You are on page 1of 13

Continuum Mech. Thermodyn.

(1999) 4, 263–275 c Springer-Verlag 1999

Constitutive models for granular materials including quasi-static


frictional behaviour: Toward a thermodynamic theory of plasticity
B. Svendsen1 , K. Hutter2 , L. Laloui3
1 Federal Institute for Materials Research and Testing, D-12200 Berlin, Germany
2 Institute of Mechanics, Darmstadt University of Technology, D-64289 Darmstadt, Germany
3 Laboratory of Soil Mechanics, Swiss Federal Institute of Technology, CH-1015 Lausanne, Switzerland

Received January 22, 1999

This work deals with the thermodynamic formulation of constitutive models for materials
whose quasi-static behaviour is governed by internal friction, e.g., dry granular materials.
The process of internal friction is represented here phenomenologically with the help of a
second-order, symmetric-tensor-valued internal variable. A general class of models for the
evolution of this variable is considered, including as special cases a hypoelastic-like form
for this relation as well as the hypoplastic form of Kolymbas (1991). The thermodynamic
formulation is carried out in the context of the Müller-Liu entropy principle. Among other
things, it is shown that for the hypoelastic-type models, a true equilibrium inelastic Cauchy
stress exists. On the other hand, such a stress does not exist for the hypoplastic model due to its
rate-independence and incremental non-linearity. With the help of a slight generalization of the
notion of thermodynamic equilibrium, i.e., to thermodynamic “quasi-equilibrium,” however,
such a Cauchy stress can be formulated for the hypoplastic model. As it turns out, this quasi-
equilibrium for the Cauchy stress represents a thermodynamic generalization of the so-called
quasi-static stress postulated for example by Goddard (1986) in the context of his viscoplastic
model for a frictional-dissipative, and in particular for granular, materials.

1 Introduction

The behaviour of granular materials at low to moderate grain number density or volume fraction, and high
(kinetic) energy, i.e., in the grain inertia regime of Bagnold (1954), where short-term, nearly-elastic colli-
sional interactions between the grains are dominant, appears to be modeled quite well by the application of
Enksog’s theory of dense kinetic gases to the problem (e.g., Lun et al., 1984; Jenkins & Richmann, 1985).
Unfortunately, the extension of this approach to high grain density and moderate to low energy, where fric-
tional interactions become dominant, appears intractable (see discussion in, e.g., Hutter & Rajagopal, 1994).
Limited in this regard as well are computer-based particle-dynamics simulations (see, e.g., Herrmann & Lud-
ing, 1998). As such, simple Mohr-Coulomb-type phenomenological continuum models for the quasi-static
frictional behaviour of granular materials enjoy wide use in the continuum modeling of granular flows (e.g.,
Savage & Hutter, 1989; Gray et al., 1998). Among the generalizations of the Mohr-Coulomb idea in the
realm of granular material modeling, one finds the viscoplastic model of Goddard (1986), and more recently,
the hypoplastic approach of Kolymbas (see, e.g., Kolymbas, 1991, 1993; Gudehus, 1996; Wu et al., 1996).
The formulation of all these models has been based up to this point on statistical mechanical and/or direct
phenomenological constitutive considerations (e.g., rate-independence); the purpose of the current work is
a formulation and analysis of such models from a phenomenological-thermodynamic point of view in the
context of the Müller-Liu entropy principle (e.g., Müller, 1985).
264 B. Svendsen et al.

To begin, the basic constitutive assumptions of the formulation are introduced and discussed (Sect. 2), in
particular the form of the evolution relation for the internal variable accounting for internal friction, and the
assumption that the material behaviour is isotropic. Next, we formulate the corresponding form of the entropy
inequality in the context of the Müller-Liu entropy principle (Sect. 3). The exploitation of the inequality yields
direct restrictions on the forms of the coefficients of the so-called potential and flux one-forms defined on the
manifold of independent constitutive variables (Sect. 3). In particular, the so-obtained restrictions on the flux
one-form yield in turn restrictions on the constitutive forms of the entropy flux and energy Lagrange multiplier
(Sect. 4). The exploitation of the entropy principle is completed by the retrictions obtained from the residual
inequality in the context of thermodynamic equilibrium (Sect. 5). Lastly, the special cases of hypoelastic-
like and hypoplastic-like evolution relations for the internal variable accounting for internal fricition are
investigated and discussed (Sect. 6), and the results compared with previous work.
Before we begin, a word on notation. If W and Z represent two finite-dimensional linear spaces, let
Lin(W , Z) represent the set of all linear mappings from W to Z. If W and Z are inner product spaces,
the inner products on W and Z induce the transpose AT ∈ Lin(Z, W ) of any A ∈ Lin(W , Z), as well as
the inner product A · B := trW (AT B ) = trZ (AB T ) on Lin(W , Z) for all A, B ∈ Lin(W , Z). In the context
of three-dimensional Euclidean vector space V , let a ∧ b := a ⊗ b − b ⊗ a represent the wedge product of
any two Euclidean vectors a, b ∈ V , and I ∈√Lin(V , V ) the second-order Euclidean identity tensor. Further,
use is made of the Euclidean norm |M | := M · M , the “direction,” dir(M ) := M /|M |, and deviatoric part
dev(M ) := M − 13 tr(M ) I , of any M ∈ Lin(V , V ). Finally, we work with the Jacobi hA, B i := AB + B A
and Lie [A, B ] := AB − B A brackets of any two A, B ∈ Lin(V , V ) in what follows.

