The combined compression
and shear deformation is governed by the macroscopic deformation gradient

0:002 0:01
F ¼1þk : ð109Þ
0 0:002
The deformation process is characterized as follows: After each load-increasing step 19 relaxation steps
are executed to reach a static equilibrium state of the system. In the relaxation steps only time is increased,
not the load. Load is increased in increments of Dk ¼ 0:1, the time step lengths for both load- and
relaxation steps is chosen to be Dt ¼ 0:01. The initial configuration of the assembly of circular discs is
shown in Fig. 12(a). Fig. 12(b) and (c) illustrates the deformation states at F 12 ¼ 0:17 and F 12 ¼ 0:34. The
final configuration of the deformed microstructure at F 12 ¼ 0:5 is illustrated in Fig. 12(d). During defor-
mation many contacts break and others form due to steady alterations of the microstructure. Fig. 12(e) and
(f) show the actual branch meshes that consist of the two-noded elements discussed in above. These ele-
ments connect the centers of the particles currently in contact. The branch mesh shown here is given at the
deformation state F 12 ¼ 0:5. In Fig. 12(e) the widths of the mesh lines are associated with the magnitude of
the contact normal force P c acting between these particles, in Fig. 12(f) they are associated with the ones of
the contact tangential force T c at the end of the deformation process.
Fig. 12. Irregular package of 400 circular disks under shear deformation mode: (a) Initial configuration F 12 ¼ 0, and deformed
configurations at (b) F 12 ¼ 0:17, (c) F 12 ¼ 0:34 (d) F 12 ¼ 0:5. Branch meshes and magnitudes of discrete contact forces at final con-
figuration: (e) contact normal forces P c and (f) contact tangential forces T c .
Fig. 13. Irregular package of 256 circular disks under shear deformation mode: (a) Cauchy-stresses r11 (solid line) and r22 (dashed
line) versus F 12 , (b) Cauchy-stresses versus r12 versus F 12 .
Fig. 13 visualizes the Cauchy stresses against the shear deformation component F 12 . Fig. 13(a) illustrates
the r11 and r22 normal stresses. The shear stress r
12 curve as depicted in Fig. 13(b) is characterized by a
slightly less smooth behavior that is explained by the preservative rearrangement of the discrete granules
during the shear deformation process of the microstructure. Fig. 14 shows the rose diagrams of the dis-
tributions of the unit contact normals corresponding to deformation states at (a) F 12 ¼ 0:17, (b) F 12 ¼ 0:34
and (c) F 12 ¼ 0:5. The diagrams plotted represent the number of unit contact normals for each load step.
They reflect the sums of the orientations over ranges of ten degree (10°). The fabric develops strong bias as
Fig. 14. Rose diagrams of the distributions of the unit contact normals deformation states (a) F 12 ¼ 0:17, (b) F 12 ¼ 0:34 and
(c) F 12 ¼ 0:5.
Fig. 15. Assembly of eight deformed periodic unit cells of the irregular particle aggregate at a shear deformation state F 12 ¼ 0:5.
Obviously, the periodicity constraints (33) for the particle fluctuations w ¼ wþ and rotations # ¼ #þ on the boundaries of the unit
cells are satisfied.
the microstructure is sheared. As has been experimentally shown in Oda et al. [39], the evolution of the
fabric is closely related to the variation in the distribution of contact normals. Note that shearing produces
a biased distribution as the particles slide and roll on each other. Observe that this fabric strongly influences
the dilatancy in frictional granules (see for example [39] and the references therein). By further shearing the
assembly, the distribution of contact normals changes such that a greater concentration of the unit normals
along the direction of maximum principal compression is attained. Note that the average direction of these
contact normals tends to concentrate in the direction of the principal compressive force trajectories, which
results in the observed strong fabric anisotropy. In Fig. 15 the compatibility of the periodicity constraints
(33) is demonstrated by an assembly of eight deformed cells.
8. Conclusion
A framework for the modelling of the overall macroscopic response of periodic granular materials based
on micro–macro transitions has been presented. We considered a homogenized macro-continuum with
locally attached microstructure representing a periodic cell aggregate of discrete solid granules. Key con-
cern of the paper was the derivation of the new representations (40), (43) and (53) of the overall stresses and
the microheterogeneity condition for periodic particle aggregates of granular materials. These expressions
appear in an extremely compact format in terms of a periodicity frame and three resultant support forces
associated with the faces of the periodic unit cell. We showed that these new representations of overall
properties of granular aggregates are identical to those of continuous microstructures in terms of stress
resultants and may formally be obtained by a limit of microstresses to microforces on the boundary of the
microstructure. Basis for the derivation of these expressions was a precise formulation of periodicity
conditions for center-displacement-fluctuations and rotations of particles associated with a well-defined
boundary of the unit cell, which induced the anti-periodicity of fictitious support forces and couples. With
these definitions at hand, it was shown that both static and power-type approaches yield the identical
symmetric macro-stress tensor.
On the computational side we constructed an implicit update algorithm which provides equilibrium
states of the particle aggregate for quasi-static, macro-deformation-driven processes. This development
included the formulation of an elastic–plastic contact law for the interaction of disc-shaped particles in an
implicit time-discrete setting. Highly complex local stability problems of particle clusters within the quasi-
static approach were handled by an artificial dynamic relaxation technique based on several artificial
damping mechanisms. The proposed framework is suitable for a qualitative analysis of periodic cells of
granular particle aggregates. The numerical examples considered showed characteristic effects and diffi-
culties of micro-to-macro transitions for aggregates with a small and moderate number of disc-shaped
particles and simple contact interaction laws. Future research towards a realistic quantitative analysis of the
overall response of granular particle aggregates must incorporate more advanced non-linear contact laws,
generalized shapes of the particles and an extension of the micromechanical model to the three-dimensional
case.
Acknowledgements
We acknowledge the contribution of J€

