Granular Matter (2020) 22:83

https://doi.org/10.1007/s10035-020-01045-7

ORIGINAL PAPER

Representation of stress and strain in granular materials using

functions of direction
E. T. R. Dean1

Received: 5 May 2019 / Published online: 24 September 2020

© The Author(s) 2020

Abstract
This paper proposes a new way of describing effective stress in granular materials, in which stress is represented by a continuous
function of direction in physical space. The proposal provides a rigorous approach to the task of upscaling from particle mechan-
ics to continuum mechanics, but is simplified compared to a full discrete element analysis. It leads to an alternative framework
of stress–strain constitutive modelling of granular materials that in particular considers directional dependency. The continu-
ous function also contains more information that the corresponding tensor, and thereby provides space for storing information
about history and memory. A work-conjugate set of geometric rates representing strain-rates is calculated, and the fundamental
principles of local action, determinism, frame indifference, and rigid transformation indifference are shown to apply. A new
principle of freedom from tensor constraint is proposed. Existing thermo-mechanics of granular media is extended to apply for
the proposed functions, and a new method is described by which strain-rate equations can be used in large-deformations model-
ling. The new features are illustrated and explored using simple linear elastic models, producing new results for Poisson’s ratio
and elastic modulus. Ways of using the new framework to model elastoplasticity including critical states are also discussed.

Keywords Anisotropy · Continuum mechanics · Critical states · Discrete element method · Elasticity · Elasto-plasticity ·
Orientational averaging · Particle mechanics · Spin · Thermo-mechanics

Abbreviations F Deformation gradient tensor

FCM Functions-based constitutive model
Keyboard symbols
FILE Fully isotropic linear elastic
a Area
g Geometric factor associated with the bunching
a, b Quantities illustrating linearity (Eq. 12)
effect (dimensionless)
A,B,C (Subscripted) functions of the directions identi-
H Helmholtz free energy per unit particle volume
fied by the subscripts
(units of stress)
A–F Parameters in elasticity matrix (Appendices 5
I 3 × 3 identity tensor
and 6 and Table 2)
j, k Dimensionless constants
d Infinitesimal of
K Modulus (units of stress)
DEM discrete element method
𝐋c Branch vector for cth contact (units of length)
det() Determinant
M Number (dimensionless)
e Void ratio
𝐦𝜁 Unit material vector in direction ζ at the refer-
E Function whose directional average is zero
ence configuration (Appendix 3)
(Eq. 14)
n Element of unit vector
E Receipt (Eq. 58)
n Dimensionless constant (Appendix 5)
E Young’s modulus (Eq. 95)
𝐧𝜓 Unit vector in direction ψ
𝐅c Vector of inter-particle force at c-th contact
N Number of contacts
(Eq. 17)
p/ Mean normal effective stress
P–U Parameters in elasticity matrix (Table 2)
* E. T. R. Dean Q A general quantity or function
[email protected] Q Frame transformation matrix
1 r Radius (dimensions of length)
Caribbean Geotechnical Design Limited, Ammanford, UK

13
Vol.:(0123456789)
83 Page 2 of 25 E. T. R. Dean

𝐫− Value of a vector in the {x,y,z} coordinate q Deviatoric; a function deviator (Eq. 15) is not
frame (element dimensions of length) the same as a tensor deviator
sym(𝐗) The symmetric part of the tensor X r Radial in the triaxial cell
t Time REP Representative
TCM Tensor-based constitutive model rs One of xx, xy, xz, yx, yy, yz, zx, zy, or zz
TILE Transversely isotropic linear elastic vol Volumetric
tr() Trace operator y At yield
u Pore pressure β Function of direction β in physical space
uic Uniaxial stress tensor associated with the ith ψ Function of direction ψ in physical space
principal direction of the stress tensor associ-
Superscripts
ated with the cth inter-particle contact
* Inner
V (Un-subscripted) specific volume
/ Effective
(dimensionless)
T Transpose
V (Subscripted with a direction) directional spe-
−1 Inverse
cific volume (dimensionless)
VREP Representative small volume (units of volume) Other
W Work per unit particle volume ⟨…⟩ Directional average (Eq. 6)
x,y,z Cartesian coordinates (units of length) ⊗ Dyadic product; the dyadic product of two
column vectors 𝐚 and 𝐛 is the matrix formed by
Greek symbols
post-multiplying 𝐚 by the transpose of 𝐛 , The
α Spreading coefficient (dimensionless)
elements of 𝐚 ⊗ 𝐛 are the products ­aibj
β Direction in physical space
⋯|C Evaluated with C constant (if C represents a
γ Small engineering strain (typically expressed
variable) or according to condition C (if C is an
as an angle)
equation)
Γ Locator constant (dimensionless)
→0 Tends to zero
δψ Second-order tensor variations over the patch
→ And based on the information calculated so far,
for direction ψ (Eq. 18)
the next step is to calculate (Eqs. 35, 36, and
δψ Function equal to a non-zero constant on patch
46)
ψ but zero for all other patches (Eq. 44)
ε Small strain Conventions and notes
𝛆̇ Strain rate tensor (Eq. 26) Notation Matrices, and tensors expressible as matrices,
ζ Direction in physical space are denoted using bold type. Vectors are bold,
θ Polar angle (Fig. 1) underlined. Elements of these are in nor-
θ Small rotation (Eq. 26) mal type, subscripted, for example n𝜓x is the
𝛉̇ Rotation rate tensor (Eq. 26) x-component of the unit vector 𝐧𝜓 . Effective
μ Poisson’s ratio stresses are denoted using the prime symbol,
σ Stress (force per unit area in the current which is also used for functions and param-
configuration) eters that relate to effective stress. However,
ϕ Azimuthal angle (Fig. 1) un-primed symbols may be used where context
ψ Direction in physical space clearly indicates an effective parameter.
ω Direction in physical space
Ω Solid angle (units of steradians) Compressive principal stress and principal
strains and strain-rates are taken positive. Work
Subscripts
is taken positive when done on a material.
a Axial in the triaxial cell
Energy and work are both expressed per unit
alt Expressed in the alternate coordinate frame
particle volume, and therefore have the same
c Contact identification number
units as stress. Rates are denoted using an over-
d Dissipated
dot, equivalent to differentiation with respect to
e Elastic
time t, so that:
ep Elasto-plastic
o Initial, or reference
∫
dq
q̇ = , q= qdt
̇
dt

13
Representation of stress and strain in granular materials using functions of direction Page 3 of 25 83

for a given q (which may also be a function of other relate effective stress to inter-granular forces. Some limita-
variables, such as direction ψ). An over-bar with an tions have been explored by Singh and Wallender [66] and
over-dot denotes a rate of change of the expression others.
beneath the bar, so that: Effective stress “at a point” is a local concept, though
. referring to spatially distributed effects at microscopic scale
d(ab)
ab = in geomaterials. At macroscopic scale, let F represent the
dt deformation gradient tensor from a reference configuration
to a current configuration of a representative elementary
A special case occurs in Eq. 50 in which a small volume of granular material [70]. Then the rate W ′ of effec-
solid angle 𝛿𝛺𝜓 is differentiated with respect to tive work done on the soil per unit particle volume is, in the
time, with the following notation: absence of fluid flow:
. ( )
d ̇ −1
W � = −Vtr 𝛔�T 𝐅.𝐅 (2)
𝛿𝛺𝜓 = (𝛿𝛺𝜓 )
dt
The negative sign is used when taking compressive stress
as positive and work as positive when done on the material.
A quantity y is said to be a function of another V = 1+e is the specific volume, equal to the ratio of macro-
quantity x if a single value of y is associated with scopic volume to particle volume, e is the void ratio, and tr is
each possible value of x. The association may be the trace operator. Houlsby [39] discusses the complicating
determined by an algebraic equation, for instance, effects of fluid flow.
or in other ways, such as by a stress–strain his- The new proposals made herein are intended to provide
tory. Quantities with subscript ψ or θ, such as 𝛼𝜓′ new connections between the microscopic and macroscopic
or 𝛼𝜃′ , represent functions of direction ψ or polar scales. Note however that the new proposals involve some
angle θ respectively, while the corresponding un- simplifications and therefore that the new connections will
subscripted quantity 𝛼 would be independent of not be as explicit as those available in either the full discrete
direction. element method (DEM) or in a macroscopic model deduced
solely from particle mechanics. This therefore represents a
different approach to the ones taken by Jiang et al. [42], Liu
1 Introduction and Chang [47],Wu et al. [82] and others who propose new
continuum models based on results obtained with these more
This paper proposes new continuum descriptions of stress, fundamental methods.
strain, and related quantities that are intermediate in detail The present starting point is a proof, outlined by
between tensor mechanics and particle mechanics. The O’Sullivan [54] and attributed by her to several previous
descriptions help to bridge the gap between these disci- independent authors. The proof is widely accepted in DEM
plines, and provide new ways to use information from both modelling and states that the effective stress in a dry granu-
in the construction of continuum constitutive models. lar matrix can be expressed in terms of the statistics of inter-
Cauchy’s [13] concept and equations for continuum particle forces in one of the following ways (differing for
mechanical stress assumes that effects of granularity become different researchers):
smoothed out if large numbers of particles are considered.
This assumption serves well for linear elastic materials, but 1 ∑
N
1 ∑
N

does not appear to have been formally proved for more com- 𝛔� = 𝐅c ⊗ 𝐋c = �
𝐋c ⊗ 𝐅c (3)
VREP c=1
VREP c=1
plex behaviours. Terzaghi [72, 73] proposed a first modifica-
tion, the now well-established Principle of Effective Stress, where VREP and VREP ′
are measures of macroscopic volume
which implies that the symmetric stress tensor 𝝈 that is enclosing a representative group of particles, 𝐅c is the inter-
′

effective in terms of the stress–strain–strength behaviours particle force at the cth contact, ⊗ is the dyadic product, 𝐋c
of fully saturated, uncemented granular soils is related to is the inter-particle branch vector for that contact, and N is
the total (Cauchy) stress tensor σ by: the number of contacts in the volume. A dyadic product is
not in general symmetric, but the sum of many such products
𝛔� = 𝛔 − uI (1)
can be. Each contact is counted once.
where u is the pore fluid stress, and I is the identity tensor. The proposals herein are expected to be particularly use-
This principle is also used for sedimentary rocks [40, 85]. ful in resolving two modelling problems for soils. One is
Explanations for its success include those of Bishop [3], that of induced anisotropy, defined as anisotropy induced
Skempton [67], Mitchell and Soga [51], and others, who by strain [12]. All natural soils appear to be anisotropic,

13
83 Page 4 of 25 E. T. R. Dean

with most being transversely anisotropic in situ, due partly Figure 1 illustrates some concepts associated with physi-
to vertical deposition and partly to a subsequent history of cal direction. A direction emanating from a material point
one-dimensional vertical compression [31, 45, 51]. More M may be represented by an arrow, or by a label ψ, or by
complex forms of anisotropy can develop with different dep- a unit vector 𝐧𝜓 , or by a point P on a sphere centred on M.
ositional and/or stress–strain histories [24]. Anisotropy and The unit vector may be represented in a variety of external
fabric evolution are active topics in granular matter research, coordinate systems, including a cartesian system {x,y,z} or
see for example Cottechia et al. [20], Papadimitriou et al. a spherical system {r,θ,ϕ}:
[56], Chen et al. [15], Vijayan et al. [78], Wang et al. [79,
80], Zhao and Kruyt [84] and others. ⎡ n𝜓x ⎤ ⎡ sin 𝜃 cos 𝜙 ⎤
The second is “plastic spin”, described by Dafalias and 𝐧𝜓 = ⎢ n𝜓y ⎥ = ⎢ sin 𝜃 sin 𝜙 ⎥ (4)
⎢ ⎥ ⎢ ⎥
Aifantis [23] and Dafalias [22] for soils and by Soare [68] ⎣ n𝜓z ⎦ ⎣ cos 𝜃 ⎦
for crystalline metals. Dafalias [21] argued that plastic spin
where the polar angle 𝜃 runs from 0 to π radians, and the
is a “necessary constitutive ingredient” for large deformation
azimuthal angle ϕ runs from 0 to 2π radians. The set of all
anisotropic elasto-plasticity with tensorial internal variables.
possible directions may also be represented by the set of all
Soare [68] interpreted plastic spin as the rate of rotation of
points on the sphere.
a material frame, relative to the material itself, where the
Let Q𝜓 be a function of direction ψ, such that for every ψ
material frame is defined by axes that characterize a yield
there is a value of the quantity Q denoted as Q𝜓 . This value
surface.
may be a scalar, or a vector (such as the unit vector 𝐧𝜓 ).
The following text starts with a description of the notation
It may be a tensor if the tensor can be associated in some
used herein for the familiar process of orientational averag-
way with direction, and an example used later is the tensor
ing. This process is then applied in a new way to the above
𝐧𝜓 ⊗ 𝐧𝜓 where ⊗ is the dyadic (outer) product. This tensor
equation, resulting in a new relation between inter-particle
can represent a unit uniaxial stress in the ψ direction, with:
point forces and effective stress. The familiar algebra for the
deformation gradient is then used to deduce new expres- 2
⎡ n𝜓x n𝜓x n𝜓y n𝜓x n𝜓z ⎤
sions for strain rate and work rate in granular media. Some 𝐧𝜓 ⊗ 𝐧𝜓 = 𝐧𝜓 𝐧T𝜓 = ⎢ n𝜓y n𝜓x n2𝜓y n𝜓y n𝜓z ⎥ (5)
general implications for modelling are then considered. ⎢ ⎥
⎣ n𝜓z n𝜓x n𝜓z n𝜓y n2𝜓z ⎦
Finally, some simple constitutive assumptions are explored
that reveal new information about simple behaviours, includ- The function Q𝜓 may alternatively be determined by some
ing about the origins of Poisson’s ratio for elastic granular different process, such as by a stress–strain ⟨history.
materials. Application to elastoplastic models and critical ⟩ Its aver-
age over all directions ψ is here denoted as Q𝜓 and could
states is also discussed.