2 Balance relations and constitutive assumptions

Since we are concerned in this work with isotropic material behaviour (see discussion below), which the
majority of models for granular materials either tacitly or explicitly posit, it is convenient to base the formu-
lation in the current configuration C ⊂ E of the material in three-dimensional Euclidean point space E with
translation vector space V . In this context, the spatial forms

0 = %̇ + % div v ,

0 = % v̇ − div T − b ,

0 = T −T T
, (2.1)

0 = % ε̇ + div q − T · D − r ,

π = % η̇ + div φ − σ ,
of the balance relations for mass, momentum, moment of momentum, internal energy, and entropy, respec-
tively, are relevant. The classical form (2.1)3 of moment of momentum balance is assumed here for simplicity;
a more realistic (and complicated) model, e.g., a micropolar or Cosserat-type one, would account for (average
or effective) particle rotation and the corresponding friction as well, i.e., moment effects, something neglected
here. In (2.1), % represents the mass density, v the spatial velocity, T the Cauchy stress, b the momentum
supply rate density, ε the specific internal energy, q the heat flux, D := 12 (L + LT ) the deformation rate,

L = F˙ F −1 = ∇v = D + W (2.2)

the velocity gradient, F the deformation gradient, W := 12 (L − LT ) the continuum spin, r the internal energy
supply rate density, η the specific entropy, φ the entropy flux, and σ the entropy supply rate density. In the
current context, the mass balance (2.1) can also be expressed as
p
%r = det(F ) % = det(B ) % , (2.3)

where %r represents the mass density of the reference configuration, and


Constitutive models for granular materials: Thermodynamic formulation 265

B := F F T (2.4)

the left Cauchy-Green deformation tensor. Since %r is constant, (2.3) implies that % is a function of F alone
here, as in the standard case. In soil mechanics, one often works with the void ratio e instead of %. This ratio
and the mass density γ of the solid grains determine the mass density % of the granular material as
γ
%= . (2.5)
(1 + e )
Approximating γ as constant, (2.1)1 takes the form

ė = (1 + e ) tr(D ) (2.6)

in terms of ė . This can be integrated to obtain


p
e = (1 + er ) det(B ) − 1 (2.7)

via (2.3), again with γ = γr constant. Such purely continuum modeling of e , or alternatively, the solid volume
fraction ν = 1/(1 + e ), is less sophisticated than that found in other works (e.g., Goodman & Cowin, 1972;
Svendsen & Hutter, 1995). Indeed, on the basis of (2.6), no “micromechanical” effects influence the evolution
of e . In contrast to these earlier works, e is treated here via (2.7) simply as a function of B , and as such
does not appear among the independent constitutive variables to follow (see (2.9) below).
The phenomenological generalization of the Mohr-Coulomb model for internal friction in a granular
material at low energy and high grain volume fraction pursued in this work is based upon a Euclidean frame-
indifferent, stress-like, symmetric-tensor-valued spatial internal variable Z . This variable is by interpretation
associated with the effective “contact” stress in the granular material; as long as the corresponding non-
conservative intergrain forces (depending only on the direction of the relative grain velocities in the rate-
independent case) satisfy action-reaction, statistical mechanical considerations (e.g., Pitteri, 1986; Svendsen,
1999) show that the corresponding stress is in fact symmetric. In the current phenomenological setting, Z is
modeled constitutively by an incremental relation of the form

Z˙ − [Ω , Z ] = Φ . (2.8)

Here, Φ represents the constitutive part of (2.8), and the left-hand side a so-called “corrotational” objective
time derivative of the spatial tensor field Z , Ω being the corresponding spin. For example, Ω is given by
W in the Jaumann case (relevant, e.g., to the case of hypoplasticity).
Assuming next that the collective behaviour of the class of materials under consideration here, and in
particular granular materials, is elastoviscoplastic, constitutive relations in this work take the general form

C = C˜ (θ, F , Z , g, θ,
˙ L) (2.9)

for the dependent spatial constitutive fields C ∈ {T , ε, q , η, φ, Φ}, where g := ∇θ represents the spatial
temperature gradient. The spatial velocity v has been left out of the above relations since these are eliminated
from the constitutive relations by the requirement of material frame-indifference. On the other hand, the
constitutive dependence of the dependent fields on θ,˙ and in particular that of ε̃, is necessary (but of course
not sufficient) for a hyperbolic temperature equation (Müller, 1985).
To incorporate the assumption of isotropic material behaviour into the formulation, consider now the
material symmetry of C˜ . As usual, any change H of local placement preserving density and orientation
represents a symmetry transformation of C˜ if

C˜ (θ, F , Z , g, θ,
˙ L) = C˜ (θ, F H T , Z , g, θ,
˙ L) (2.10)

holds. In particular, for the case of isotropic material behaviour, H is an arbitrary rotation, in which case we
may choose H = R from the (left) polar decomposition F = V R of F , yielding

C˜ (θ, F , Z , g, θ,
˙ L) = C˜ (θ, V , Z , g, θ,
˙ L) = Cˆ (θ, B , Z , g, θ,
˙ L) (2.11)
266 B. Svendsen et al.