org Schr€
oder who was involved in the initial phase of this work.
Support for this research was provided by the Deutsche Forschungsgemeinschaft (DFG) under grant
SFB404/A8 and C6.
References
[1] A. Anthoine, Derivation of in-plane characteristics of masonry through homogenization theory, Int. J. Solids Struct. 32 (1995)
137–163.
[2] K. Bagi, Stress and strain in granular assemblies, Mech. Mater. 22 (1996) 165–177.
[3] J.P. Bardet, Introduction to computational granular mechanics, in: B. Cambou (Ed.), Behaviour of Granular Materials, CISM
Courses and Lectures No. 385, Springer-Verlag, Wien, New York, 1998.
[4] J.P. Bardet, J. Proubet, Adaptive dynamic relaxation for statics of granular materials, Comput. Struct. 39 (1991) 221–229.
[5] J.P. Bardet, J. Proubet, A numerical investigation of the structure of persistent shear bands in granular materials, Geotechnique
41 (4) (1991) 599–613.
[6] J.P. Bardet, I. Vardoulakis, The asymmetry of stress in granular media, Int. J. Solids Struct. 38 (2001) 353–367.
[7] R.I. Borja, J.R. Wren, Micromechanics of granular media. Part I: Generation of overall constitutive equation for assemblies
of circular disks, Comput. Methods Appl. Mech. Engrg. 127 (1995) 13–36.
[8] R.J. Bathurst, L. Rothenburg, Micromechanical aspects of isotropic granular assemblies with linear contact interactions, J. Appl.
Mech. 55 (1988) 17–23.
[9] B. Cambou, P. Dubujet, F. Emeriault, F. Sidoroff, Homogenization for granular materials, Euro. J. Mech., A/Solids 14 (1995)
255–276.
[10] B. Cambou, Micromechanical approach in granular materials, in: B. Cambou (Ed.), Behaviour of Granular Materials, CISM
Courses and Lectures No. 385, Springer-Verlag, Wien, New York, 1998.
[11] C.S. Chang, C.L. Liao, Constitutive relation for a particulate medium with the effect of particle rotation, Int. J. Solids Struct. 26
(1990) 437–453.
[12] C.S. Chang, L. Ma, A micromechanical-based micropolar theory for deformation of granular solids, Int. J. Solids Struct. 28 (1991)
67–86.
[13] J. Christoffersen, M.M. Mehrabadi, S. Nemat-Nasser, A micromechanical description of granular material behavior, J. Appl.
Mech. 48 (1981) 339–344.
[14] P.A. Cundall, O.D.L. Strack, A discrete numerical model for granular assemblies, Geotechnique 29 (1) (1979) 47–65.
[15] P.A. Cundall, Numerical experiments on localization in frictional materials, Ing. Arch. 59 (1989) 148–159.
[16] H. Deresiewicz, Mechanics of granular matter, in: Advances in Applied Mechanics, vol. V, Academic Press, 1958, pp. 233–306.
[17] F. Emeriault, B. Cambou, A. Mahboudi, Homogenization for granular materials: non reversible behaviour, Mech. Cohesive–
Frictional Mater. 1 (1996) 199–218.
[18] H.J. Herrmann, S. Luding, Modeling granular media on the computer, Continuum Mech. Thermodyn. 10 (1998) 189–231.
[19] R. Hill, On constitutive macro-variables for heterogeneous solids at finite strains, in: Proceedings of the Royal Society of London,
Series A, vol. 326, 1972, pp. 131–147.
[20] T.J.R. Hughes, The Finite Element Method: Linear Static and Dynamic Finite Element Analysis, Prentice-Hall, Englewood Cliffs,
NJ, 1987.
[21] K. Kanatani, A micro-polar continuum theory for the flow of granular materials, Int. J. Engrg. Sci. 17 (4) (1979) 419–432.
[22] A. Krawietz, Materialtheorie. Mathematische Beschreibung des ph€anomenologischen thermomechanischen Verhaltens, Springer-
Verlag, Berlin, 1988.
[23] N.P. Kruyt, L. Rothenburg, Micromechanical definition of the strain tensor of granular materials, Trans. ASME 118 (1996) 706–
711.
[24] N.P. Kruyt, L. Rothenburg, Statistical theories for the elastic moduli of two-dimensional assemblies of granular materials, Int. J.