2 Proposed functional representations

2.1 Directional notation for orientational averaging

The idea of averaging quantities over physical directions has

a long history, and is part of the process of “homogeniza-
tion” or “upscaling” by which micro-scale features can be
averaged to create macroscale results. Examples for soils
include Pande and Sharma’s [55] multi-laminate model in
which macroscopic constitutive behaviour is constructed
from simple behaviours that are individually associated
with physical direction. Faria [29] discusses applications in
mixture theory, and Altenbach et al. [1] discuss orientation
averaging for fiber suspensions. Göodert and Hutter [30] use
orientational averaging in their model of induced anisotropy
in ice shields. Yimsiri and Soga [83] upscale their proposals
for contact mechanics using a spherical integration method.
Blumenfeld [4] proposed a specialised upscaling method to
solve a problem related to fabric in isostaticity theory.
Fig. 1  Notation and concepts associated with physical direction

13
Representation of stress and strain in granular materials using functions of direction Page 5 of 25 83

⟨ ⟩ ⟨ ⟩ ⟨ ⟩
be calculated by integrating over the surface of the sphere aQ1𝜓 + bQ2𝜓 = a Q1𝜓 + b Q2𝜓 (12)
and dividing by the surface area a:
In the constitutive context, a and b may be variables that
⟨ ⟩ da𝜓 d𝛺𝜓 change with time. A directional average is not itself a func-
∫ ∫
Q𝜓 = Q𝜓 = Q𝜓 (6)
a 4𝜋 tion of direction, and a useful result later will be that:
⟨ ⟨ ⟩⟩ ⟨ ⟩⟨ ⟩ ⟨⟨ ⟩ ⟩
where da𝜓 is the area of an infinitesimal patch on the sphere A𝜓 B𝜓 = A𝜓 B𝜓 = A𝜓 B𝜓 (13)
centred on the point P representing the direction ψ, and d𝛺𝜓
is the solid angle in steradians that is subtended by the patch. where the average inside an averaging operator is done first.
The solid angle is the area of the patch divided by the square One might call this a “swap property” since the change from
of the radius r𝜓 to the centre of the patch, with r𝜓 = 1 unit for left to right involves swapping the position of the inner aver-
a sphere of unit radius: aging operator. Another property follows from the fact that
there are many functions of direction whose average is zero,
da𝜓 3dv𝜓 for example cos 𝜃 . If two functions have the same average,
d𝛺𝜓 = = (7)
r𝜓2 r𝜓3 their difference is a function whose average is zero:
⟨ ⟩ ⟨ ⟩
where dv𝜓 is the volume of a cone subtended at the centre by If Q1𝜓 = Q2𝜓 , then Q1𝜓 = Q2𝜓 + E𝜓 (14)
the patch. The full spherical surface subtends a solid angle ⟨ ⟩
of 4π steradians. where E𝜓 is some function with E𝜓 = 0. One might call
Equation 6 is just a definition of a directional average this the “extraction property”, since it allows a directional
in three dimensions. Some details are nevertheless worth equation to be extracted from an equality of averages.
exploring. The area of a rectangular patch illustrated in In tensor mechanics an invariant is a quantity whose value
Fig. 1 is the product of its polar dimension and its circum- does not depend⟨ ⟩ on physical direction or coordinate frame.
ferential dimension. This product is sin 𝜃d𝜙d𝜃 for a unit The average Q𝜓 is a first invariant of the function Q𝜓 . But
radius, so: a directional function can also be associated with an infinite
number of invariants, for example the average of its mth
⟨ ⟩
2𝜋 𝜋
power is invariant, for any given m. The variance Q2q of Q𝜓 is:
4𝜋 ∫ ∫ 𝜓
1
Q𝜓 = Q sin 𝜃d𝜃d𝜙 (8) ⟨( ⟨ ⟩)2 ⟩ ⟨ 2 ⟩ ⟨ ⟩2
𝜙=0 𝜃=0
Q2q = Q𝜓 − Q𝜓 = Q𝜓 − Q𝜓 (15)

Using this, the following results are obtained: The first expression
⟨ for⟩ Qq is the average of
⟨ the⟩square of
2

⟨ ⟩ ⟨ ⟩ ⟨ ⟩ the deviation Q𝜓 − Q𝜓 from the average Q𝜓 , and the

1
n2𝜓x = n2𝜓y = n2𝜓z = (9) second is then a standard result for variance in statistics. The
3
deviator Qq is a second invariant of the function, and would
⟨ ⟩ ⟨ ⟩ ⟨ ⟩ normally be taken non-negative. Its value does not depend
n𝜓y n𝜓z = n𝜓z n𝜓x = n𝜓x n𝜓y = 0 (10) on any physical direction.
A complement A†𝜓 of a function A′𝜓 can be defined as the
⟨ ⟩
1 average of A′𝜓 over directions that are normal to ψ. This is
so ∶ 𝐧𝜓 ⊗ 𝐧 𝜓 = 𝐈 (11)
3 defined more formally in Appendix ⟨ 1. ⟩
The label ψ in the expression Q𝜓 is a dummy variable,
where 𝐈 is the 3 × 3 identity matrix. The trace⟨of a matrix
⟩ is and may
⟨ be ⟩ replaced by any other ⟨ if ⟩the meaning is clear.
the sum of its diagonal terms, so the trace of 𝐧𝜓 ⊗ 𝐧𝜓 is Thus Q𝛽 would be the same as Q𝜓 if the first average is
1. By transforming to a coordinate frame in which 𝐧𝜓 acts understood as being taken for all directions β. Where there
along a coordinate axis, it can readily be verified that is ambiguity, the variable over which the average is taken
𝐧𝜓 ⊗ 𝐧𝜓 is a tensor with two principal values equal to zero can also be placed outside as a subscript. Functions of more
and one equal to 1. than one direction can also be useful, and the algebra of
If a function Q𝜓 has the same value for all directions, then such functions can be very similar to that of tensors. For
its average is that value (the average of a constant is that example, a function C𝜓 may be defined in terms of functions
constant). If a and b are two quantities that are independent A𝜓𝛽 and B𝛽 as:
of direction, and Q1𝜓 and Q2𝜓 are two functions of direc-
tion, then the average of the linear combination is the linear ⟨ ⟩ d𝛺𝛽
∫
C𝜓 = A𝜓𝛽 B𝛽 𝛽 = A𝜓𝛽 B𝛽 (16)
combination of the averages: 4𝜋

13
83 Page 6 of 25 E. T. R. Dean

(( ) )
This is analogous to the multiplication of two tensors, or ∑ ∑ ∑
1 1
the post-multiplication of a matrix A by a vector 𝐁 , except 𝛔� = ��
𝐮ic = ��
𝐮ic𝜓 + 𝛅𝜓 (18)
VREP VREP
that the result is an average rather than a sum. c,i 𝜓 c,i,𝛾

where the first summation on the right is taken over all the
2.2 Effective stress
groups. VREP
′′
depends on which original expression is used
from Eq. 3, and the quantity in brackets is the summation
Figure 2 shows the sign convention used herein for positive
taken over all the members of the ψth group. Inside the sec-
components of the effective stress tensor. Normal stresses
ond sum on the right, 𝐮ic𝜓 is a uniaxial tensor with its one
are positive in compression. The off-diagonal components
non-zero principal value in direction ψ, so it is a scalar mul-
are in the opposite directions to shear stresses denoted as
tiple of 𝐧𝜓 ⊗ 𝐧𝜓 . The last term 𝛅𝜓 describes the second-
𝜏rs = −𝜎rs.
In Eq. 3, since 𝛔′ is symmetric, only the symmetric parts order deviations from direction ψ over the area of the patch.
of the tensors 𝐅c ⊗ 𝐋c or 𝐋c ⊗ 𝐅c are needed in calculat- A stress magnitude 𝜎𝜓′ can then be defined for the ψth group
ing the macroscopic stress. Consistently with Cauchy [13], as:
every symmetric stress tensor can be regarded as the sum ∑ da𝜓
1
of three uniaxial stress tensors, the ith of which represents ��
𝐮ic𝜓 = 3𝜎𝜓� 𝐧𝜓 ⊗ 𝐧𝜓 (19)
VREP a
a stress in a principal direction of the original tensor. Let 𝐮ic c,i,𝛾

be the uniaxial stress tensor representing a uniaxial stress in

Combining this with the previous equation, and then put-
the ith principal direction of the dyadic product for the cth
ting the result in the DEM Eq. 3 for the stress tensor 𝛔′ ,
contact, so that:
gives:
( ) ( ) ∑3 ( )
∑ da𝜓
sym 𝐅c ⊗ 𝐋c = sym 𝐋c ⊗ 𝐅c = 𝐮ic (17) 𝛔� = 3𝜎𝜓� 𝐧𝜓 ⊗ 𝐧𝜓 + 𝛅𝜓 (20)
i=1 𝜓
a
Substituting this into Eq. 3 gives a summation over 3N Taking the limit as smaller and smaller patch sizes are
uniaxial stress tensors. Let these uniaxial stress tensors now considered, the deviations from direction ψ tend to zero, so
be organized into groups, the ψth of which contains only the second order term 𝛅𝜓 is expected to collapse to zero, and
stress directions that correspond to points in a particular
the summation will become an integral, giving:
patch on the unit sphere of area da𝜓 . Then:
⟨ ⟩
𝛔� = 3𝜎𝜓� 𝐧𝜓 ⊗ 𝐧𝜓 (21)

where 𝜎𝜓′ has now become a continuous scalar function

of direction ψ. It will be convenient to call this the “outer
stress” function. Its derivation involves no special constraint,
other than that it be integrable in the way described. It may
for example be continuous or discontinuous, provided the
integration remains possible and gives a definite result.
Equations 19 and 21 represent a new way of upscaling
from particle mechanics to continuum mechanics, but the
summation in Eq. 19 means that the method does not pro-
vide the inverse path, from continuum to particle mechan-
ics. The upscaling is based on orientational averaging, and
in this sense it follows many previous models in the litera-
ture. For example, Pande and Sharma’s [55] multi-laminate
model involves averaging of the behaviours of laminates
each of which is affected by several stresses. A key feature
is that Eq. 21 has been proved, essentially simply from the
assumption that macroscopic stress represents point forces
at inter-particle contacts—the assumption underlying Eq. 3.
Equation 21 does not involve any additional constitutive
assumption. And the stress function 𝜎𝜓′ has a clear physical
meaning in terms of the statistics of the principal values and
Fig. 2  Sign convention for components of the effective stress tensor

13
Representation of stress and strain in granular materials using functions of direction Page 7 of 25 83

directions of stresses that are set up in the soil matrix by Ẇ ′ is the first invariant of Ẇ 𝜓′ . The product 𝐅.𝐅
̇ −1 can be
different combinations of differently oriented inter-particle expressed as:
contact forces and different brace vectors.
Equation 21 implies that each element of the stress tensor ⎡ 0 −𝜃̇ z 𝜃̇ y ⎤ ⎡ 𝜀̇ xx 𝜀̇ xy 𝜀̇ xz ⎤
is a directional ̇𝐅.𝐅−1 = 𝛉̇ − 𝛆̇ = ⎢ 𝜃̇ z 0 −𝜃̇ x ⎥ − ⎢ 𝜀̇ yx 𝜀̇ yy 𝜀̇ yz ⎥ (26)
⟨ average⟩ of the outer stress function, for exam- ⎢ ̇ ̇ ⎥ ⎢ ⎥
ple, 𝜎xx = 3𝜎𝜓 n𝜓x . Since the symmetric tensor has only
� � 2 ⎣ −𝜃y 𝜃x 0 ⎦ ⎣ 𝜀̇ zx 𝜀̇ zy 𝜀̇ zz ⎦
six independent elements, Eq. 21 puts only six constraints This rate equation has an equivalent form in terms of
on the function, and the function can contain infinitely more infinitesimals, and a way of using it in the context of a theory
information than the tensor. Indeed, the tensor contains no of large deformations appropriate to soils is discussed later.
information at all about the function deviator 𝜎q′ , which is The first matrix is the anti-symmetric part of 𝐅.𝐅 ̇ −1, and
defined as in the general
⟨ Eq.⟩ 15, or about any of the infinite contains rotation rates. The second is the symmetric parts
number of invariants 𝜎 ′ m𝜓 for m ≠ 1. (𝜀̇ rs = 𝜀̇ sr ), and contains strain rates with compression posi-
The stress function will have a basic symmetry that its tive. In some soil mechanics conventions, shear strain rates
value in a given direction is equal to its value in the diametri- may be represented as 𝛾̇ rs = −2𝜀̇ rs [45]. Evaluating the trace
cally opposite direction. Taking the trace of both sides of in Eq. 2 gives:
Eq. 21, noting that the trace of (𝛔′ is three)times the mean �
normal effective stress p’, and tr 𝐧𝜓 ⊗ 𝐧𝜓 = 1, gives: ⎧ 𝜎xx 𝜀̇ xx + 𝜏yz 𝛾̇ yz +⎫
� ⎪ � ⎪
⟨ ⟩ W = −V ⎨ 𝜎yy 𝜀̇ yy + 𝜏zx 𝛾̇ zx +⎬ (27)
𝜎𝜓� = p� (22) ⎪ � ⎪
⎩ 𝜎zz 𝜀̇ zz + 𝜏xy 𝛾̇ xy ⎭
Also, if 𝜎𝜓′ equals p’ for all directions ψ, then from This is consistent with the familiar result that rotation
Eqs. 11 and 21, the Cauchy stress tensor will be hydrostatic. rates are not associated with any working [70]. Using Eqs. 5
However, the converse is not necessarily true. Many dif- and 26 to expand the right side of Eq. 25 gives the rate of
ferent functions can satisfy the same six constraints, so a outer volume strain as:
hydrostatic tensor does not necessarily imply that the stress
function equals p’ in all directions. 2
⎧ n𝜓x 𝜀̇ xx + 2n𝜓y n𝜓z 𝜀̇ yz +⎫
⎪ ⎪
V̇ 𝜓 = −3V ⎨ n2𝜓y 𝜀̇ yy + 2n𝜓z n𝜓x 𝜀̇ zx +⎬
2.3 Strain rate and work ⎪ 2 ⎪ (28)
⎩ n𝜓z 𝜀̇ zz + 2n𝜓x n𝜓y 𝜀̇ xy ⎭
For tensor models, it is conventional to use strain rate or = −3Vtr(𝐧𝜓 ⊗ 𝐧𝜓 𝛆)
̇
incremental strain parameters that are “work-conjugate”
with the chosen measures of stress (e.g. [57, 70]. An exten- If special axes are considered with the x-axis in direction
sion of this idea to functions is straightforward. Using Eq. 21 ψ, then n𝜓x = 1 and n𝜓y = n𝜓z = 0, and the strain rate 𝜀̇ xx will
to substitute for the stress tensor in Eq. 2, and organizing the equal the normal compressive strain rate 𝜀̇ 𝜓 in direction ψ.
result appropriately, gives: Hence the right side simplifies to:
⟨ ⟩
Ẇ � = Ẇ 𝜓� (23) V̇ 𝜓 = −3V 𝜀̇ 𝜓 (29)