from (2.10) via (2.2) and the fact that V = B . Requiring next that Cˆ be material frame-indifferent yields
the restriction
˙ L) = Q ∗ Cˆ (θ, Q B Q T , Q Z Q T , Q g, θ,
C˜ (θ, F , Z , g, θ, ˙ Q LQ T + Q̇ Q T ) (2.12)

for all time-dependent rotations Q in the case of isotropic material behaviour via (2.11), where Q ∗ represents
the (pull-back) action induced by Q . Since Q is arbitrary, we are in particular free to choose Q W Q T = −Q̇ Q T ,
reducing (2.12) to
˙ L) = Q ∗ Cˆ (θ, Q B Q T , Q Z Q T , Q g, θ,
C˜ (θ, F , Z , g, θ, ˙ Q D Q T) (2.13)

via (2.10) and the isotropy of Cˆ . With Q (t) = I for some time t, (2.13) yields the reduced form

C˜ (θ, F , Z , g, θ,
˙ L) = Cˆ (θ, B , Z , g, θ,
˙ D) (2.14)

for C˜ via (2.11). Together, then, (2.8) and (2.14) define the current constitutive class. With the basic con-
stitutive relations in hand, we now turn to the restriction of these relations in the context of the entropy
principle.

3 Entropy inequality and exploitation

As usual, the entropy principle is based on the general local form

π = % η̇ + div φ − σ ≥ 0 (3.1)

of the entropy inequality from (2.1)5 . Assuming that all solutions of the system of algebraic equations arising
from combination of the balance (2.1)1−4 , evolution (2.8) and constitutive (2.14) relations also satisfy the
algebraic inequality obtained from (2.8), (2.14) and (3.1), Liu (1972) showed that (3.1) can be expressed in
the alternative equivalent form1

π = % η̇ + div φ − σ

− λε {% ε̇ + div q − T · D − r }

− λv · {% v̇ − div T − b } (3.2)

− ΛZ · {Z˙ − [Ω , Z ] − Φ}

≥ 0.

Here, linear momentum, and internal energy, balance, as well as the incremental constitutive relation (2.8),
for the structure variables, appear as constraints on the entropy inequality. The quantities λε , λv and ΛZ
represent the corresponding constraint coefficients or “Lagrange multipliers.” In general, each of these may
depend on the independent constitutive fields appearing in (2.14) as well as on v and the external supply rate
densities r and b . On the basis of (2.14) and (3.2), π and the material behaviour will be independent of such
supplies when (i), λε , λv and ΛZ are so, and (2), σ takes the form

σ = λε r + λv · b . (3.3)

Indeed, in this case, the external supply rate densities vanish from (3.2) identically. For the material behaviour
to be independent of the supplies, this must necessarily be the case.
The evaluation of (3.2) on the basis of (2.14) involves the generalized Gibbs’ relations
1 Since we are treating % as a function of the independent constitutive field B here via (2.3), the mass balance (2.1) does not appear
1
as a constraint in (3.2).
Constitutive models for granular materials: Thermodynamic formulation 267

% dη = λε % d ε + P ,
(3.4)
dφ = λε (d q ) + F ,

where
P = Pθ d θ + PB · d B + PZ · d Z + Pg · d g + Pθ˙ d θ˙ + PD · d D ,
(3.5)
F = Fθ d θ + FB · d B + FZ · d Z + Fg · d g + Fθ˙ d θ˙ + FD · d D
represent linear-space-valued one-forms on the manifold of independent constitutive variables appearing in
(2.14). Indeed, substitution of (2.14) into (3.2) yields the form

π = Pθ θ˙ + {hPB , B i + λε T − sym(g ⊗ Pg )} · D

+ {PZ − ΛZ } · Z˙ + Pθ˙ θ̈ + PD · Ḋ − λv · % v̇

+ {[PB , B ] − skw(g ⊗ Pg )} · W + [ΛZ , Z ] · Ω


(3.6)
+ Fθ · g + FB · ∇B + FZ · ∇Z + Fg · ∇2 θ + FD · ∇D + {Pg + Fθ˙ } · ∇θ˙

+ λv · div T + ΛZ · Φ

≥ 0

of the inequality via (2.2) and (3.3) involving the coefficients of P and F . To obtain this last form, use has
been made of the symmetry of PB = %η, B − λε %ε, B via isotropy, the result

Ḃ = LB + B LT (3.7)

from (2.2) and (2.4), as well as the identity

∇θ˙ = ġ + LT g . (3.8)

In the context of (2.14), then, one can exploit (3.6) in the standard fashion. To this end, we first note
that π is linear in the independent quantities v̇ and Z˙ . Consequently, (3.1) could be violated during some
thermodynamic process unless the corresponding coefficients vanish identically, yielding

λv = 0,
(3.9)
ΛZ = PZ ,

for the Lagrange multipliers associated with linear momentum balance and the evolution of Z , respectively,
via (3.4)1 and (3.5)1 . Similarly, note that π is linear in the independent fields θ̈, and Ḋ , yielding the restrictions

Pθ˙ = 0,
(3.10)
PD = 0,

on the coefficients of P . In particular, the first of these takes the alternative form