Engrg. Sci. 36 (1998) 1127–1142.
[25] M. L€ atzel, S. Luding, H.J. Herrmann, Macroscopic material properties from quasi-static, microscopic simulations of a two-
dimensional shear-cell, Granul. Matter 2 (2000) 123–135.
[26] M.M. Mehrabadi, B. Loret, S. Nemat-Nasser, Incremental constitutive relations for granular materials based on micromechanics,
J. Appl. Mech. 48 (1993) 339–344.
[27] C. Miehe, Strain-driven homogenization of inelastic microstructures and composites based on an incremental variational
formulation, Int. J. Numer. Methods Engrg. 55 (2002) 1285–1322.
[28] C. Miehe, Computational micro-to-macro transitions for discretized micro-structures of heterogeneous materials at large strains
based on the minimization of averaged incremental energy, Comput. Methods Appl. Mech. Engrg. 192 (2003) 559–591.
[29] C. Miehe, J. Schr€ oder, J. Schotte, Computational homogenization analysis in finite plasticity. Simulation of texture development
in polycrystalline materials, Comput. Methods Appl. Mech. Engrg. 171 (1999) 387–418.
[30] C. Miehe, J. Schotte, M. Lambrecht, Homogenization of inelastic solid materials at finite strains based on incremental
minimization principles. Application to the texture analysis of polycrystals, J. Mech. Phys. Solids 50 (2002) 2123–2167.
[31] C. Miehe, J. Schr€oder, M. Becker, Computational homogenization analysis in finite elasticity. Material and structural instabilities
on the micro- and macro-scales of periodic composites and their interaction, Comput. Methods Appl. Mech. Engrg. 191 (2002)
4971–5005.
[32] R.D. Mindlin, Compliance of elastic bodies in contact, J. Appl. Mech. 71 (1949) 259–268.
[33] R.D. Mindlin, H. Deresiewicz, Elastic spheres in contact under varying oblique forces, J. Appl. Mech. 20 (1953) 327–344.
[34] J.J. Moreau, Some numerical methods in multibody dynamics: Application to granular materials, Euro. J. Mech., A/Solids 13
(1994) 93–114.
[35] H.B. M€ uhlhaus, I. Vardoulakis, The thickness of shear bands in granular materials, Geotechnique 37 (1987) 271–283.
[36] A. Munjiza, D.R.J. Owen, N. Bicanic, A combined finite-discrete element method in transient dynamics of fracturing solids,
Engrg. Comput. 12 (1995) 145–175.
[37] A. Munjiza, K.R.F. Andrews, NBS contact detection algorithm for bodies of similar size, Int. J. Numer. Methods Engrg. 43
(1998) 131–149.
[38] S. Nemat-Nasser, M. Hori, Micromechanics: Overall properties of heterogeneous materials, North-Holland Series in Applied
Mathematics and Mechanics, vol. 37, 1993.
[39] M. Oda, J. Konishi, S. Nemat-Nasser, Experimental micromechanical evaluation of strength of granular materials: effects of
particle rolling, Mech. Mater. 1 (1982) 269–283.
[40] M. Papadrakakis, A method for the automatic evaluation of the dynamic relaxation parameters, Comput. Methods Appl. Mech.
Engrg. 25 (1981) 35–48.
[41] K.C. Park, A family of solution algorithms for nonlinear structural analysis based on relaxation equations, Int. J. Numer.
Methods Engrg. 18 (1982) 1337–1347.
[42] C. Thornton, The conditions of failure of a face-centered cubic array of uniform rigid spheres, Geotechnique 29 (1979) 441–
459.
[43] P. Underwood, Dynamic relaxation, in: T. Belytschko, T.J.R. Hughes (Eds.), Computational Methods for Transient Analysis,
Elsevier, 1983, pp. 245–265.
[44] J.R. Williams, R. OÕConnor, Discrete element simulation and the contact problem, Arch. Comput. Methods Engrg. 6 (4) (1999)
279–304.
[45] J.R. Wren, R.I. Borja, Micromechanics of granular media. Part II: Overall tangential moduli and localization model for periodic
assemblies of circular disks, Comput. Methods Appl. Mech. Engrg. 141 (1997) 221–246.