The work rate Ẇ 𝜓′ in Eq. 24 might then be interpreted as

with ∶ Ẇ 𝜓� = −𝜎𝜓� V̇ 𝜓 (24) a measure of external work that is done on the material in
compressing it in this direction, per unit particle
⟨ volume.
⟩
Using Eqs. 9, 10 and 28 to compute the average V̇ 𝜓 gives:
̇ −1 )
V̇ 𝜓 = −3Vtr(𝐧𝜓 ⊗ 𝐧𝜓 𝐅.𝐅 (25) ⟨ ⟩ { }
V̇ 𝜓 = −V 𝜀̇ xx + 𝜀̇ yy + 𝜀̇ zz = V̇ (30)
V̇ 𝜓 might be conveniently called an “outer” volumetric
rate. It can be different for different directions. There may Thus the directional average of the outer volumetric rates
be some relationship with Pietruszczak and Krucinski’s [59] V̇ 𝜓 is the rate of change V̇ of the overall specific volume V.
proposal of “directional porosity”, but there is no general
implication that V̇ 𝜓 is a differential of any quantity that is
independent of the choice of reference configuration. Ẇ 𝜓′ is
a work rate per unit particle volume associated with direc-
tion ψ. The first equation shows that the overall work rate

13
83 Page 8 of 25 E. T. R. Dean

3 General aspects of constitutive modelling than the order of magnitude of a large point. It would then
with functions follow that the outer volume rates in Eq. 25 are also local
variables.
3.1 Noll’s axioms
3.3 Objectivity
Noll’s [53] four axioms for the constitutive modelling of
“simple materials” have largely been accepted [49, 77]. The concept of objectivity is that behaviour that is the property
The Axioms of Local Action, Frame Indifference (Objec- of a macroscopic physical system does not depend on the way
tivity), and Indifference to Rigid Transformation would it is observed [18, 34, 50, 52]. Wooseok et al. [81] argue that
seem to be directly transferable to functions through large errors occur in some finite element models due to the
Eqs. 21 and 25, because the tensors in these equations use of non-objective stress rates. For the present proposals, the
satisfy those axioms. algebra by which Eq. 21 was proved from the objective Eq. 3
ensures that the stress function 𝜎𝜓′ is objective. Indeed, this is
3.2 Local action implicit in the definition of ψ as a physical direction. The fol-
lowing text verifies this formally.
The concept of local action is that the stress and other Noll’s [53] Axiom of Material Frame-Indifference also
descriptions of the state of a material can be represented expresses the expectation that constitutive behaviour of a
by descriptions that apply at infinitesimal points. Bower [7] material does not depend on the frame in which it is observed.
states that, if the principle holds, “the stress at a point in the Let Q be the frame transformation tensor that relates the value
solid depends only on the change in shape of a vanishingly 𝐫 of a vector in a given frame to its value 𝐫 alt = 𝐐𝐫 in an
small volume element surrounding the point. It must there- alternative frame. The tensor transformation law for stress is
fore be a function of the deformation gradient or a strain 𝛔�alt = 𝐐𝛔� 𝐐T (e.g. [70]. Using Eq. 21 for 𝛔′ gives:
measure that is derived from it.” Another implication is that ⟨ ⟩
the stress–strain behaviour at a point does not depend on 𝛔� alt = 𝐐 3𝜎𝜓� 𝐧𝜓 ⊗ 𝐧𝜓 𝐐T (31)
the spatial gradient of stress at that point. So, for example,
the equilibrium equations are formally independent of the Since Q is independent of direction it can be brought inside
constitutive behaviour. the averaging operator. Now the dyadic product of two column
It is easy to imagine that granular media might not obey vectors equals the result of post-multiplying the first vector by
this principle. Consider for example a hard particle that, as a the transpose (to row vector) of the second (e.g. Eq. 5). Using
result of a small change of loading conditions, slips at some this property then gives:
inter-particle contact. If the particle is hard enough, there ⟨ ⟩
𝛔� alt = 3𝜎𝜓,alt (32)
�
will not be enough local elasticity to take up the movement 𝐧𝜁,alt ⊗ 𝐧𝜁,alt
in the immediate vicinity of the slip, and the movement will
necessarily spread to nearby particles. Their movements will
spread further, and there seems to be no a priori reason to
with ∶ 𝐧𝜓,alt = 𝐐𝐧𝜓 (33)
suppose that this spreading effect is limited in scale.
This paper explores the concept of spreading later. Lambe �
𝜎𝜓,alt = 𝜎𝜓� (34)
and Whitman [44] describe the usual assumption in practi-
cal engineering in the following way: “When we talk about The first equation has the same form as the original Eq. 21,
the stresses at a point within a soil, we often must envision confirming frame-indifference. The second equation defines
a rather large point”. On this basis, Eq. 3 is a local equation a one-to-one correspondence between the values of the unit
in the sense that it involves a summation over a representa- vector in direction ψ as evaluated in the two different frames.
tive volume that is small enough to be considered as a “large The third relates the stress functions in the two frames.
point”—small enough in comparison to the scale of a practi-
cal engineering problem, but large enough that the statistics
3.4 Determinism
involved in the summation are not sensitive to the actual size
of the REV. It follows that Eq. 21 is also a local equation,
Together with Locality, the Axiom of Determinism implies
since it applies to the same elementary volume.
that a model for isothermal stress–strain behaviour needs to
The calculation of work rate in Eq. 2 involves the assump-
be capable of predicting the rates of all constitutive variables
tion that the deformation gradient has physical meaning. The
given the deformation rate and their current values. This
present paper assumes that this is the case, implying that the
might be represented by the following calculation sequence,
gradient is not measured over dimension scales that are less
where TCM represents a tensor-based model:

13
Representation of stress and strain in granular materials using functions of direction Page 9 of 25 83

( ) ⟨ ⟩
̇ −1 → TCM → 𝛔̇ �
𝐅𝐅 (35) V̇ 𝜓∗ = 1 − 𝛼𝜓 V̇ 𝜓 + 𝛼𝜓 V̇ 𝜓 (37)
⟨ ⟩
This “roadmap” is the opposite compared to models of where V̇ 𝜓 is the same as the rate V̇ of change of overall
elasto-plasticity in which strain rates are determined from specific volume. The function 𝛼𝜓 may depend on other fac-
given stress rates, though this form can be achieved by tors in the model, and is herein called a “spreading coef-
inverting a model’s compliance matrix [28]. ficient”, because the effect of the last term is a spreading of
Equation 21 implies that the stress function is not neces- strain rates from different directions. Taking averages gives:
sarily determinable uniquely solely from the stress tensor. ⟨ ⟩ ( ⟨ ⟩) ⟨ ⟩
Equations 21 and 25 would imply that the above process can V̇ 𝜓∗ = 1 + 𝛼𝜓 V̇ − 𝛼𝜓 V̇ 𝜓 (38)
be expanded as follows, where FCM represents a functions-
based model: Hence, while the average inner rate equals the overall
volumetric rate in a purely volumetric deformation, inde-
̇ −1 → V̇ 𝜓 → FCM → 𝜎̇ � → 𝛔̇ �
𝐅𝐅 (36) pendently of the spreading coefficient, it does not equal this
𝜓
for a⟨general
⟩ deformation
⟨ ⟩ unless it happens to be the case
This raises the following issue. It would seem natural to that 𝛼𝜓 V = 𝛼𝜓 V𝜓 .
̇ ̇
expect that the FCM would need to be able to calculate the Let 𝜎𝜓∗ be a function that is work-conjugate with V̇ 𝜓∗ . One
rate of change of the outer stress function for any given outer might call this an “inner” stress function. In the context of
volumetric rate function V̇ 𝜓 . However, Eq. 25 implies that functions, “work-conjugate” would mean that, for all possi-
the only functions that are involved in practice will be those ble outer functions V̇ 𝜓 , the rate of working can alternatively
that can be specified by six independent tensor components be calculated using the following equations analogous to
of strain rate. Since the set of all possible functions V̇ 𝜓 is Eqs. 23, 24:
infinitely larger than the set of functions that satisfy the six ⟨ ⟩
constraints, the question arises as to whether to impose the Ẇ � = Ẇ 𝜓∗ (39)
six constraints on the FCM or not.
Three arguments exist in favour of not imposing these
six constraints on FCM. First, the calculation of V̇ 𝜓 from Ẇ 𝜓∗ = −𝜎𝜓∗ V̇ 𝜓∗ (40)
̇ −1 automatically applies the constraints, so Occam’s razor
𝐅𝐅
suggests there is no need to apply them again in the FCM. Using Eq. 37 to substitute for V̇ 𝜓∗ , then taking the aver-
Second, as shown by example in the next section, applying age, gives:
them is algebraically inconvenient, and can prevent infer- ⟨ ( ) ⟨ ⟩⟩
ences which seem natural to make. Third, the behaviour of Ẇ � = − 𝜎𝜓∗ 1 − 𝛼𝜓 V̇ 𝜓 + 𝛼𝜓 𝜎𝜓∗ V̇ 𝜓 (41)
a soil element in a laboratory test is only controlled on its
boundary. The six constraints are not explicitly enforced Using Eqs. 23, 24 to substitute for the work rate on the
inside a sample, which may be free to deform in ways that left, and using the swap property (Eq. 13) on the last term
use more freedoms. inside the outer averaging operator on the right gives:
Accordingly, the present paper assumes that it is not ⟨ ⟩ ⟨{ ( ) } ⟩
appropriate to impose the tensor constraints on an FCM. − 𝜎𝜓� V̇ 𝜓 = − 𝜎𝜓∗ 1 − 𝛼𝜓 + 𝛼𝜓 𝜎𝜓∗ V̇ 𝜓 (42)
This might be called a principle of “freedom from tensor
constraint”. It seems natural to expect that the outer stress is given
by the quantity inside the curly brackets. However, Eq. 25
3.5 An example of using freedom from tensor imposes six constraints of V̇ 𝜓 , and the extraction property
constraint (Eq. 14) implies that:
{ ( ) }
Composite quantities are used in many tensor-based models. 𝜎𝜓� V̇ 𝜓 = 𝜎𝜓∗ 1 − 𝛼𝜓 + 𝛼𝜓 𝜎𝜓∗ V̇ 𝜓 + Ẇ 𝜓�� (43)
A simple example is the triaxial deviator stress which con- ⟨ ⟩
sists of a combination of principal stresses. In the context of where Ẇ 𝜓�� = 0 . If the principle of freedom from tensor
the modelling schematics above, some FCMs might involve constraint is adopted, one can now proceed as follows. First,
a process of combining outer geometrical rates from differ- referring to the patch on the sphere in Fig. 1, a function 𝛿𝜓
ent directions. for direction ψ is defined such that it is large and finite and
A simple example, developed for the purposes of illustra- constant for the patch, and zero everywhere. Using this as
tion and not from any particular data, might be the following V̇ 𝜓 gives, on calculating the integrals:
definition of a new “inner volumetric rate” V̇ 𝜓∗ , in terms of
the outer geometric rate V̇ 𝜓 and another function, 𝛼𝜓 :

13
83 Page 10 of 25 E. T. R. Dean

{ ( ) ⟨ ⟩}
⟨ ⟩ ∑ 𝛿𝛺𝜓
−𝜎𝜓� 𝛿𝜓 = − 𝜎𝜓∗ 1 − 𝛼𝜓 + 𝛼𝜓 𝜎𝜓∗ 𝛿𝜓 (44) Q𝜓 = lim Q𝜓 (47)
𝛿𝛺𝜓 →0
patches
4𝜋
Dividing each side by 𝛿𝜓 then gives a relation for the
patch around direction ψ. Doing the same for each patch, Differentiating with respect to time, and noting that the
then taking the limits as the patch sizes tend to zero (assum- solid angle has a rate, gives:
ing that any fractal effects can be managed) gives: .
. ⎛ ⎞
( ) ⟨ ⟩ � � � ⎜ 𝛿𝛺𝜓 𝛿𝛺𝜓 ⎟
𝜎𝜓� = 1 − 𝛼𝜓 𝜎𝜓∗ + 𝛼𝜓 𝜎𝜓∗ (45) Q𝜓 = lim ⎜Q̇ 𝜓 4𝜋 + Q𝜓 4𝜋 ⎟ (48)
patches ⎜ ⎟
𝛿𝛺𝜓 →0
⎝ ⎠
This is an assembly equation for the outer stresses in
terms of the inner stresses, ensuring work conjugacy. It hap- The rate of change of the solid angle can be obtained by
pens to have the property that the average outer stress equals expressing Eq. 7 in terms of finite patch sizes, so that:
the average inner stress (which means that both averages
𝛿v𝜓
always equal the mean normal effective stress). Appendix 2 𝛿𝛺𝜓 = 3 (49)
derives the inverse, giving the inner stress function in terms r𝜓3
of the outer stress function.
This is an exact equation involving finite quantities, so
there is no difficulty in differentiating it with respect to time,
3.6 Determinism and locality revisited giving:
.
If the representation of spreading using a function 𝛼𝜓 is .
𝛿v𝜓 ṙ 𝜓
valid, then Eq. 37 implies that the inner variables are local 𝛿𝛺𝜓 = 𝛿𝛺𝜓 − 3 𝛿𝛺𝜓 (50)
𝛿v𝜓 r𝜓
variables just as the outer ones are. In effect the spreading
is assumed to happen only within a “large point”, and to not The first fraction on the right is the rate of change of vol-
violate the principle of local action. ume divided by volume, which can be expressed in terms of
In a continuum context, Eq. 37 is one example of a wide specific volume as V∕V̇ . The second fraction is the rate of
range of possibilities of relating outer functions to functions extension strain of a material vector in direction ψ, which
that may have some more direct role in continuum constitu- can be calculated from Eq. 29 (recalling that 𝜀̇ 𝜓 is positive
tive behaviour. This raises the possibility that, in general, in compression while ṙ 𝜓 ∕r𝜓 is positive in expansion). Using
the constitutive roadmap of Eq. 36 can be further expanded these results in Eq. 48, then taking the limit as the patch
as follows: sizes tend to zero, gives:
̇ −1 → V̇ 𝜓 → V̇ ∗ → ICM → 𝜎̇ ∗ → 𝜎̇ � → 𝛔̇ �
𝐅𝐅 (46) ⟨ ⟩ ⟨
.
⟩
𝜓 𝜓 𝜓
Q𝜓 = Q̇ 𝜓 + Q𝜓 ġ 𝜓 (51)
where ICM represents an “inner constitutive model” that
is somehow simpler than the models at other levels. For .