η, θ˙ = λε ε, θ˙ (3.11)

via (3.4)1 . In a similar fashion, the linearity of π in ∇θ˙ leads to the restriction

Pg + Fθ˙ = 0 (3.12)

on P and F . Further, the linearity of π in ∇2 θ, ∇B , ∇Z , and ∇D yields


268 B. Svendsen et al.

Fg · a ⊗ b = −Fg · b ⊗ a ,

FB · a ⊗ b ⊗ c = −FB · a ⊗ c ⊗ b ,
(3.13)
FZ · a ⊗ b ⊗ c = −FZ · a ⊗ c ⊗ b ,

FD · a ⊗ b ⊗ c = −FD · a ⊗ c ⊗ b ,

for all non-zero a, b, c ∈ V . Lastly, since π is linear in W , and the inequality must be satisfied for all spins
Ω , the restrictions
[PB , B ] = skw(g ⊗ Pg ) ,
(3.14)
PZ Z = Z PZ ,
hold for the case Ω 6= W , and that

[PB , B ] + [PZ , Z ] = skw(g ⊗ Pg ) (3.15)

for the Jaumann case Ω = W . The necessary conditions (3.9), (3.10), (3.12), (3.13), and either (3.14) or
(3.15), reduce (3.6) to its so-called residual form

π = Pθ θ˙ + {hPB , B i + λε T − sym(g ⊗ Pg )} · D + Fθ · g + PZ · Φ (3.16)

for the current constitutive class defined by (2.8) and (2.14). To investigate the restrictions imposed by (3.10)–
(3.15) on the constitutive relations, we now turn to consideration of the associated integrability conditions.

4 Flux one-form and entropy flux

As indicated by the results (3.10)–(3.15) of the last section, (3.2) places restrictions on the coefficients of the
one-forms P and F defined in (3.4)-(3.5). The next step is to transform these restrictions into ones on the
constitutive fields. In particular, we begin by examining the restrictions (3.13) on the flux one-form F . To
this end, consider the condition of isotropy

Q fˆ(θ, Aα , g, θ)
˙ = fˆ(θ, Q Aα Q T , Q g, θ)
˙ (4.1)

on vector-valued constitutive fields fˆ such as φ or q , holding for all rotations Q , with

Aα ∈ {B , Z , D } (4.2)

and α = 1, 2, 3. To exploit (3.13) in the context of (4.1), we employ the approach of Liu (1996) and work
with the “differential” form of (4.1), i.e.,

Ωf = (f , Aα )[Ω, Aα ] + (f , g )Ωg (4.3)

(sum on α = 1, 2, 3) holding for all skew-symmetric tensors Ω. The fact that Ω is skew-symmetric yields
the alternative form
a ∧ f = 2 [(f , Aα )T a, Aα ] + (f , g )T a ∧ g (4.4)
of (4.3) for all non-zero a ∈ V . In particular, the forms of (4.4) holding for φ and λε q yield that

a ∧ k = 2 [(FAα )T a, Aα ] − Fg a ∧ g (4.5)

for the extra entropy flux


k := φ − λε q (4.6)
in terms of FAα and Fg via (3.4)2 . In terms of k , note that F takes the form

F = d k + q d λε (4.7)
Constitutive models for granular materials: Thermodynamic formulation 269

also via (3.4)2 .


Consider next the restrictions (3.13)2,3,4 . In contrast to these, the symmetry of Aα implies

FAα · a ⊗ b ⊗ c = FAα · a ⊗ c ⊗ b (4.8)

for all non-zero a, b, c ∈ V . The only way both (3.13)2,3,4 and (4.8) can be satisfied is if

FAα = 0 (4.9)

i.e., iff FAα vanishes identically for α = 1, 2, 3. As such, (4.5) reduces to

k ∧ a = Fg a ∧ g , (4.10)

again for all a ∈ V . With respect to the Cartesian basis (e1 , e2 , e3 ), this last relation yields that

δij k − ki ej = gi Fg ej − (Fg )ij g (4.11)

for i , j = 1, 2, 3, with δij := ei · ej , ki := ei · k , gi := ei · g, and (Fg )ij := ei · Fg ej . In particular, (4.11) implies


the system
k − k1 e1 = g1 Fg e1

k − k2 e2 = g2 Fg e2 (4.12)

k − k3 e3 = g3 Fg e3
for i = j via the skew-symmetry of Fg from (3.13)1 . Summing these three relations together and rearranging,
one obtains the form
k = 12 Fg g (4.13)
for k . Next, (4.11) yields (Fg )ij gk + (Fg )jk gi = 0 for i 6= j 6= k , and so the system
    
(Fg )23 0 (Fg )12 g1 0
    
    
 (Fg )23 (Fg )31 0   g2 = 0  . (4.14)
    
0 (Fg )31 (Fg )12 g3 0
For arbitrary g, then, the matrix in this last relation must be identically zero, implying

Fg = 0 , (4.15)

and so
k =0 (4.16)
from (4.13). In particular, (4.16) implies the proportionality

φ = λε q (4.17)

between the entropy flux and heat flux for the current constitutive class via (4.6), and so the reduced form

F = Fθ d θ + Fθ˙ d θ˙ = q d λε (4.18)

for F from (3.5)2 , (4.7), (4.9), (4.15) and (4.16). Necessarily, then,

λε = λ̂ε (θ, θ)
˙ (4.19)

holds for λ̂ε since q 6= 0. In addition, note that (3.12), (4.18) and (4.19) result in the form

Pg = −λε, θ˙ q =⇒ %η, g = λε %ε, g − λε, θ˙ q (4.20)

for the g-coefficient of P .