A Framework For Micro-Macro Transitions in Periodic Particle Aggregates of Granular Materials

Uploaded by

Copyright:

Available Formats

A Framework For Micro-Macro Transitions in Periodic Particle Aggregates of Granular Materials

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Framework For Micro-Macro Transitions in Periodic Particle Aggregates of Granular Materials

Uploaded by

Copyright:

Available Formats

Comput. Methods Appl. Mech. Engrg.

193 (2004) 225–256

A framework for micro–macro transitions in periodic

A granular medium is considered to be an aggregate of solid granules as visualized in Figs. 1 and 2,

2. Micro–macro-transition for continuous microstructures

2.1. Deﬁnition of a continuous periodic microstructure

jVj ¼ P1 ðP2 P3 Þ > 0: ð2Þ

into d pairs foVþ 

Outward normals at corresponding points X þ   þ

2.2. Periodic ﬂuctuations and anti-periodic tractions

Let ut : V ! R3 denote the non-linear deformation map of the microstructure V R3 at time t 2 R. ut

in terms of the deformed Eulerian periodicity frame

pa ðtÞ ¼ FðtÞPa ; ð9Þ

jvj :¼ det½FðtÞjVj ¼ p1 ðp2 p3 Þ > 0: ð10Þ

2.3. Equilibrium conditions for the microstructure

div½r ¼ 0 and rT  r ¼ 0 in ut ðVÞ; ð12Þ

in terms of the stress resultants ff a ga¼1...d deﬁned by

2.4. Deﬁnition of the macro-stresses

2.5. Hill–Mandel-type microheterogeneity condition

for homogenized stresses alternatively to (18).

3. Micro–macro-transition for granular particle aggregates

3.1. Deﬁnition of a periodic aggregate of particles

3.2. Periodic ﬂuctuations/rotations and anti-periodic forces/couples

3.3. Equilibrium conditions for the microstructure

3.3.1. Equilibrium for particles in the interior domain

3.3.2. Equilibrium for particles on the boundary

3.3.3. Equilibrium of the particle aggregate

in terms of resultant support forces ff a ga¼1...d deﬁned by

3.4. Deﬁnition of the macro-stresses

we obtain the representation

3.5. Hill–Mandel-type microheterogeneity condition

4. Micromechanical model for quasi-static particle interaction

4.1. Elastic normal contact-force mechanism

P c ¼ w0p ðec Þ for c 2 A: ð59Þ

4.2. Frictional–cohesive tangential contact-force mechanism

4.2.1. Continuous formulation

c_ c ¼ RM ð#_ M  #_ c Þ þ RS ð#_ S  #_ c Þ for c 2 A: ð60Þ

the case of contact, from (60) we obtain the result

c_ c ¼ RM #_ M þ RS #_ S  e3 ðl c l_ c Þ=ðRM þ RS Þ for c 2 A: ð61Þ

T c ¼ w0t ðcce Þ for c 2 A ð64Þ

where k is determined by the Kuhn–Tucker-type loading–unloading conditions

4.2.2. Algorithmic formulation

for c 2 A. The incremental step is elastic for

Hcn :¼ fccpn ; an ; ccn ; xMn ; #Mn ; xSn ; #Sn g ð74Þ

ccp ¼ cc ¼ 0 for c 62 A: ð75Þ

5. Implicit discrete element formulation for rigid particles

5.1. Two-particle contact element for plane problems

pðd; staten Þ :¼ A pc ðd c ; Hcn Þ ¼ 0 with d :¼ A d c ; ð80Þ

5.2. Algorithmic tangent stiﬀness operators

5.2.1. Tangent of contact normal force mechanism

oxM P c ¼ w00p n and oxM n ¼ ð1  n  nÞ=jl c j: ð85Þ

5.2.2. Tangent of contact tangential force mechanism

oxM T c ¼ Ctep d and oxM t ¼ ð~

with the deﬁnitions

6. Dynamic regularization of the quasi-static response

6.1. Global damping mechanism and equations of motion

md€ þ cd_ þ pðd; staten Þ ¼ 0 ð93Þ

6.2. Local contact damping mechanism

P c ( P c þ Pvc and T c ( T c þ Tvc : ð100Þ

into d pairs foVþ

Outward normals at corresponding points X þ þ

jvj :¼ det½FðtÞjVj ¼ p1 ðp2 p3 Þ > 0: ð10Þ

div½r ¼ 0 and rT r ¼ 0 in ut ðVÞ; ð12Þ

c_ c ¼ RM ð#_ M #_ c Þ þ RS ð#_ S #_ c Þ for c 2 A: ð60Þ

c_ c ¼ RM #_ M þ RS #_ S e3 ðl c l_ c Þ=ðRM þ RS Þ for c 2 A: ð61Þ

oxM P c ¼ w00p n and oxM n ¼ ð1 n nÞ=jl c j: ð85Þ