example, the inner equations for a given direction ψ will not d𝛺𝜓 V̇ − V̇ 𝜓
ġ 𝜓 = = (52)
necessarily need to use all the information available from d𝛺𝜓 V
all other directions.
The first equation shows that, when ψ represents material
3.7 Mechanical consistency direction, the rate of change of an average is the sum of the
average of the rate of change of the quantity and an extra,
Nothing in the above algebra restricts the direction ψ to geometric effect. The second shows that this is due to the
be fixed in relation to an external coordinate frame. It may rate of change of solid angle. This might be called a “bunch-
alternatively be the direction of a material vector, which can ing effect”, with ġ 𝜓 being negative if material directions in
change in a stress–strain process. the neighbourhood of ψ are changing in a way that makes
In this second case, ψ itself has a rate of change. In Fig. 1, them bunch together, and positive if they are spreading.
the directions and the size of the patch will all change, and The above results are new in the sense that they show
this will affect the calculation of the change of the average that the bunching effect is a real feature of behaviour that is
of any general function Q𝜓 . To calculate the effect, it is con- implied by classical tensor mechanics but is not explicitly
venient to re-state the integral of Eq. 6 as a limit as the sizes recognized in classical tensor mechanics. Equation 51 might
of patches covering the sphere tend to zero, or equivalently be called a “general bunching equation”. Its effects for the
as the solid angles that they subtend tend to zero: Cauchy stress rate are described in Appendix 3. Taking the

13
Representation of stress and strain in granular materials using functions of direction Page 11 of 25 83

⟨ ⟩
⟨average
⟩ of both sides of Eq. 52, and using Eq. 30, gives H � = H𝜓� (54)
ġ 𝜓 = 0 . This is expected since there is no change to the
total solid angle subtended by the full sphere of Fig. 1.
⟨ ⟩
Ẇ d� = Ẇ 𝜓d (55)
�

3.8 Thermodynamic consistency
Differentiating the first equation with respect to time, and
using the general bunching Eq. 51, gives:
Collins and Houlsby [19] developed a thermodynamic analy-
⟨ ⟩
sis for granular materials subject to continuum stress–strain
Ḣ � = Ḣ 𝜓� + H𝜓� ġ 𝜓 (56)
events. They showed that, under conditions of equilibrium
and constant temperature (isothermal), work is used up as
with ġ 𝜓 given by Eq. 52 if ψ represents a material direction,
follows:
or with ġ 𝜓 = 0 if ψ represents direction fixed with respect to
Ẇ � = Ḣ � + Ẇ d� (53) an external coordinate frame. Ḣ 𝜓′ would be the rate of change
of a Helmholtz free energy that is associated with direction
where Ḣ ′ is the change of the Helmholtz free energy H ′ per ψ, and Ẇ 𝜓d
′
would be a rate of dissipation of work or energy
unit particle volume. Ẇ d′ is the rate of work or energy loss in in particulate processes associated with direction ψ. Putting
irreversible, dissipative processes such as those associated these results into Eq. 53 then gives:
with frictional sliding, contact damage, or particle crushing. ⟨ ⟩ ⟨ ⟩ ⟨ ⟩
Ẇ d′ is truly irreversible. It is non-negative in the sign conven- Ẇ 𝜓∗ = Ḣ 𝜓� + H𝜓� ġ 𝜓 + Ẇ 𝜓d �
(57)
tion used herein.
Collins and Houlsby [19] derived their equation on the The averaging operator is linear (Eq. 12), so the two
basis of the familiar assumptions of small deformation averages on the right can be combined into one. Using the
theory, including the assumptions represented (in different extraction property (Eq. 14) then gives:
notation) in Eq. 26 of the present paper. In effect then, their
result is applicable as a local theory. An important implica- Ẇ 𝜓∗ + Ė 𝜓� = Ḣ 𝜓� + H𝜓� ġ 𝜓 + Ẇ 𝜓d
�
(58)
tion relates to the concept that has become known as the ⟨ ⟩
“exploitation of the entropy principle” [17, 75]. The issue where Ė � = 0. This is now an equation that involves only
𝜓
was addressed by Coleman and Noll [18] in the context of quantities for direction ψ. It might be regarded as the funda-
heat conduction, which requires that there be a spatial gradi- mental work-balance equation for a postulated class of mate-
ent of temperature, and by Liu [46] and others, and the two rials for which Eqs. 54, 55 have physical meaning.
papers were discussed by Triani et al. [76]. In the present The balance is between two inputs on the left and three
paper, the deformation gradient is the only spatial gradient outputs on the right. On the left, the inner work Ẇ 𝜓∗ can be
involved: the assumption is therefore formally required that known in a calculation once the strain rates are known. It
this gradient does not affect the basic constitutive behaviour is calculated using Eq. 25 and whatever is found to be the
of a large material point. relation between outer and inner geometric rates, an example
Collins and Houlsby’s [19] derivation is such that Ḣ ′ being Eq. 37. The second input is Ė 𝜓′ . This is interpreted as
includes the reversible component of the change of entropy energy that is transferred into particle mechanical processes
in addition to change of strain energy. Consequently, the associated with direction ψ from particle mechanical pro-
terms on the right of their equation are not in general the cesses associated with other directions. These would be local
same as the elastic and plastic components of work that are transfers, occurring inside the “large point”. Experimental
often assumed in plasticity models [37]. In those models, evidence that supports their existence is discussed later.
plastic work is the net work done in a closed stress cycle. The right side of Eq. 58 contains three ways the inputs are
By contrast, dissipative work is here taken to be associated used. One is the change the Helmholtz energy, one is to
with the irreversible component of change of entropy that account for the bunching effect. The third is in dissipated
occurs in a stress–strain process that is not necessarily a work. Appendix 4 derives the relation between Ẇ 𝜓d ′
and a
closed stress cycle, and is not necessarily a loading process. cumulative dissipated work function.
If the inner work equals the outer work, no loss of gener- Equation 58 provides the following guide as to what a
ality occurs if the work rate in Collins and Houlsby’s⟨ ⟩ [19] continuum constitutive model is required to do, and what
original equation is taken to be the average Ẇ 𝜓∗ of the kinds of assumptions it needs to provide. Since the inner
inner work rate. It seems natural to consider whether there work rate and the bunching effect can be calculated if the
might be materials for which the Helmholtz energy and dis- strain rates are known, all that remains for a model to do is:
sipated work are also averages, so that:

13
83 Page 12 of 25 E. T. R. Dean

1. provide rules for calculating two of the functions Ė 𝜓′ , 4 Examples of simple constitutive models
Ḣ 𝜓′ , and Ẇ 𝜓d
′
for a given input function Ẇ 𝜓∗ , (the balance
equation will then provide the means of calculating the 4.1 Preliminary considerations
third of the three functions), and
2. provide rules for inferring the inner stress rates from the The aim of the following presentation is to explore the use
results. of the functional representation of stress in the context of
some relatively simple constitutive behaviours. This is found
For example, if a relation between energy and stress is to lead to potentially new understandings of the particle
used to solve the second requirement, the rules needed in the mechanical origins of some elastic parameters and of criti-
first would need to result in equations for the receipts and the cal and steady states. However, these results would need to
dissipated work rate, given the inner work rate. be verified by future DEM simulations. The exploration also
sheds potential light on anisotropy and plastic spin. Again,
3.9 Large deformations these results would need to be confirmed in further work.
For simplicity in the following, strains and rotations are
The derivations above are in terms of work rates, and the assumed to be small so that their products can be neglected.
equations can be readily converted to infinitesimal (incre- Direction ψ is taken to be relative to a fixed external frame,
mental) form by multiplying by an infinitesimal increment so that the bunching effect does not appear in the equations.
of time. There remains the question of whether the results
are applicable to realistic stress–strain processes for granular 4.2 Linear elastic behaviours
media, which typically involve finite deformations, and often
involve very large deformations [5, 10, 62]. Linear elastic models, sometimes called Hooke’s law, are
A simple procedure by which the present equations can described for soils by Davis and Selvadurai [26], Lings [45],
be used is proposed as follows. First, a large deformations and others. The most general model has 21 independent
process is discretized into a number of small increments. In material constants, but the most complex type usually con-
an increment, small deformations are obtained by multiply- sidered for soils is typically either the transversely isotropic
ing rates by the relevant small time increment. The proposed linear elastic (TILE) model, with rotational symmetry about
procedure will use the directions ψ as material directions one axis, or the fully isotropic linear elastic (FILE) model.
for the increment, and these will in general also have a rate Orthotropic behaviour is also sometimes relevant [24].
of change. The bunching effect will therefore need to be Linear combinations of well-connected linear elastic sys-
included. tems are typically linear elastic, so linear combinations of
At the end of the constitutive calculation for the incre- linear elastic relations for different directions ψ might be
ment, the material directions are now different, and all the expected to be linear elastic. Since linear elastic behaviour
functions of direction (such as Helmholtz energy) now refer is reversible, the dissipated work rate in the work balance
to the new directions. Let us re-name the new directions ζ, would be zero, and a simple approach would be to take the
so that the new Helmholtz energy is now H𝜁′ . The relation receipts as zero. A natural assumption would be that the
between ζ and the original directions ψ at the start of the inner stress rate would be linearly related to the inner geo-
increment is known because the incremental strain is known. metric rate:
So it is now possible to re-cast H𝜁′ as a new function H𝜓′ of
V̇ 𝜓∗
the directions ψ as they were at the start of the increment. 𝜎̇ 𝜓∗ = −K𝜓∗ (59)
This is now the energy function that will apply at the start V
of the next increment. where the modulus K𝜓∗ could be a function of direction. For
If this proposed procedure is done for every increment on simplicity, the following analysis assumes that the volume
the deformations path, the final result is a function H𝜓′ that strains are sufficiently small that that changes of K𝜓∗ and
describes the material state at the end of the path in terms of K𝜓∗ ∕V are of secondary significance during a stress–strain
directions that have not changed since the start of the path. process, and can be ignored in a linear analysis. Integrating
The procedure is similar in concept to a re-meshing over time then gives:
method used in a large-strain finite element analysis. Further
aspects of the use of meshes to represent the continuous con- K𝜓∗ ( )
𝜎𝜓∗ = ∗
V𝜓0 − V𝜓∗ (60)
stitutive functions are discussed later. V
where V𝜓0
∗
could be interpreted as a value of V𝜓∗ at zero stress.
Using Eqs. 25 and 37 the right side can be expressed in

13
Representation of stress and strain in granular materials using functions of direction Page 13 of 25 83

terms of tensor strain rates. If the spreading coefficient is a 4.3 Special linear elastic behaviours
constant, there are no non-linearities involved, and the result
can be integrated to give a path-independent relation Constitutive properties of a transversely isotropic material are
between inner stress and tensor strains. symmetric under rotation about an axis of symmetry [45, 70].
Using Eqs. 40 and 59 to calculate the inner work rate, then Taking this as the z-axis, the spreading coefficient and the
integrating the balance equation over time taking the receipts, modulus function would be a function of, at most, the polar
bunching effect, and dissipated work as zero, and assuming angle θ (Fig. 1), and might be written as 𝛼𝜃 and K𝜃∗. The inte-
that energy is zero at zero inner stress, gives: grals implied by Eq. 63 then simplify considerably. Appen-
( )2 dix 5 confirms that the symmetries in the matrix of moduli
𝜎𝜓∗ calculated in this way are consistent with the symmetries in
∫
H𝜓� = −𝜎𝜓∗ V̇ 𝜓∗ dt = (61) the compliance matrix given by Lings [45] and others for a
2K𝜓∗ ∕V
time transversely isotropic linear elastic (TILE) material.
If the spreading coefficient and modulus function are inde-
This result could be interpreted as energy associated with pendent of direction, they may be denoted simply as α and
direction ψ. Taking the average, Eq. 54, then gives the Helm- K. Appendix 6 confirms that model then reduces to one of
holtz energy H ′. a standard fully isotropic linear elastic (FILE) material. The
( )2 inner modulus K is then found to equal the elastic bulk modu-
⟨ ⟩
𝜎𝜓∗ lus. The shear modulus is found to be given by:
H� =
1 (62)
2 K𝜓∗ ∕V 3
G= (1 − 𝛼)2 K (64)
5
Because the inner stress is related to inner geometry
through Eq. 60, and provided that the inner geometry can be The isotropic Poisson’s ratio μ is found to be given by:
related to cumulative strain by integration over time, it fol-
5 − 2(1 − 𝛼)2
lows that H ′ is a function of strain, implying that this model 𝜇= (65)
would satisfy the familiar continuum mechanical concept of 10 + 2(1 − 𝛼)2
a Cauchy elastic material (e.g. [70].
The spreading coefficient has a range from minus infin-
Lings [45] discusses the thermodynamic constraint of
ity to plus infinity. This equation shows that this range cor-
non-negative strain energy, and shows how this provides
responds to the familiar limits of − 1 to + 1/2 on the fully
limits of Poisson’s ratios for transversely isotropic and fully
isotropic Poisson’s ratio.
isotropic linear elastic materials. These limits include those
The above results represent new proposed understandings
by Love [48], Pickering [58], and Raymond [61]. If H ′ can
of what determines the isotropic shear modulus and Pois-
be interpreted as strain energy, the above equation shows
son’s ratio. The above equation shows that the isotropic Pois-
that the thermodynamic constraint is automatically satisfied
son’s ratio is 0.25 if the spreading coefficient is zero. This is
if the inner modulus K𝜓∗ is non-negative for all directions.
in the range that is typical for loose to medium dense sand
The detailed stress–strain equations are obtained by first
or sandy clay [8, 25]. In terms of the functions approach, it
substituting for the inner functions in Eq. 59. Using the par-
suggests that there may be little or no reversible interaction
ticular example of Eq. 37 for the geometrical relation, and
between directions in such soils.
using the differential of Eq. 45 for the stress relation, gives:

⎧ � 4.4 Elastic behaviours: discussion

�2 V̇ ∗ � � ̇ ⎫
⎪ 1 − 𝛼� K ∗ 𝜓 + 𝛼� 1 − 𝛼� K ∗ V ⎪
⎪ 𝜓 𝜓 V 𝜓 𝜓 𝜓V
⎪ Many previous authors have attempted to relate macroscopic
𝜎̇ 𝜓� = −⎨ � � � V̇ 𝜓∗
�
� � ̇ ⎬ (63)
⎪ + 𝛼� 1 − 𝛼� K ∗ V⎪ elastic parameters to particle mechanical ones. Bathurst and
+ 𝛼𝜓�2 K𝜓∗
⎪ 𝜓 𝜓 𝜓
V V⎪ Rothenburg [2] extended Horne’s [38] work and showed how
⎩ ⎭
a simple linear contact elasticity law could be used to gener-
This shows that, while outer stress is associated with ate the macroscopic elastic properties of an assemblage of
direction ψ, it can be determined by changes of geometry discs. Their proposals involve a two-dimensional version
that occur in other directions too. This equation can now be of the three-dimensional Eq. 3 used herein. Chang and Gao
used with Eq. 21, in rate form, and Eq. 28 to determine the [14] and Liu and Chang [47] propose expressions for con-
rate of change of the Cauchy stress tensor in terms of the tinuum elastic constants based on micromechanics includ-
tensor strain rates. ing a version of Eq. 3 of the present paper. Stránský et al.
[71] and Kruyt [43] developed equations for an assemblage
of spheres. Gu and Yang [32] investigated experimentally

13
83 Page 14 of 25 E. T. R. Dean

(a) Along the elastic lines such as EB and FC, reversible

elastic behaviour would occur, governed by some rela-
tion that would be a development of Eq. 60 and in
which a constant analogous to V𝜓0 ∗
would occur, differ-
ent for different elastic lines. The soil state at a given
time may be on different elastic lines for different direc-
tions ψ,
(b) If the stress reaches the yield limit, for example at point
B for line EB or point C for line FC, further compres-
sion in direction ψ would cause the state in that direc-
tion to move down the elastoplastic limit line ABCD.
On subsequent unloading, the state in that direction
Fig. 3  Adapting familiar ideas to create a simple elastoplastic inner
would move up a new elastic line,
constitituve relation (c) A no-tension limit might be imposed in each direction
ψ. The derivation of Eq. 21 showed that 𝜎𝜓′ represent
a combination of principal values of tensors that are
and through DEM modelling a problem with the classical determined by contact forces and branch vectors. If
Hertz–Mindlin contact law [27]. It therefore seems relevant contact forces cannot be tensile, this would suggest a
to ask how the spreading function and inner elastic moduli limit of 𝜎𝜓′ ≥ 0 in uncemented granular soils. A limit
might relate to these previous works. 𝜎𝜓∗ ≥ 0 might alternatively be postulated.
An answer will need to take account of the loss of par-
ticle mechanical information that is inherent in the process Item (b) introduces plastic behaviour relative to a direc-
of defining the stress function 𝜎𝜓′ in terms of the average on tion ψ, so a model with this feature would certainly exhibit
the left of Eq. 19. This loss means that one cannot deduce elasto-plasticity in its overall macroscopic behaviour, though
detailed particle mechanical information from macroscopic not necessarily in accordance with conventional elasto-plas-
parameters of this type of model, though a deduction in the tic model assumptions. The modelling framework would
opposite direction remains possible, from particle to con- allow for different behaviours in different physical direc-
tinuum parameters. It follows that the spreading coefficients tions, perhaps giving a result akin to Callladine’s [11] micro-
and inner stiffnesses are not necessarily fundamental micro- structural model. Whether or not a model that includes (b)
scopic particle mechanical properties. Instead, they may pos- would be realistic may depend on other assumptions, as
sibly relate to mesoscopic scales, i.e. to aggregate structure discussed in general terms earlier in relation to the work
or fabric. balance Eq. 58.
One difference between this framework and some previ-
4.5 Elasto‑plastic behaviours (1) some preliminary ous elasto-plastic models might be in its incremental calcu-
ideas lation procedure. A simple procedure for this model would
be as follows. Given the tensor strain rates 𝛆̇ over a time
The constitutive map of Eq. 46 indicates that there are two increment dt, the outer and inner changes of geometry are
places where more complex behaviours can be introduced. calculated. Then, independently for each direction ψ, three
One is the spreading relation V̇ 𝜓 → V̇ 𝜓∗ and the consequent steps would be carried out:
conjugate stress assembly relation 𝜎𝜓∗ → 𝜎𝜓� . The second is
the inner relation V̇ 𝜓∗ → 𝜎̇ 𝜓∗ . Whatever properties or behav- 1. a trial calculation: the stress response to the given
iours are introduced in these places will appear in the tensor change of geometry is calculated assuming the soil state
model 𝛆̇ → 𝛔̇ ′, typically in more complicated form due to the in that direction remains on its elastic line
spreading and assembly processes. 2. a yield check: if the result is greater than the yield stress
A simple way to start to introduce elasto-plasticity with for the line, the response is recalculated using the elas-
functions would be to adapt a linear-elastic model by add- toplastic line ABCD
ing a yield stress 𝜎𝜓y ∗ as a function of direction ψ (as well
3. a tension check: if the result is tensile, the stress is re-
as of stress–strain history in that direction). Figure 3 illus- set to zero. This would mean that this state in direction
trates this for a given direction ψ, based on direct adaptation ψ would move up the geometric axis, such as from F
of basic features of the hydrostatic parts of the Cam–Clay towards E.
models of Schofield and Wroth [64] and Roscoe and Burland
[63]:

13
Representation of stress and strain in granular materials using functions of direction Page 15 of 25 83

This simple procedure omits consideration of the receipts compression (𝜀̇ r > 0 with 𝜀̇ a = 0). In pure volumetric com-
in the work balance equation, and some important aspects pression, 𝜀̇ a = 𝜀̇ r = 𝜀̇ vol ∕3, and so
of this omission will be considered later. The simple model
includes hardening along ABCD, softening by movement up V̇ 𝜓∗ −V̇
− = 𝜀̇ vol = (67)
the geometric axis, and induced anisotropy due to the fact V V
that different directions can evolve differently.
for this special case. This suggests that, if the inner relations
are also the same in all directions, the inner volumes can be
4.6 Elasto‑plastic behaviours (2) preliminary identified with the overall specific volume in this special
discussion type of hydrostatic process for isotropic soils.
The next section will confirm that this is a special case,
The preliminary considerations suggest that the develop- and that V𝜓∗ cannot be identified with V in more complex
ment of ideas about plasticity in the context of functions stress–strain processes, but it is useful to first explore the
may require very different ways of thinking compared to simple case of hydrostatic processes for isotropic samples.
previous conventional plasticity approaches. If yielding is The asymptotic isotropic compression curve assumed in the
something that occurs in terms of inner, direction-related Cam–Clay models by Schofield and Wroth [64] and Roscoe
variables, then and Burland [63] can be expressed as:

1. the question of whether loading or unloading occurs will V + 𝜆 ln p� = 𝛤 (68)

need to be answered in terms of inner variables rather where λ is a slope constant and Γ is a positional constant.
than tensor stresses, and the answer may be different for If the material is isotropic then all the inner stresses will
different directions ψ. evolve as the mean normal effective stress in these hydro-
2. a macroscopic yield envelope and flow rule may static events. The corresponding inner asymptotic compres-
be likely to be emergent properties rather than funda- sion curve would therefore be:
mental ones.
V𝜓∗ + 𝜆 ln 𝜎𝜓∗ = 𝛤 (69)
These features illustrate both the potential richness and
the potential complexity of the functional approach. But also Alternative formulae may be developed in the same way,
there can be relative simplicity: for example, the concept of such as the double-log proposals by Butterfield [9], Hashi-
“plastic spin” would likely not be needed, because the model guchi [33], Butterfield and Marchi [10]. Other inner proper-
structure in Eq. 46 is already complete, and so is expected ties might be inferable similarly, depending on modelling
to automatically create any changes of symmetry that may assumptions, such as the relation between Helmholtz free
result from the different responses in different material energy function H𝜓′ and the inner stress and inner volume.
directions ψ. Finally, it is worth noting that the energy receipts term Ė 𝜓′
in Eq. 58 was not used in the linear elastic models discussed
earlier. This term provides a second connection between
4.7 Elasto‑plastic behaviours (3) further physical directions, the first being the spreading coefficient.
preliminary explorations It therefore seems possible that the spreading coefficient
might represent reversible aspects of spreading, while the
It can be interesting to briefly explore standard triaxial receipts represent irreversible aspects. A simple particle
behaviours. If the spreading coefficient is a constant α, as in mechanical explanation might be as follows:
the FILE model, then the rate of change of inner geometry
due to axial and radial strain rates 𝜀̇ a and 𝜀̇ r is calculated in • the spreading coefficients model, in some average way,
Appendix 7 as follows: the mechanics by which a given particle equilibrates one
inter-particle force by a combination of inter-particle
V̇ 𝜓∗ ( )
− = 3(1 − 𝛼) 𝜀̇ a cos 2 𝜃 + 𝜀̇ r sin 2 𝜃 + 𝛼 𝜀̇ vol (66) forces at other contacts and in different directions,
V • the receipts represent energy changes caused by the
where 𝜀̇ vol = 𝜀̇ a + 2𝜀̇ r is the volumetric strain rate. If changes of these forces that would be induced if particles
0 < 𝛼 < 1, directions close to axial will experience faster slide frictionally at inter-particle contacts.
compression in one-dimensional axial compression (𝜀̇ a > 0
with 𝜀̇ r = 0), indicating that yielding will begin in directions Further work outside the scope of the present paper would
close to vertical. By contrast, directions close to radial will be needed to explore this further. It would imply that Ė 𝜓′
experience faster compression in two-dimensional, radial would be determined by dissipated work rates in different

13
83 Page 16 of 25 E. T. R. Dean

directions, subject to the thermodynamic

⟨ ⟩ requirement that ⎛ ⎞
the average receipt is zero, Ė 𝜓� = 0. ⎜ H𝜓� ⎟� �
Ė 𝜓� = j ⎜1 − � � ⎟ Ẇ 𝜓d
�
(71)
⎜ H𝜓� ⎟
⎝ ⎠
4.8 Critical and steady states as experimental
evidence where j could be a positive dimensionless material constant.
This describes a system in which energy is transferred from
Schofield and Wroth [64] define a “critical state” as one at high energy directions to low energy directions. This might
which a soil element can be sheared continuously at constant be sufficient to prevent the energy continually increasing in
volume and constant effective stress. An analogous concept directions for which Ẇ 𝜓∗ > 0, and also prevent energy reach-
is that of “steady states” [41, 60]. A critical state can also be ing zero in directions for which Ẇ 𝜓∗ < 0.
interpreted as a special case of proportional straining, other However, this cannot be a complete answer, for the fol-
examples of which include isotropic and one-dimensional lowing reason. It means that the work balance equation for
compression [16, 74]. general processes would become:
If the state of the soil does not change as shearing con-
tinues, then V𝜓∗ cannot be a description of state, since any ⎛ ⎞
⎜ H𝜓� ⎟� �
definite equation like Eq. 37 would mean that it does change. + j ⎜1 − � � ⎟ Ẇ 𝜓d
Ẇ 𝜓∗ (72)
� �
= Ḣ 𝜓� + Ẇ 𝜓d
It follows that the elastoplastic curve ABCD would need ⎜ H𝜓 ⎟
�
to be defined in a different way, one for which V̇ 𝜓∗ = V̇ for ⎝ ⎠
asymptotic hydrostatic compression if the soil is isotropic,
The problem now is that the three-step calculation proce-
but for which V𝜓∗ and V become uncoupled for asymptotic
dure for the simple plasticity model can no longer be applied
critical state deformations.
in the same way, because the average dissipated work would
During critical state straining, the constancy of the spe-
need to be known in order to do the trial calculation in step
cific volume and effective stress would suggest that inner
1, but the dissipated work will not be known until step 2 has
parameters such as the Helmholtz energy function are also
been done.
constant as the straining occurs. But some directions will
One possible way to address this new challenge would
receive a negative inner work rate Ẇ 𝜓∗ . From Eq. 66 with
be through modelling the rate of change of energy. If Ḣ 𝜓′
𝜀̇ vol = 𝜀̇ a + 2𝜀̇ r = 0 , this will occur in triaxial compression
involves a term that offsets the average on the left, then the
for directions with tan2 𝜃 > 2 . If the rate of change of the
simplicity of the calculation procedure might be recovered.
Helmholtz energy function is zero, the work balance Eq. 58
This would be consistent with the proposed physical expla-
in the absence of the bunching effect then reduces to:
nation of receipts discussed earlier. It might relate to the
Ẇ 𝜓∗ + Ė 𝜓� = Ẇ 𝜓d
�
(70) familiar phenomenon of plastic hardening, and if so would
indicate a difference compared to other plasticity models in
The right side cannot be negative, so this implies that the literature. Further work on this is outside the scope of
sufficient energy transfers must occur between directions the present paper.
such that the receipts for the directions experiencing nega-
tive inner work rates offset the negative rates and make the
left sides positive. In effect, therefore, critical and steady 5 Discussion
states are prima facie experimental evidence for the occur-
rence of receipts. 5.1 Nature of the new approach proposed herein

4.9 Critical and steady states as challenges Equation 21 for the functional representation of effective
stress was derived from the well-established DEM Eq. 3,
The elastic models discussed earlier did not need to involve which might be regarded as a mathematical representation
receipts, suggesting that receipts may depend only on dissi- of the familiar explanation of Terzaghi’s [72, 73] Principle
pated work. This might be consistent with a particle mechan- of Effective Stress in terms of inter-particle forces. Equa-
ics explanation in which receipts represent the re-distribution tion 30 for the outer volumetric rates followed directly from
of forces in a particle aggregate after frictional slip at particle work-conjugacy in the absence of Cosserat stress effects.
contacts or due to other irreversible processes. For purposes So these continuum variables provide a modelling basis
of discussion, a simple example of a formula might be: that is consistent with a widely accepted approximation to
the micro-mechanics of aggregates of hard, inert particles.
Bolton [6] argues that “continuum soil mechanics is
insufficient”. Soga and O’Sullivan [69] divide models