270 B. Svendsen et al.

In summary, the general restrictions (3.13) from the entropy principle together with the isotropy of the
material behaviour lead to the reduced forms
P = Pθ d θ + PB · d B + PZ · d Z − λε, θ˙ q · d g ,
(4.21)
F = q {λε, θ d θ + λε, θ˙ d θ}
˙ ,

of P and F via (3.9), (3.10), (4.18), (4.19) and (4.20) for the current constitutive class. Likewise, the
residual entropy inequality (3.16) reduces to

π = Pθ θ˙ + {hPB , B i + λε T + λε, θ˙ sym(g ⊗ q )} · D + λε, θ q · g + PZ · Φ ≥ 0 , (4.22)

via (4.20) and (4.21)2 . As usual, further restrictions on the form of the constitutive relations can be obtained
from (4.22) in the context of thermodynamic equilibrium, to which we now turn.

5 Thermodynamic equilibrium

The results of this section are based on the condition of thermodynamic equilibrium, i.e.,

πE = π̂E (e ) := π̂(e , 0) = 0 , (5.1)

via the split x = (e , n ) of the independent constitutive variables

x := (θ, B , Z , g, θ,
˙ D) (5.2)

appearing in (2.14) into equilibrium e := (θ, B , Z ) and nonequilibrium n := (g, θ,


˙ D ) subsets. Note that any
dependent constitutive quantity C = C (x ) can be represented in the form
ˆ

Cˆ (x ) = Cˆ E (e ) + Cˆ N (e , n ) , (5.3)

where
Cˆ E (e ) := lim Cˆ (e , n ) (5.4)
n →0
and
Cˆ N (e , n ) := Cˆ (e , n ) − Cˆ E (e ) (5.5)
represent its equilibrium and nonequilibrium parts, respectively. In what follows, we also use the notation

Cˆ |E (e ) := lim Cˆ (e , n ) . (5.6)
n →0

The representation (5.3) as based on (5.4) and (5.5) implies the relations C, eα |E = CE, eα and C, nα = CN, nα
for the partial derivatives of Cˆ with respect to the equilibrium and non-equilibrium variables, respectively,
as well as CN |E = 0. Note that e1 = θ, e2 = B , and so on.
The residual form (4.22) of π fulfills (5.1) in particular since Φ is a production-like quantity, i.e.,

ΦE = 0 =⇒ Φ = Φ̂N (θ, B , Z , g, θ,
˙ D) (5.7)

holds. With the help of the isotropic form

Φ = φ0 I + φ1i Aα + φ2i Aα
2

+ φ1ij hAα , Aβ i + φ2ij hAα


2
, Aβ i + φ3ij hAα , Aβ2 i (5.8)

+ φ3i hg ⊗ g, Aα i + φ4i hg ⊗ g, Aα
2
i,

for Φ, the equilibrium part of Φ is given by


Constitutive models for granular materials: Thermodynamic formulation 271

ΦE = φ0 |E I + φ11 |E B + φ12 |E Z + φ21 |E B 2 + φ22 |E Z 2


(5.9)
+ φ112 |E hB , Z i + φ212 |E hB 2 , Z i + φ312 |E hB , Z 2 i .
For this to vanish, then, the restrictions
φ11 |E = 0, φ12 |E = 0, φ21 |E = 0, φ22 |E = 0,
(5.10)
φ112 |E = 0, φ212 |E = 0, φ312 |E = 0,
on the corresponding coefficients appearing in (5.8) pertain in general. For the particular cases of Φ examined
in the next section, these conditions are in fact satisfied identically.
The condition (5.1) of thermodynamic equilibrium motivates the expansion

π̂(x ) = π̂N (e , n ) = π̂, n (e , 0) · n + 12 n · π̂, nn (e , 0)n + · · · (5.11)

of π̂ about thermodynamic equilibrium via (5.3). In the context of (5.11), n = 0 represents a minimum of π̂
when the necessary conditions

π, n |E = (π, g |E , π, θ˙ |E , π, D |E ) vanishes ,
 

 π, gg |E π, gθ˙ |E π, gD |E 

 


 
 (5.12)

 π, gθ˙ |E π, θ˙θ˙ |E π, θ˙D |E 

π, nn |E = 
 
 non-negative definite ,

 


 

 π | π, θ˙D |E π | 
, gD E ,DD E

hold. The evaluation of (5.12) in what follows takes advantage of the fact that, for any scalar- or symmetric-
tensor-valued constitutive relation Cˆ ,
C, g |E = 0 ,

C, gθ˙ |E = 0, (5.13)

C, gD |E = 0,

follow from the isotropy of Cˆ . Indeed, in this case, Cˆ is “quadratic” in g. Since π̂ as given in (4.22) is
such an isotropic function, (5.13) holds in particular for it.
In the context of (2.14) and (4.22), the restrictions (5.12) are evaluated as follows. Begin for example
with the case π, g |E = 0. Given that λεE, θ 6= 0, q E vanishes identically via isotropy, and PZ |E 6= 0, (5.12)1 yields

Φ, g |E = 0 (5.14)

in this case. On the other hand, as just discussed, this also follows directly from the fact that the coefficients
of Φ in (5.8) are isotropic constitutive functions, and so in particular “quadratic” in g. Next, the case π, θ˙ |E = 0
from (5.12)1 implies
Pθ |E = −PZ |E · Φ, θ˙ |E . (5.15)
And π, D |E = 0 leads to the result

λεE T E = −hPB |E , B i − (Φ, D )T |E PZ |E (5.16)

for the equilibrium Cauchy stress T E . The result (5.15) leads in particular to the equilibrium form

P |E = −(PZ · Φ, θ˙ )|E d θ + PB |E · d B + PZ |E · d Z (5.17)

for P from (4.21)1 , implying

%(d η)|E = λεE %(d ε)|E − PZ |E · Φ, θ˙ |E d θ + PB |E · d B + PZ |E · d Z (5.18)


272 B. Svendsen et al.

for the generalized Gibbs’ relation (3.4)1 . So, except for the terms involving the inelastic behaviour, this last
result takes the form of the classical (i.e., thermostatic) Gibbs’ relation, identifying

λεE = λ̂εE (θ) = θ−1 (5.19)

as the absolute coldness (e.g., Müller, 1985).


To cast the results up to this point in a more familiar form, it is useful to introduce the referential free
energy density
ψ := εE − θηE = ψ̂(θ, B , Z ) . (5.20)
In terms of ψ, we then have
− θP |E = % d ψ + %ηE d θ (5.21)
for the equilibrium part P |E of P from (3.4)1 and (5.19). In the equilibrium context, the restriction (3.14)
reduces to
ψ, B B = B ψ, B ,
(5.22)
ψ, Z Z = Z ψ, Z ,
for the case Ω 6= W , and (3.15) to
[ψ, B , B ] + [ψ, Z , Z ] = 0 (5.23)
for the Jaumann case Ω = W in terms of ψ in the equilibrium context, both representing restrictions on the
form of ψ. In addition, the relations (5.15) and (5.21) then yield

ηE = −{ψ, θ + ψ, Z · Φ, θ˙ |E } (5.24)

for the equilibrium part of the entropy density η. As such, in addition to the usual “elastic” part −ψ, θ ,
frictional processes could contribute in equilibrium to η via a θ-dependence
˙ of Φ̂. Lastly, the result (5.16)
reduces to
T E = % hψ, B , B i + % (Φ, D )T |E ψ, Z (5.25)
for T E via (5.21). The first term on the right-hand side clearly represents the elastic, and the second the
frictional or contact, contribution to T E , as mediated by the dependence of Φ̂ on D . To summarize, the results
(5.19), (5.24) and (5.25) yield the forms

λε = θ−1 + λεN ,

η = −ψ, θ − ψ, Z · Φ, θ˙ |E + ηN , (5.26)

T = % hψ, B , B i + % (Φ, D )T |E ψ, Z + T N ,

for the decomposition of λε , η, and T , respectively, into equilibrium and non-equilibrium parts, via (5.3).
In particular, since λε depends constitutively only on θ and θ, ˙ note that λεN could be expressed in general
as quasi-linear form in θ.
˙ Further, T N represents a viscous contribution to the Cauchy stress including for
example the Bagnold (1954) contribution to the stress in the grain-inertia regime.
We turn next to the condition (5.12)2 . In particular, this yields the restriction

2 sym(q , g )|E + θ(Φ, gg )T |E ψ, Z non-positive definite (5.27)

on the symmetric part of the equilibrium thermal conductivity tensor q , g |E via (5.19). Further, it requires that

ε, θ˙ |E ≤ −θλε, θ˙ |E θεE, θ − ψ, Z · θΦ, θ˙θ˙ |E (5.28)

hold for ε, θ˙ |E via (3.11). In the particular case that Φ̂ is independent of θ, ˙ for example, this last result implies
ε
that ε, θ˙ |E will be positive only if λ, θ˙ |E is negative, and so lead to a hyperbolic temperature evolution relation.
Beyond (5.27) and (5.28), (5.12)2 leads to the condition
Constitutive models for granular materials: Thermodynamic formulation 273

2 sym(T N, D )|E − (Φ, D D )T |E ψ, Z non-negative definite (5.29)

on the symmetric part of the “viscosity” tensor T N, D . Finally, the sole non-zero “off-diagonal” element of
π, nn |E takes the form

θπ, θ˙D |E = −θ−1 ε, D |E + θλε, θ˙ |E {hεE, B , B i + (Φ, D )T |E εE, Z } + θ(λε T ), θ˙ |E − (Φ, D θ˙ )T |E ψ, Z (5.30)

via (3.11), (5.28) and (5.29). Its Cartesian components satisfy the restriction

(π, θ˙D |E )ij (π, θ˙D |E )kl ≤ (π, θ˙θ˙ |E ) (π, D D |E )ijkl (5.31)

in the context of (5.12)2 .


This last restriction completes the investigation and exploitation of thermodynamic equilibrium. Now we
turn to the investigation and discussion of two major special cases of the above formulation having to do
with the constitutive form for Φ in light of the above general results.