13
Representation of stress and strain in granular materials using functions of direction Page 17 of 25 83

into macro and micro ones. Liu and Chang [47] note that of this new approach to constitutive modelling of granular
continuum plasticity approaches involve parameters that media.
“describe directly the behavior of a particle assembly ele-
ment (macro-scale)”, which are typically measured in a 5.2 Achievements and some further considerations
laboratory element test. By contrast, the micro-mechanical
approach includes parameters that describe behaviour of an The simple models developed herein confirm that modelling
inter-particle contact (micro-scale), and which may be “dif- is possible and can produce useful results when functions
ficult or impossible to be obtained in laboratory directly”. are used to represent stress and strain rate. The elastic model
The aim is to use an upscaling or other procedure to deduce reproduces the typical transversely isotropic behaviour of
macroscopic continuum behaviour. soil and the elasto-plastic one includes yield behaviours and
Another recent example is given in the detailed microme- other realistic features. The challenges of using functions
chanics-based model by He et al. [35, 36], who identify a to model critical and steady states have also been explored.
problem of an “overwhelming number of parameters in the Features that have not been explored herein include
model”. They attribute this to the complexity of granular the possibility of a more complex approach to the spread-
materials, but it does present a challenge in terms of practi- ing equations such as making the spreading coefficient 𝛼𝜓′
cal engineering application. Micro-mechanical approaches depend on energy ratios. It seems clear that any behav-
to continuum modelling differ from DEM approaches in that ioural features that are included in the inner relation will
the latter do not necessarily involve an intermediate stage also appear in the overall behaviour, in more complex
of a continuum model. Huge efforts by many authors over form. For instance, an entire model would be time-depend-
many decades have been made using all of these approaches. ent if the inner relation between V̇ 𝜓∗ and 𝜎̇ 𝜓∗ was.
In the fully isotropic linear elastic model explored ear- In numerical implementation, a function of direction
lier, the inner modulus was equal to the macroscopic bulk might be represented by its values for a finite number of
modulus, and the Poisson’s ratio was related to the spreading directions. Several models in the literature already involve
coefficient. This suggests that the functional representation multiple directions, including the multi-laminate models
can involve a hybrid approach, which [55, 65]. It therefore seems likely that functions of direc-
tion can be modelled using simple adaptations of existing
• uses the basic micro–macro relationships of Eqs. 21 and finite element or other numerical methods.
30, and Figure 4 illustrates this for four symmetry cases. If only
• can include mesoscopic concepts such as the postulated triaxial stress–strain processes are to be explored on a fully
spreading coefficients, and isotropic or transversely isotropic material aligned with the
• can have its parameters measurable in macroscopic, ele- axis, then only the polar angle need be discretized, and this
ment-scale testing would correspond to a system in which the sphere of Fig. 1
was discretized into hoops as in Fig. 4a. If more complex
There is no implication of conflict between the various processes are applied, both the polar and azimuthal angles
approaches, and one would expect that there might eventu- need to be varied, and the deformation symmetries then
ally be ways by which all three approaches can agree in their indicate discretizations of one-eighth of the sphere for true
predictions for engineering behaviours of interest. Indeed, triaxial process, or one quarter for plane strain processes
one might expect that advances in one approach may assist that include simple shear. For general processes and mate-
in the development of one or more of the others. rial symmetries, only one half of the sphere need to be
An example of potential mutual assistance and consist- discretized, because of the basic symmetry that the value
ency might be as follows. If the spreading coefficient 𝛼𝜓 is a of a function in a given direction will be the same as its
function of direction, then it may relate to the way particles value in the diametrically opposite direction.
interact and therefore to fabric. One might use it to define a
variety of tensors, for example:
⟨ ⟩ 6 Concluding remarks
𝛂 = 3𝛼𝜓 𝐧𝜓 ⊗ 𝐧𝜓 (73)
This paper has proposed a continuum mechanics that is
This would transform like stress (Eqs. 32–34). It may intermediate in detail between particle mechanics and ten-
be that, if 𝛼𝜓 is also a function of stress–strain history as sor continuum mechanics. The proposal provides a more
experienced in terms of inner quantities, that function’s evo- rigorous but also simplified approach to the task of upscaling
lution would constitute a fabric evolution law. Whether this from particle mechanics to continuum mechanics. This leads
particular 𝛂 is useful or not, it illustrates the potential power to an alternative framework for the stress–strain constitutive

13
83 Page 18 of 25 E. T. R. Dean

Fig. 4  Discretizations of direc-

tion for various symmetries a (a) (b)
standard triaxial b true triaxial
c plane strain in the yz plane d
general

(c) (d)

modelling of granular materials that in particular considers was derived, based on Collins and Houlsby’s [19] thermo-
directionality. mechanics. A “bunching” effect was identified and included
The proposals are based on Terzaghi’s Principle of Effec- in this equation.
tive Stress and on a well established result in the discrete Familiar linear elastic models of soils—fully isotropic
element method, Eq. 3, which relates inter-particle point and transversely isotropic—were shown to be readily
forces to macroscopic tensors. The proposed descriptions expressed in terms of functions of direction. The assump-
of stress and deformations result from organizing the infor- tion of non-negative strain energy, which produces all the
mation according to the directions of the principal stresses known limits on Poisson’s ratio [45], is expressed by the
associated with the effects of each particle contact in a rep- simple criterion that the inner modulus function K𝜓∗ is non-
resentative elementary volume. negative. For the fully isotropic linear elastic (FILE) model,
The organization involves a summation (Eq. 19) that the spreading coefficient is independent of direction, and
loses some particle mechanical information, so the proposed determines Poisson’s ratio though Eq. 65.
descriptions lose part of the explicit link between particu- By adding a yield stress function to the isotropic model,
late-level and macroscopic behaviours that a full discrete it was argued that a simple elasto-plastic model can cer-
element method or a fundamental micro-mechanical deri- tainly be obtained. A model so constructed will be able to
vation of macroscopic behaviour might provide. However, allow for different hardening in different physical directions,
analysis of elastic behaviour suggested that the new frame- thereby providing opportunities for modelling induced ani-
work may provide a link to mesoscopic, aggregate or fabric sotropy. The structure of Eq. 46 also means that the model
properties. is complete, so that one can expect that the phenomenon
The framework was shown to satisfy Noll’s Axioms of that is interpreted as “plastic spin” in a tensor context would
constitutive modelling. A principle of freedom from tensor emerge automatically from a functional plasticity model,
constraint was proposed. A general structure for constitutive without the need for a separate concept of spin. The new
models employing functions was inferred from the Axiom proposals also create the potential for new approaches in
of Determinism, Eq. 46. A new work balance equation numerical analysis methods, which were briefly discussed.

13
Representation of stress and strain in granular materials using functions of direction Page 19 of 25 83

A future challenge will be to find the required additional Appendix 2: Inverse of Eq. 45
constitutive assumptions that can lead to a good match
( )
between model and data. Proposing such assumptions in Multiplying both sides of Eq. 45 by 𝛼𝜓 ∕ 1 − 𝛼𝜓 , (assuming
detail is outside the scope of this paper, but some general that 𝛼𝜓 is not 1 for any direction), then taking the average, gives:
aspects have been briefly discussed in relation to the model- ⟨ ⟩ ⟨
𝛼𝜓 𝜎𝜓� ⟩⟨ ⟩
ling of critical or steady states. 1
= 𝛼𝜓 𝜎𝜓∗ (75)
1 − 𝛼𝜓 1 − 𝛼𝜓
⟨ ⟩
Funding No external funding was used to support the work described
Using this to substitute for 𝛼𝜓 𝜎𝜓∗ in the original Eq. 45
in this paper.
and re-arranging then gives an expression by which the inner
Compliance with ethical standards function can be calculated if the outer function is known.

Conflict of interest The author declares that he has no conflicts of in-

terest in relation to this paper. Appendix 3: Material directions
Human and animal rights This paper does not refer to techniques relat-
and the calculation of stress rate
ing to research involving humans or animals, and does not use results
from research methods involving humans or animals. The general bunching Eq. 51 applies whenever the directions
ψ are considered to be material directions, which can change
Informed consent This research does not involve anything that requires
informed consent.
during a stress–strain process. Applying this when differen-
tiating Eq. 21 for the Cauchy stress tensor gives:
⟨ ⟩
Open Access This article is licensed under a Creative Commons Attri-
𝛔̇ � = 3𝜎̇ 𝜓� 𝐧𝜓 ⊗ 𝐧𝜓 +
bution 4.0 International License, which permits use, sharing, adapta-
tion, distribution and reproduction in any medium or format, as long ⟨ ( )⟩ ⟨ ⟩ (76)
d
as you give appropriate credit to the original author(s) and the source, 3𝜎𝜓� 𝐧𝜓 ⊗ 𝐧 𝜓 + 3𝜎𝜓� ġ 𝜓 𝐧𝜓 ⊗ 𝐧𝜓
provide a link to the Creative Commons licence, and indicate if changes
dt
were made. The images or other third party material in this article are The first term on the right involves the rate of change of
included in the article’s Creative Commons licence, unless indicated
otherwise in a credit line to the material. If material is not included in the stress function. It is proposed that this be interpreted
the article’s Creative Commons licence and your intended use is not as the constitutive effect in the context of the functions
permitted by statutory regulation or exceeds the permitted use, you will approach. This is the only term that is needed if directions
need to obtain permission directly from the copyright holder. To view a ψ are fixed relative to an external frame.
copy of this licence, visit http://creat​iveco​mmons​.org/licen​ses/by/4.0/.
The second and third terms apply if ψ represents material
direction. The second term is explored below. The third term
is the bunching effect.
Appendix 1: The complement function
Let 𝐧𝜓 be the value of a unit material vector at the current
configuration. That material vector may have been in a dif-
The complement A†𝜓 of a function A′𝜓 is conveniently defined
ferent direction ζ and been associated with a different unit
as the average of the function over directions normal to ψ.
vector 𝐦𝜁 at the reference configuration, such that:
One way to calculate its value for a given ψ would be to
define a new coordinate frame whose z-axis is in direction 𝐅𝐦𝜁
ψ. One might denote polar and azimuthal angles measured 𝐧𝜓 =
| | (77)
in this frame as 𝜃𝜓 and 𝜙𝜓 . Then: |𝐅𝐦𝜁 |
| |
2𝜋
where the denominator is needed because 𝐧𝜓 is defined as
|
2𝜋 ∫
1
A†𝜓 = A�𝜓 | d𝜙 (74) a unit vector. Using the facts that the dyadic product of the
|𝜃𝜓 =𝜋∕2 𝜓
𝜙𝜓 =0 vectors is the first vector post-multiplied by the transpose
of the second, and that the squared-length of a vector is its
where A′𝜓 inside the integral is evaluated on 𝜃𝜓 = 𝜋∕2. The scalar product with itself, gives
derivation of the stress Eq. 21 shows that the value of the
stress function 𝜎𝜓′ in direction ψ represents principal val- 𝐅𝐦𝜁 𝐦T𝜁 𝐅T
𝐧𝜓 ⊗ 𝐧𝜓 = (78)
ues of tensors one of whose principal directions is ψ. The 𝐦T𝜁 𝐅T 𝐅𝐦𝜁
complementary stress function 𝜎𝜓† might be interpreted as
representing an average of all the principal values normal
to direction ψ.

13
83 Page 20 of 25 E. T. R. Dean

Differentiating this expression and using Eq. 26 and This confirms that the second law of thermodynamics is
the fact that the rotation rate 𝛉̇ is anti-symmetric while the obeyed. The net rate of dissipated work is:
strain rate 𝛆̇ is symmetric, gives: ⟨ ⟩
d𝛺𝜓
∫
( ) { ( ) ( ) } Ẇ d� = �
Ẇ 𝜓d �
= Ẇ 𝜓d (83)
d 4𝜋
𝐧𝜓 ⊗ 𝐧𝜓 = 𝛉̇ 𝐧𝜓 ⊗ 𝐧𝜓 − 𝐧𝜓 ⊗ 𝐧𝜓 𝛉̇
{dt( ) ( ) ( )} (79)
− 𝛆̇ 𝐧𝜓 ⊗ 𝐧𝜓 + 𝐧𝜓 ⊗ 𝐧𝜓 𝛆̇ − 2𝜀̇ 𝜓 𝐧𝜓 ⊗ 𝐧𝜓 However, material directions have changed relative to a
fixed external frame, the solid angle subtended by these mate-
rial directions
( has changed,
) and from Eq. 52 the new solid
where 𝜀̇ 𝜓 is the direct strain rate in the direction ψ (Eqs. 28,
angle is d𝛺𝜓 1 + ġ 𝜓 dt . Let C𝜓d
�
+ Ċ 𝜓d
�
dt be the new value of
29). The first term on the right is a simple effect of rota-
tion. It is expected because a body rotation will cause the the cumulative dissipated work function, so that the cumula-
principal directions of the tensors 𝐮ic in Eq. 17 to rotate too. tive dissipated work for these material directions is:
The second term is an effect of shear straining relative to ( ( ))
( ) d𝛺𝜓 1 + ġ 𝜓 dt
the direction ψ. � �
C𝜓d + Ċ 𝜓d dt (84)
4𝜋

at the end of the time increment. Equating this to the value

Appendix 4: Material directions given by Eq. 82 gives, after re-arrangement:
and cumulative dissipated work �
C𝜓d + Ẇ 𝜓d
�
dt
�
C𝜓d �
+ Ċ 𝜓d dt = (85)
The work balance Eq. 58 contains a dissipated work rate. 1 + ġ 𝜓 dt
If an analysis uses ψ as a fixed direction in relation to an
Using Taylor’s theorem to expand the denominator as a
external coordinate frame, then a straightforward integral
multiplier, cancelling the term C𝜓d
′
which then appears on both
over time will give the cumulative dissipated work associ-
sides, dividing the result by dt, and then taking the limit as dt
ated with any direction, and the average will give the net
tends to zero, gives:
cumulative dissipated work.
However, the calculation is more complex if an analysis �
Ċ 𝜓d �
= Ẇ 𝜓d �
− C𝜓d ġ 𝜓 (86)
uses ψ as a material direction. To clarify this, it can be
helpful to start by defining a function C𝜓d
′
such that, for the Re-arranging, then taking averages of both gives:
directions contained within a solid angle d𝛺𝜓 , the cumula- ⟨ ⟩ ⟨ ⟩
tive dissipated work is: Ẇ d� = Ẇ 𝜓d
� �
= Ċ 𝜓d �
+ C𝜓d ġ 𝜓 (87)
d𝛺𝜓
′
C𝜓d (80) This is consistent with the general bunching Eq. 51. It
4𝜋 shows that cumulative dissipated work is the function C𝜓d ′

The net cumulative dissipated work Wd′ is just the inte- whose rate satisfies Eq. 86 for a given dissipated work rate
gral over all directions: Ẇ 𝜓d
′
. C𝜓d
′
may in some cases reduce with time for some direc-
tions, even though the rates Ẇ d′ and Ẇ 𝜓d
′
remain non-negative.
d𝛺𝜓
∫
Wd� = �
C𝜓d (81) This appears to be a subtle effect of bunching that remains
4𝜋 consistent with the laws of thermodynamics.
Consider an increment of behaviour that takes place
over a time increment dt. If the dissipated work rate is Ẇ 𝜓d
′
Appendix 5: Transversely isotropic linear
for this increment, then the cumulative dissipated work per
elastic (TILE) model
unit particle volume for these material directions at the
end of the time increment is:
A method for constructing linear elastic models with func-
( ) d𝛺𝜓 tions was discussed in the main text, with the simple spread-
�
C𝜓d �
+ Ẇ 𝜓d dt (82) ing relation in Eq. 37 being used in combination with Eq. 59
4𝜋
to give the stress function in Eq. 63.
If Ẇ 𝜓d
′
is non-negative, this cumulative value will not be For a transversely isotropic material with the z-direction
less that the cumulative value at the start of the increment, and as the axis of symmetry, the relevant material functions will
it follows that the integral sum over all directions also will not also have this symmetry, and so be functions of polar angle
be smaller at the end of an increment than it was at the start. only. To denote this, it can be helpful to denote them as

13
Representation of stress and strain in granular materials using functions of direction Page 21 of 25 83

Table 1  Calculation grid for term 1 in the TILE model (color table online)

· · · · · ·

Table 2  Parameters in equations for simple linear elastic models 𝛼𝜃 = 𝛼𝜋−𝜃 (88)
(a) Transversely A = 3P + 2S + U
isotropic model B = P + 2S + U K𝜃∗ = K𝜋−𝜃
∗
C =R+S+T +U
(89)
D = Q + 2T + U
E = 2R
A simple example would be 𝛼𝜃 = 1 + k cos 2n𝜃 , where k
F = 2P is a dimensionless constant and n is an integer. Replacing θ
with π−θ leaves the value of the function unchanged.
∫ 1 − 𝛼𝜃 K𝜃∗ sin5 𝜃d𝜃
𝜋 ( )2
9
P = Ghh = 16
Equation 63 has four terms on the right. Each is used in
∫
𝜃=0
9
𝜋 ( )2 the integration processes needed to calculate the six compo-
Q= 1 − 𝛼𝜃 K𝜃∗ sin 𝜃 cos4 𝜃d𝜃
2
nents of the Cauchy stress tensor. Table 1 gives a convenient
∫
𝜃=0
𝜋 ( )2
R = Gvh = 9
4
1 − 𝛼𝜃 K𝜃∗ cos2 𝜃 sin3 𝜃d𝜃 way of assembling the integrals involved in the first term on
the right. Expanding this term using Eq. 28 gives:
∫ 𝛼𝜃 1 − 𝛼𝜃 K𝜃∗ sin3 𝜃d𝜃
𝜃=0
3
𝜋 ( )
S= 4
�2 V̇ 𝜓
𝜃=0 ∗
�
∫ 𝛼𝜃 1 − 𝛼𝜃 K𝜃∗ cos2 𝜃 sin 𝜃d𝜃
𝜋 ( )
T= 3
2
1 − 𝛼𝜓 K𝜓∗ =
𝜃=0 V
∫ 𝛼𝜃2 K𝜃∗ sin 𝜃d𝜃
𝜋
U= 1 ⎧ n2 𝜀̇ + 2n n 𝜀̇ + ⎫
2
𝜃=0 � �2 ∗ ⎪ n2 𝜀̇ + 2n n 𝜀̇ + ⎪
𝜓x xx 𝜓y 𝜓z yz

(b) Fully isotropic 1( ) 3 1 − 𝛼𝜃 K𝜃 ⎨ 𝜓y yy 𝜓z 𝜓x zx ⎬

A=D= 9 − 8𝛼 + 4𝛼 2 K
model 5( ) ⎪ n2 𝜀̇ zz + 2n𝜓x n𝜓y 𝜀̇ xy ⎪
B=C= 1
3 + 4𝛼 − 2𝛼 2 K ⎩ 𝜓z ⎭
5
E=F= 6
(1 − 𝛼)2 K (90)
5

𝛼𝜃 and K𝜃∗ respectively. In combination with the symmetry This is written in stylized form along the second line of
that a relevant function has the same value in diametrically the table, and the first line contains the expressions for the
opposite directions, this means that the functions will be unit matrix components. Down the first column, the multi-
such that: pliers needed for the six components of the Cauchy tensor
are written. In the central cells of the table, the relevant

13
83 Page 22 of 25 E. T. R. Dean

products of the unit vector components are written. These to a constant K. The integrals simplify dramatically, as indi-
would be integrated in order to obtain the relevant compo- cated in Table 2b, so that the result takes the following form:
nent of the answer. For example, the top left cell contains
the terms involved in the following integral for the effect of ⎛ 𝜎̇ � ⎞ ⎛ A B B 0 0 0 ⎞⎛ 𝜀̇ xx ⎞
⎜ 𝜎̇ xx
� ⎟
𝜀̇ xx on 𝜎̇ xx : ⎜ yy ⎟ ⎜⎜ B 0 ⎟⎜ 𝜀̇ yy ⎟
′
A B 0 0
⎜ 𝜎̇ zz� ⎟ ⎜ B ⎟⎜ ⎟
⟨ { ( )2 } ⟩ B A 0 0 0 ⎟⎜ 𝜀̇ zz ⎟
3 3 1 − 𝛼𝜃 K𝜃∗ n2𝜓x 𝜀̇ xx n2𝜓x = ⎜ 𝜎̇ � ⎟ = ⎜ 0 0 0 E 0 0 ⎟⎜ 𝜀̇ yz ⎟ (94)
⎜ yz ⎟ ⎜
⎜ 𝜎̇ zx
�
⎟ 0 0 0 0 E 0 ⎟⎜ 𝜀̇ zx ⎟
⎜ 𝜎̇ � ⎟ ⎜⎝ 0 ⎟⎜ ⎟
𝜋 2𝜋
( )2 (91) 0 0 0 0 E ⎠⎝ 𝜀̇ xy ⎠
∫ ∫
sin 𝜃d𝜃d𝜙 ⎝ xy ⎠
9 1 − 𝛼𝜃 K𝜃∗ sin4 𝜃 cos4 𝜙
4𝜋
𝜃=0 𝜙=0
This stiffness equations for the standard FILE model
where sin 𝜃 cos 𝜙 is written in the top left cell. Because the
4 4 can be obtained by inverting the compliance equations
spreading coefficient and inner modulus depend only on θ, given for example by Davis and Selvadurai [26], and the
the integral can be readily separated into one over θ and one results can be written in the following form:
over ϕ. Doing the one over ϕ then gives: �
⎛ 𝜎̇ xx ⎞ ⎛1−𝜇 𝜇 𝜇 ⎞⎛ 𝜀̇ xx ⎞
⟨ { ( )2 } ⟩ ⎜ 𝜎̇ � ⎟ = E ⎜ 𝜇 1 − 𝜇 𝜇 ⎟⎜ 𝜀̇ yy ⎟ (95)
3 3 1 − 𝛼𝜃 K𝜃∗ n2𝜓x 𝜀̇ xx n2𝜓x = ⎜ yy ⎟ (1 + 𝜇)(1 − 2𝜇) ⎜ ⎟⎜ ⎟
⎝ 𝜎̇ zz� ⎠ ⎝ 𝜇 𝜇 1 − 𝜇 ⎠⎝ 𝜀̇ zz ⎠
𝜋
( )2 (92)
8 ∫
9 3𝜋
× 1 − 𝛼𝜃 K𝜃∗ sin5 𝜃d𝜃 𝜏̇ xy 𝜏̇ yz
4𝜋 𝜏̇ zx E
𝜃=0 G= = = = (96)
𝛾̇ xy 𝛾̇ yz 𝛾̇ zx 2(1 + 𝜇)
The remaining integral will depend on the particular forms
of the spreading coefficient and modulus function. The 36 ele- where E is Young’s Modulus, μ is Poisson’s ratio, and G
ments Table 1 as a whole form a symmetric matrix which can is the shear modulus. Comparing the two equations for the
be considered to consist of four 3x3 blocks. The integrals for normal stress rates shows that the ratio B/A in the first equals
the cells in the blocks at top right and bottom left evaluate to the ratio 𝜇∕(1 − 𝜇) in the second. Using Table 2 for A and B
zero for the present symmetries, as do the integrals for the off- then leads to Eq. 65 for the Poisson’s ratio.
diagonal cells in the block at bpttom right. The relation for the bulk modulus K can be deduced
The integrals for the second, third, and fourth terms in similarly, or by taking averages of both sides of Eq. 59,
Eq. 63 are simpler, and the final results for the effective with K𝜓∗ = K , giving:
stress tensor rate have the following form: ⟨ ⟩
⟨ ⟩ V̇ 𝜓∗
𝜎̇ 𝜓∗ = −K (97)
⎛ 𝜎̇ � ⎞ ⎛ A B C 0 0 0 ⎞⎛ 𝜀̇ xx ⎞ V
⎜ 𝜎̇ xx
� ⎟
⟨ ⟩
⎜ yy ⎟ ⎜⎜ B A C 0 0 0 ⎟⎜ 𝜀̇ yy ⎟
⎟⎜ ⎟
⎜ 𝜎̇ zz� ⎟ ⎜ C C D 0 0 0 ⎟⎜ 𝜀̇ zz ⎟ If 𝛼𝜓 = 𝛼 , Eq. 45 implies that 𝜎̇ 𝜓∗ is the rate of change
⎜ 𝜎̇ � ⎟ = ⎜ 0 (93) of mean
⎜ yz ⎟ ⎜ 0 0 E 0 0 ⎟⎜ 𝜀̇ yz ⎟ ⟨ ⟩ normal effective stress, and Eq. 37 implies that
⎜ 𝜎̇ zx
�
⎟ 0 0 0 0 E 0 ⎟⎜ 𝜀̇ zx ⎟ V̇ 𝜓∗ ∕V is the negative of the volumetric strain rate.
⎜ 𝜎̇ � ⎟ ⎜⎝ 0 0 0 0 0
⎟⎜ ⎟
F ⎠⎝ 𝜀̇ xy ⎠
⎝ xy ⎠ Hence the above equation implies that K is the bulk
modulus.
where parameters A–F are given in Table 2a. These six param-
eters are related because A = B + F , so there are five inde-
pendent parameters. This is consistent with Lings [45] who
also found five (though expressed differently from herein). Appendix 7: Standard triaxial conditions

Equation 66 applied for the special case of a speading

coefficent that is independent of direction, and was derived
Appendix 6: Fully isotropic linear elastic as follows. Firstly, the deformation rate tensor 𝐅𝐅 ̇ −1 and
(FILE) model the strain rate tensor 𝛆̇ for a standard triaxial process with
direction z being the axial direction satisfy the following
For a fully isotropic material, no constitutive properties relations:
can depend on direction, so the spreading coefficient 𝛼𝜃
reduces to a constant α and the inner modulus K𝜓∗ reduces

13
Representation of stress and strain in granular materials using functions of direction Page 23 of 25 83

⎡ 𝜀̇ r 0 0 ⎤ MM-ALE formulations. Arch. Appl. Mech. 84, 743–755 (2014).

https​://doi.org/10.1007/s0041​9-014-0830-5
= −𝛆̇ = −⎢ 0 𝜀̇ r 0 ⎥
−1
̇
𝐅𝐅 (98)
⎢ ⎥ 6. Bolton, M.D.: The role of micro-mechanics in soil mechan-
⎣ 0 0 𝜀̇ a ⎦ ics. In International Workshop on Soil Crushability, Yamagu-
chi University, Japan, July 1999, also available as Bolton MD
Using this in Eq. 25, and with Eqs. 4, 5, gives the cor- (2000) Technical Report CUED/Soils-TR313, Cambridge Uni-
responding expression for V̇ 𝜓 . Using the result to substitute versity Engineering Department (1999)
7. Bower, A.: Continuum Mechanics 6. constitutive laws. Online
for V̇ 𝜓 in Eq. 37 then gives Eq. 66. lecture notes, Course EN 2210, Brown University. https​://www.
If the material is isotropic or transversely isotropic, the brown​. edu/Depar ​t ment​s /Engin​e erin​g /Cours​e s/En221​/ Notes​/
inner stress state will have rotational symmetry about the Const​i tuti​ve_Laws/Const​i tuti​ve_Laws.htm Accessed 4 May
vertical axis, and the inner stress function will be express- 2019
8. Bowles, J.E.: Foundation Analysis and Design. McGraw-Hill,
ible as a function 𝜎𝜃∗ of the polar angle θ (Fig. 1). With the New York (1996)
spreading coefficient independent of direction, equation 45 9. Butterfield, R.: A natural compression law for soils an advance
then gives: on e-log p’. Géotechnique 29, 469–480 (1979)
10. Butterfield, R., Marchi, M.: The compressibility of natural and
⟨ ⟩
𝜎𝜓� = (1 − 𝛼)𝜎𝜃∗ + 𝛼 𝜎𝜃∗ reconstituted marine clays. Geotech. Res. 4(3), 172–177 (2017).
(99)
https​://doi.org/10.1680/jgere​.17.00010​
11. Calladine, C.R.: Microstructural view of the mechanical proper-
Taking averages confirms that the average of the inner ties of saturated clay. Géotechnique 21, 391–415 (1971)
stress equals the average of the outer, which was shown 12. Casagrande, A., Carillo, N.: Shear failure of anisotropic materi-
in the main text to equal the mean normal effective stress. als. Proc. Boston Soc. Civ. Eng. 31, 74–87 (1944)
Using this result in Eq. 21 then gives the following results 13. Cauchy, A.L.: Recherches sur l’équilibre et le movement inté-
rieur des corps solides or fluides, élastiques ou non élastiques.
for the axial and radial components of the effective stress Bulletin de la Societé Philomatique, pp. 9–13. In: Cauchy AL
tensor: (1882–1974), Oeuvres complètes, s II, vol 2 Gauthier-Villars,
Paris, pp 300–304 (1822/3)
𝜋
{ } 14. Chang, C.S., Gao, J.: Second-gradient constitutive theory for