6 Hypoelastic and hypoplastic special cases

The exploitation of thermodynamic equilibrium presented in the last section is based on the tacit assumption
that π̂(e , n ) as given by (4.22) is k -times continuously differentiable (k ≥ 2) in n at n = 0. This is of course
the case only when all constitutive functions, and in particular Φ̂, are such. For example, the “hypoelastic”
form
Φ̂(B , Z , D ) = L̂ (B , Z )D (6.1)
for Φ̂ fulfills this requirement since it is linear in D . Here, L̂ (B , Z ) represents a fourth-order-tensor-valued
isotropic function. In the soil mechanics context, the dependence of (6.1) (and that of the hypoplastic form
(6.3) considered below) on B is via one on the void ratio e as given by (2.7). From (6.1) follow in particular
the forms
ηE = −ψ, θ
(6.2)
T E = % hψ, B , B i + % L ψ, Z
T

for the equilibrium entropy and Cauchy stress via (5.24) and (5.25), respectively, with ψ given by (5.20).
Models for granular materials based on (6.1) in the realm of soil mechanics have been considered by, e.g.,
Stutz (1972), Romano (1974), as well as Davis & Mullenger (1978), and criticized by Gudehus (1979). The
basic problem with (6.1) is that it cannot capture the fact that the material behaviour of granular materials is
in general different in extension than in compression.
In contrast to (6.1), the hypoplastic form

Φ̂(B , Z , D ) = L̂ (B , Z )D + Nˆ (B , Z ) |D | (6.3)

for Φ̂ does account for the fact that the material behaviour of granular materials is in general different in
extension than in compression, i.e., via the second term non-linear in D . Particular forms of L̂ and Nˆ include
those
Z ⊗Z
L̂ (B , Z ) = c1 (B ) tr(Z ) I + c2 (B )
tr(Z )
(6.4)
Z2 dev(Z )2
N (B , Z ) = c3 (B )
ˆ + c4 (B )
tr(Z ) tr(Z )
used by Wu et al. (1996) in modeling the failure of various types of soils. Here, I represents the fourth-order
identity tensor, and the material coefficients c1−4 depend on B through the void ratio e via (2.7). These
coefficients are determined, e.g., with the help of triaxial extension-compression tests. Note that, in contrast
to (6.1), (6.3) is not (Fréchet) differentiable in D at D = 0 since the Euclidean norm is not. Indeed, we have

Φ̂, D (B , Z , D ) = L̂ (B , Z ) + Nˆ (B , Z ) ⊗ dir(D ) . (6.5)


274 B. Svendsen et al.

Consequently, the concept of thermodynamic equilibrium introduced and exploited in Sect. 6 is not applicable
to the hypoplastic case; all non-equilibrium results, i.e., those from Sects. 4–5, of course still apply.
Formally, at least, one can deal with this difficulty by generalizing the notion of thermodynamic equi-
librium presented in the last section to one of “quasi-equilibrium” with the help of so-called non-standard
analysis (e.g., Robert, 1988). In particular, this involves a generalization of the limit (5.6) to the form
Cˆ |QE (e , ) := lim Cˆ (e , n ) (6.6)
n →
with  := (0, 0, A), A being an element of the set of all infinitesimal symmetric tensors, i.e., symmetric
tensors whose magnitude is infinitely near (but not equal to) zero. In the context of non-standard analysis,
such quantities are indicated by the notation A ' 0, where 0 represents the unique “standard” infinitesimal
(i.e., number, vector, tensor). Analogous to (5.5), we then have
Cˆ QN (e , n , ) := Cˆ (e , n ) − Cˆ QE (e , ) . (6.7)
In particular, (6.3) and (6.6) imply that ΦQE is infinitesimal, i.e., ΦQE ' 0. Likewise, πQE ' 0 follows from
(4.22) and (6.3) via (6.6). The corresponding analysis of thermodynamic quasi-equilibrium is then based on
the generalization2
π̂(x ) ' π̂, n (e , ) · (n − ) + 12 (n − ) · π̂, nn (e , )(n − ) + · · · (6.8)
of (5.11), with
π, n |QE ' 0,
(6.9)
π, nn |QE ' non-negative definite ,
the corresponding generalizations of (5.12). In particular, the second of these last conditions means that
π, nn |QE is infinitely close to being non-negative definite. On this basis, then, all results of the previous section
generalize accordingly; in particular,
T QE ' % hψ, B , B i + % (Φ, D )T |QE ψ, Z
(6.10)
' % hψ, B , B i + % L ψ, Z + (N · ψ, Z ) dir(A)
T

generalizes the equilibrium form (5.25) for the Cauchy stress to the case of quasi-equilibrium, the second
expression following from (6.3).
Comparing in particular (6.10) to previous work, one sees that it represents a thermodynamic general-
ization of models for the “quasi-static” stress in granular materials discussed by McTigue (1982), Sayed &
Savage (1983), and especially Goddard (1986). Comparing the current formulation with this latter work, one
establishes in particular that dir(A) appearing in (6.10) is nothing other than the (quasi-equilibrium form) of
the so-called versor
E := dir(D ) (6.11)
of Goddard (1986). Consequently, such models represent special cases of the hypoplastic approach. These
models were formulated from the point of view that the corresponding special cases of T QE obtained by them
should describe the stress in a granular material that is “at failure” everywhere as D “vanishes”. The remaining
“kinetic” (i.e., non-equilibrium) part T QN of T then determines any rate-dependent behaviour of the granular
material, and in particular that in the grain-inertia regime of Bagnold (1954). For both cases considered in this
section, the corresponding constitutive model obtained for T represents an elastoviscoplastic type of material
behaviour. As shown in the special case of (6.10) considered by Goddard (1986), such models for T QE can
account for both Mohr-Coulomb or more general yield-type behaviour as well as normal-stress effects. A yet
more encompassing formulation of such material behaviour would involve of course stability considerations,
the subject of future work.