2∫
1 granular material with random packing structure. Int. J. Solids
axial ∶ 𝜎a� = 3(1 − 𝛼) cos2 𝜃 + 𝛼 𝜎𝜃∗ sin 𝜃d𝜃 (100)
Struct. 32(16), 2279–2293 (1995)
𝜃=0 15. Chen, G., Wu, Q., Zhou, Z., Ma, W., Chen, W., Khoshnevisan,
S., Yang, J.: Undrained anisotropy and cyclic resistance of satu-
𝜋
{ } rated silt subjected to various patterns of principal stress rota-
tion. Géotechnique 70(4), 317–331 (2020)
2∫
1 3
radial ∶ 𝜎r� = (1 − 𝛼) sin2 𝜃 + 𝛼 𝜎𝜃∗ sin 𝜃d𝜃 16. Chu, J., Lo, S.-C.R.: Asymptotic behaviour of a granular soil in
2
𝜃=0 strain path testing. Géotechnique 44(1), 65–82 (1994)
(101) 17. Cimmelli, V.A., Sellitto, A., Triani, V.: A generalized Coleman–
Noll procedure for the exploitation of the entropy principle.
for this special case. Because a relevant function has the Proc. R. Soc. Lond. Ser. A 466, 911–925 (2009)
same value in diametrically opposite directions, the extra 18. Coleman, B.D., Noll, W.: The thermodynamics of elastic mate-
rotational symmetry about the z-axis due to the standard rials with heat conduction and viscosity. Arch. Ration. Mech.
triaxial condition also implies that the stress function will Anal. 13, 167–178 (1963)
19. Collins, I.F., Houlsby, G.T.: Application of thermomechanical
have the same values for a given θ as for π−θ. principles to the modeling of geotechnical materials. Proc. R.
Soc. Lond. Ser. A 453, 1975–2001 (1997)
20. Cottechia, F., Guglielmi, S., Cafaro, F., Gens, A.: Characterisa-
tion of the multi-scale fabric features of high plasticity clays.
References Géotech. Lett. 9(4), 361–368 (2019)
21. Dafalias, Y.F.: Plastic spin: Necessity or redundancy? Int. J.
1. Altenbach, H., Naumenko, K., Zhikin, P.H.: A micro-polar theory Plast. 14(9), 909–931 (1998)
for binary media with application to phase-transitional flow of 22. Dafalias, Y.F.: Finite elastic–plastic deformations: beyond the
fiber suspensions. Continuum Mech. Thermodyn. 15, 539–570 plastic spin. Theoret. Appl. Mech. 38(4), 321–345 (2011)
(2003) 23. Dafalias, Y.F., Aifantis, E.C.: On the microscopic origin of the
2. Bathurst, R.J., Rothenburg, L.: Micromechanical aspects of iso- plastic spin. Acta Mech. 82(1–2), 31–48 (1990). https​: //doi.
tropic granular assemblies with linear contact interactions. J. org/10.1007/BF011​73738​
Appl. Math. 55(1), 17–23 (1988). https:​ //doi.org/10.1115/1.31736​ 24. Dean, E.T.R.: Soil hinges: macroscopic evidence and modeling
26 considerations. Int. J. Geomech. (2019). https​://doi.org/10.1061/
3. Bishop, A.W.: The principle of effective stress. Teknisk Ukeblad (asce)gm.1943-5622.00014​81
39, 859–863 (1959) 25. Das, B.M.: Principles of Foundation Engineering. Thomson
4. Blumenfeld, R.: The unusual problem of upscaling isostaticity Brookes/Cole, Boston (2005)
theory for granular matter. Granul. Matter 22, 38 (2020). https​:// 26. Davis, R.O., Selvadurai, A.P.S.: Elasticity and Geomechanics.
doi.org/10.1007/s1003​5-020-1002-7 Cambridge University Press, Cambridge (1996)
5. Bojanowski, C.: Numerical modeling of large deformations in 27. Duffy, J., Mindlin, R.D.: Stress–strain relations and vibrations
soil structure interaction problems using FE, EFG, SPH, and of a granular medium. J. Appl. Mech. 24, 585–593 (1956)

13
83 Page 24 of 25 E. T. R. Dean

28. Einav, I.: Thermomechanical relations between stress-space and 53. Noll, W.: A mathematical theory of the mechanical behaviour of
strain-space models. Géotechnique 54(5), 315–318 (2004) continuous media. Arch. Ration. Mech. Anal. 2, 197–226 (1958)
29. Faria, S.H.: Mixtures with continuous diversity: general theory 54. O’Sullivan, C.: Particulate Discrete Element Modelling, a
and application to polymer solutions. Continuum Mech. Ther- Geomechanics Perspective. Routledge, Abingdon (2010)
modyn. 13, 91–120 (2001) 55. Pande, G.N., Sharma, K.G.: Multilaminate model of clays—a
30. Göodert, G., Hutter, K.: Induced anisotropy in large ice shields: numerical analysis of the influence of rotation of the principal
theory and its homogenization. Continuum Mech. Thermodyn. stress axes. Int. J. Numer. Anal. Methods Geomech. 7, 397–418
10, 293–318 (1998) (1983)
31. Graham, J., Crooks, J.H.A., Lau, S.L.K.: Yield envelopes: 56. Papadimitriou, A.G., Chaloulos, Y.K., Dafalias, Y.F.: A fabric-
identification and geotechnical properties. Géotechnique 38(1), based sand plasticity model with reversal surfaces within aniso-
125–134 (1988) tropic critical state theory. Acta Geotech. 14, 253–277 (2019).
32. Gu, X.Q., Yang, J.: A discrete element analysis of elastic prop- https​://doi.org/10.1007/s1144​0-018-0751-5
erties of granular materials. Granul. Matter 15, 139–147 (2013) 57. Peric, D., Owen, D.R.J., Honnor, M.E.: On work-conjugacy
33. Hashiguchi, K.: On the linear relations of V-lnp and lnV-lnp for and finite strain elasto-plasticity. In: Pande, G.N., Middleton,
isotropic consolidation of soils. Int. J. Numer. Anal. Methods J. (Eds.), NUMETA 90, Numerical Methods in Engineering:
Geomech. 19, 367–376 (1995) Theory and Applications. Elsevier, Amsterdam Vol. 2, pp.
34. Hazel, A.: Information for MATH 45061: continuum mechanics. 718–729 (1990)
Course Notes in book form, University of Manchester (2019). 58. Pickering, D.J.: Anisotropic elastic parameters for soil. Géo-
https​: //perso​n alpa​ges.manch​e ster​. ac.uk/staff​/ Andre​w.Hazel​/ technique 20(3), 271–276 (1970)
MATH4​5061/MATH4​5061_Ch5.pdf Accessed 5 May 2019 59. Pietruszczak, S., Krucinski, S.: Description of clay anisotropy
35. He, X., Wu, W., Wang, S.: A constitutive model for granular employing the concept of directional porosity. In: Pietruszczak,
materials with evolving contact structure and contact forces— S., Pande, G.N. (Eds.), Proceedings of 3rd International Con-
part I: framework. Granul. Matter 21, 16 (2019). https​://doi. ference on Numerical Models in Geomechanics (NUMOG III).
org/10.1007/s1003​5-019-0868-8 Elsevier Applied Science, pp. 61–70 (1989)
36. He, X., Wu, W., Wang, S.: A constitutive model for granular 60. Poulos, S.J.: The steady state of deformation. J. Geotechn. Eng.
materials with evolving contact structure and contact forces— Div. ASCE 107, 553–562 (1981)
part 2: constitutive equations. Granul. Matter 21, 20 (2019). 61. Raymond, G.P.: Discussion: stresses and displacements in a
https​://doi.org/10.1007/s1003​5-019-0869-7 cross-anisotropic soil. Géotechnique 20(4), 456–458 (1970)
37. Hill, R.: The Mathematical Theory of Plasticity. Oxford Clar- 62. Roscoe, K.H.: The influence of strains in soil mechanics. Géo-
endon Press, Oxford (1950) technique 20(2), 129–170 (1970)
38. Horne, M.R.: The behaviour of an assembly of rotund, rigid, 63. Roscoe, K.H., Burland, J.B.: On the generalised stress-strain
cohesionless particles I and II. Proc. R. Soc. Lond. 286, 62–97 behaviour of ‘wet’ clay. In: Heyman, J., Leckie, F.A. (eds.)
(1965) Engineering Plasticity, pp. 535–608. Cambridge University
39. Houlsby, G.T.: The work input to a granular material. Géotech- Press, Cambridge (1968)
nique 29(3), 354–358 (1979) 64. Schofield, A.N., Wroth, C.P.: Critical State Soil Mechanics.
40. Jaeger, J.C., Cook, N.G.W., Zimmerman, R.W.: Fundamentals of McGraw-Hill, New York (1968)
Rock Mechanics. Blackwell, London (2007) 65. Schweiger, H.F., Wiltafsky, C., Scharinger, F., Galavi, V.: A
41. Jefferies, M.G., Been, K.: Soil Liquefaction. Taylor & Francis, multilaminate framework for modelling induced and inherent
Routledge (2006) anisotropy of soils. Géotechnique 59(2), 87–101 (2009)
42. Jiang, M., Zhang, A., Shen, Z.: Granular soils: from DEM simu- 66. Singh, P.N., Wallender, W.W.: Effective stress from force bal-
lation to constitutive modeling. Acta Geotech. 15, 1723–1744 ance on submerged granular particles. Int. J. Geomech. 7(3),
(2020). https​://doi.org/10.1007/s1144​0-020-00951​-7 186–191 (2007)
43. Kruyt, N.P.: Micromechanical study of elastic moduli of three- 67. Skempton, A.W.: Effective stress on soils, concrete and rocks.
dimensional granular assemblies. Int. J. Solids Struct. 51, 2336– In: Selected Papers on Soil Mechanics, Thomas Telford,
2344 (2014). https​://doi.org/10.1016/j.ijsol​str.2014.03.002 reprinted from Pore pressure and suction in soils, pp. 4–16
44. Lambe, P.W., Whitman, R.V.: Soil Mechanics—SI Version. Wiley, (1961)
Hoboken (1979) 68. Soare, S.C.: On the structure of metal plasticity constitutive equa-
45. Lings, M.L.: Drained and undrained anisotropic elastic stiffness tions and the physical origin of spin (2013). arXiv​:1306.6485
parameters. Géotechnique 51(6), 555–565 (2001) 69. Soga, K., O’Sullivan, C.: Modelling of geomaterials behavior.
46. Liu, I.-S.: Method of Lagrange multipliers for exploitation of the Soils Found. 50(6), 861–875 (2010)
entropy principle. Arch. Ration. Mech. Anal. 46, 131–148 (1972) 70. Spencer, A.J.M.: Continuum Mechanics. Dover Publications,
47. Liu, Y., Chang, C.S.: Relationship between element-level and con- Mineola (1980)
tact level parameters of micromechanical and upscaled plasticity 71. Stránský, J., Jirásek, M., Šmilauer, V.: Macroscopic elastic proper-
models of granular soils. Acta Geotech. 15, 1779–1798 (2020) ties of particle models. In: Proceedings of International Confer-
48. Love, A.E.H.: A Treatise on the Mathematical Theory of Elastic- ence on Modelling and Simulation 2010, Prague (2010)
ity, 4th edn (1st edn 1892). Cambridge University Press, Cam- 72. Terzaghi, K.: Die Berechnung der Durchlassigkeitsziffer des
bridge (1927) Tones aus dem Verlauf der hydrodynamischen Spannungserschei-
49. Maugin, G.A.: Continuum Mechanics Through the Twentieth Cen- nungen. Akademie der Wissenchaften en Wien, Sitzungbrichte,
tury: A Concise Historical Perspective. Springer, Berlin (2013) Mathematisch-natuwissenschaftliche Klasse, Part IIa, Vol. 132,
50. Metzger, D.R., Dubey, R.N.: Objective tensor rates and frame No. 3/4, pp. 125–138 (in German, translation in Clayton et al.
indifferent constitutive models. Mech. Res. Commun. 13(2), 1995) (1923)
91–96 (1986) 73. Terzaghi, K.: The shearing resistance of saturated soils and the
51. Mitchell, J.K., Soga, K.: Fundamentals of Soil Behavior. Wiley, angles between the planes of shear. In: 1st International Confer-
Hoboken (2005) ence on Soil Mechanics and Foundation Engineering, Cambridge,
52. Murdoch, A.I.: On material frame-indifference. Proc. R. Soc. 1, pp. 161–165 (1936)
Lond. Ser. A Math. Phys. Sci. 380, 417–426 (1982)

13
Representation of stress and strain in granular materials using functions of direction Page 25 of 25 83

74. Topolnicki, M., Gudehus, G., Mazurkiewicz, B.K.: Observed 81. Wooseok, J., Waas, A.M., Bazant, Z.P.: On the importance of
stress–strain behaviour of remoulded saturated clay under plane work-conjugacy and objective stress rates in finite deformation
strain conditions. Géotechnique 40(2), 155–187 (1990) incremental finite element analysis. ASME J. Appl. Math. 80(4),
75. Triani, V., Cimmelli, V.A.: Entropy principle, no-regular pro- 041024-9 (2013). https​://doi.org/10.1115/1.40078​28
cesses, and generalized exploitation procedures. J. Math. Phys. 82. Wu, X.H., Cai, G., Qi, J., Kim, J.R., Zhang, D., Jiang, M.: Work–
53: D063509-1 to -8 (2012) energy analysis of granular assemblies validates and calibrates
76. Triani, V., Papenfuss, C., Cimmelli, V.A., Muschik, W.: Exploi- a constitutive model. Granul. Matter 22, 28 (2020). https​://doi.
tation of the second law: Coleman–Noll and Liu procedure in org/10.1007/s1003​5-019-0990-7
comparison. J. Non-Equilib. Thermodyn. 33, 47–60 (2008) 83. Yimsiri, S., Soga, K.: Micromechanics-based stress-strain behav-
77. Truesdell, C.: A First Course in Rational Continuum Mechanics. iour of soils at small strains. Geotechnique 50(5), 559–571 (2000)
Elsevier, Amsterdam (1977) 84. Zhao, C.-F., Kruyt, N.P.: An evolution law for fabric anisotropy
78. Vijayan, A., Gan, Y., Annabattula, R.K.: Evolution of fabric in and its application in micromechanical modelling of granular
spherical granular assemblies under the influence of various load- materials. Int. J. Solids Struct. 196–197, 53–66 (2020). https​://
ing conditions through DEM. Granul. Matter 22, 34 (2020). https​ doi.org/10.1016/j.ijsol​str.2020.04.007
://doi.org/10.1007/s1003​5-020-1000-9 85. Zoback, M.D.: Reservoir Geomechanics. Cambridge University
79. Wang, R., Wei, C., Xue, L., Zhang, J.-M.: An anisotropic plastic- Press, Cambridge (2010)
ity model incorporating fabric evolution for monotonic and cyclic
behavior of sand. Acta Geotecnica (2020). https:​ //doi.org/10.1007/ Publisher’s Note Springer Nature remains neutral with regard to
s1144​0-020-00984​-y jurisdictional claims in published maps and institutional affiliations.
80. Wang, D., Zheng, H., Ji, Y., Bares, J., Behringer, R.P.: Shear of
granular materials composed of ellipses. Granul. Matter 22, 5
(2020). https​://doi.org/10.1007/s1003​5-019-0965-8

13