Acknowledgements. We thank I.-S. Liu for reviewing the first version of the paper and providing many helpful comments leading to its
improvement. This work was partially supported by the Max Planck Society and the Alexander von Humboldt Foundation.
2 The notation a ' b indicates that a = b + , with  infinitesimal.
Constitutive models for granular materials: Thermodynamic formulation 275

References

1. Bagnold RA (1954) Experiments on a gravity free disperion of large solid spheres in a Newtonian fluid under shear, Proc. Roy.
Soc. London A225, 49–63
2. Davis RO, Mullenger G (1978) A rate-type constitutive model for soil with a critical state, Int. J. Numer. Anal. Methods Geomech.
2, 255–282
3. Gray JMNT, Weiland M, Hutter K (1999) Gravity-driven free surface flow of granular avalanches over complex topography, Proc.
Roy. Soc. London, in press
4. Goddard JD (1986) Dissipative materials as constitutive models for granular materials, Acta Mech. 63, 3–13
5. Goodman DC, Cowin S (1972) A theory of granular materials, Arch. Rat. Mech. Anal. 44, 249–266
6. Gudehus G (1979) A comparison of some constitutive laws for soils under radially symmetric loading and unloading, in: Wittke W
(ed) Proceedings of the 3rd International Conference on Numerical Methods in Geomechanics, pp 1309–1323
7. Gudehus G (1996) A comprehensive constitutive relation for granular materials, Soils and Foundations 36, 1–12
8. Herrmann H, Luding S (1998) Modeling granular media on the computer, Cont. Mech. Thermodyn. 10, 189–231
9. Hutter K, Rajagopal KR (1994) On flows of granular materials, Cont. Mech. Thermo. 6, 81–139
10. Jenkins JG, Richman MW (1985) Grad’s 13-moment system for a dense gas of inelastic spheres, Arch. Rat. Mech. Anal. 87,
355–377
11. Kolymbas D (1991) An outline of hypoplasticity, Ing. Arch. 61, 143–151
12. Kolymbas D (1993) Introduction to Hypoplasticity, in Kolymbas D (ed) Modern Approaches to Plasticity, pp 213–223
13. Liu I-S (1972) Method of Lagrange muptipliers for exploitation of the entropy principle, Arch. Rat. Mech. Anal. 46, 131–148
14. Liu I-S (1996) On entropy flux-heat flux relations in thermodynamics with Lagrange multipliers, Continuum Mech. Thermodyn. 8,
247–256
15. Lun CKK, Savage SB, Jeffrey DJ, Chepurniy N (1984) Kinetic theories for granular flow: inelastic particles in Couette flow and
slightly inelastic particles in a general flow field, J. Fluid Mech. 140, 223–256
16. McTigue DF (1982) A non-linear constitutive model for granular materials: application to gravity flow, J. Appl. Mech. 49, 291–296
17. Müller I (1985) Thermodynamics, Pitmann
18. Pitteri M (1986) Continuum equations of balance in classical statistical mechanics, Arch. Rat. Mech. Anal. 94, 291–305
19. Robert A (1988) Non-standard Analysis, John Wiley & Sons
20. Romano MA (1974) A continuum theory for granular media with a critical state, Arch. Mech. 20, 1011–1028
21. Savage SB, Hutter K (1989) The motion of a finite mass of granular material down a rough incline, J. Fluid Mech. 199, 177–215
22. Sayed M, Savage SB (1983) Rapid gravity flows of cohensionless granular materials down incline chutes, ZAMP 34, 84–100
23. Šilhavý M (1997) The Mechanics and Thermodynamics of Continuous Media, Springer-Verlag
24. Stutz A (1972) Comportment elasto-plastique des milieux granulariesn, in: Sawczuk A (ed) Foundations of Plasticity, pp 33–49,
Noordhoff
25. Svendsen B (1999) A statistical mechanical formulation of continuum fields and balance relations for granular and other materials
with internal degrees of freedom, CISM Course on Kinetic and Continuum Mechanical Approaches to Granular and Porous Media,
in press, Springer-Verlag
26. Svendsen B, Hutter K (1995) On the thermodynamics of a mixture of isotropic materials with constraints, Int. J. Engrg. Sci. 33,
2021–2054
27. Truesdell C, Toupin R (1960) The classical field theories of mechanics, in Handbuch der Physik, V. III/1, Springer-Verlag
28. Truesdell C, Noll W (1965) The non-linear field theories of mechanics, in Handbuch der Physik, V. III/3, Springer-Verlag
29. Wu W, Bauer E, Kolymbas D (1996) Hypoplastic constitutive model with critical state for granular materials, Mech. Mat. 23, 45–69

You might also like