A Canonical Framework for Modeling Elasto-Viscoplasticity in

Complex Fluids
Christopher J. Dimitriou, Gareth H. McKinley
77 Massachusetts Avenue, Cambridge, MA 02139

Abstract

A comprehensive framework for modeling elasto-viscoplasticity in complex fluids is dis-

cussed. It is based on the plasticity mechanism of kinematic hardening, which is widely
accepted in solid mechanics and accounts for transient yielding processes. We discuss a
simple one dimensional variant of the model, as well as a fully three-dimensional, frame-
invariant and thermodynamically admissible version of the model. Predictions for several
canonical rheometric test protocols are provided. We also discuss possible extensions to
account for additional rheological complexities exhibited by real fluids, such as thixotropy,
nonlinear elasticity and normal stress differences. We find that this framework has several
advantages over the more commonly used elastic Bingham-like or elastic Herschel Bulkley
models for describing elasto-viscoplasticity. First, the model can account for behavior over a
much wider range of viscometric test conditions. Second, it eliminates the flow/no flow cri-
terion inherent in Bingham-like constitutive laws, which frequently requires regularization.
Third, it is a flexible framework and allows for implementation of additional complexities,
including thixotropic behavior and other nonlinear rheological features.
Keywords: Yield-Stress Fluids, Viscoplasticity, Constitutive modeling

1. Introduction

Elasto-Viscoplastic (or EVP) behavior is frequently exhibited by a range of complex

fluids, including many foods, consumer products and industrial materials [1]. EVP behavior
is generally characterized by a yield-like transition that occurs when the stress imposed on
a soft solid exceeds a critical value. Below this stress, the material behaves primarily as
Preprint submitted to Journal of Non-Newtonian Fluid Mechanics July 27, 2018
an elastic solid, whereas above the critical stress the material flows like a liquid. Complex
fluids that exhibit EVP behavior are often called “yield stress fluids”, and a wide range of
fluids are known to exhibit this behavior [2, 1].
Most constitutive models that are used to describe viscoplastic behavior are based on the
well-known Bingham or Herschel-Bulkley models. These two models assume the following
relation between the magnitude of the stress |σ| and the magnitude of the strain rate |γ̇|,

|σ| = σy + k|γ̇|m , (1)

where σy is the critical yield stress, k is a viscosity coefficient (with units Pa.sm ), and m is
a power law exponent. For the case of m = 1, the constitutive model above reduces to the
simplest model for yield stress fluids, the Bingham model, with k → µp being the plastic
viscosity.
Oldroyd [3] was the first to propose a frame-invariant, three-dimensional generalization
of the Bingham model. This Oldroyd-Bingham model (we borrow this term from Goddard
[4]) for EVP materials specifies the following relationship between the Cauchy stress T and
the rate of deformation tensor D
√
T0 = 2GE If |T0 | < 2σy (2)
√ D √
T0 = 2kD + 2σy If |T0 | ≥ 2σy (3)
|D|
In the above equations, T0 is the deviatoric Cauchy stress defined as T0 ≡ T − 31 tr(T).
The tensor E is the infinitesimal strain tensor defined as E ≡ 21 (H + H| ), where H is the
displacement gradient, defined as H ≡ ∇u. The vector u is the displacement of material
points from an undeformed reference configuration X, and is given by u = x − X. The
vector x describes the spatial location of the material points X after a deformation. With
the notation adopted herein,1 the components of the displacement gradient H are given by

1
Throughout this manuscript we adopt the notation employed in [5] to describe various tensorial con-
stitutive models. However we include footnotes in certain instances where it is necessary to clarify the
differences between the continuum mechanics notation employed in [5], and the notation typically employed
in the rheology literature, such as in [6] and [7].
2
 ∂ui
Hij = ∂Xj
, where ui are the cartesian components of the displacement vector u, and Xj are
the cartesian components of the material reference points X.
The rate of deformation tensor is defined as D ≡ 12 (L + L| ), where L is the velocity
∂vi 2
gradient defined as L ≡ grad v. The components of L are given by Lij = ∂xj
, where vi
are the cartesian components of the velocity vector v, and xj are the cartesian components
of the spatial points x.3 In many applications, the elastic shear modulus of the material is
√
assumed to be infinite, so that for T0 ≤ 2σy the material is perfectly rigid and there is no
deformation, with E = 0 and D = 0.
The above Oldroyd-Bingham model is analogous to other frame-invariant, tensorial con-
stitutive models used for viscoelastic fluids (e.g. the Upper-Convected Maxwell model or
Oldroyd-B model). It can be used for numerical simulations of more complex flow scenarios,
which necessarily require a model of tensorial form. There are therefore a large number
of contributions to the literature in which versions of this model have served as a basis
for understanding complex flows of viscoplastic liquids [8, 9, 10, 11, 12, 13, 14, 15]. From
that perspective, it serves as the current canonical framework for modeling viscoplasticity
in complex fluids.
Despite its widespread use, the Oldroyd-Bingham model is extremely simplified, and has
several known deficiencies that prevent it from fully describing the response of real materials.
One drawback of the model form given in Eqs 2-3 is that it gives unphysical predictions for
transient viscometric responses during startup of steady flow or oscillatory flow [16, 17].
For example, when a constant deformation rate D is applied on the material, there is an
initial elastic deformation with the stress T0 increasing linearly with the imposed strain.

2
The tensor L used here (and in [5]) is the same as the tensor (∇v)† in [6]
3
Based on the approach followed in [5], we make a distinction between the grad and ∇ operator for
taking gradients of vectors. The different operators are utilized to distinguish between “spatial gradients”
∂
and “material gradients”. Hence ∇ ≡ ∂X is a gradient with respect to some “material” undeformed reference
∂
configuration, and grad ≡ ∂x is a gradient with respect to some “spatial” deformed configuration of the
body. From Eq. 9.2 in [5] and Eq. 9.2-9 in [6], ∇v = (gradv)F, where F is the deformation gradient tensor.
Note that the original Oldroyd model described in [3] appears to use the grad operator to define both E
and D (or ij and eij respectively, using the notation in [3])
3
When the material reaches the yield point, the stress then exhibits a jump discontinuity
due to the presence of the rate-dependent term 2kD in Eq. 3. These discontinuities can be
avoided by decomposing the material deformation into separate elastic and plastic parts,
which is a commonly taken approach to model plasticity. More recent models, such as those
used Marmottant and Graner [18], Benito et al. [19] and Saramito [20, 21] have taken this
approach.
Another drawback of the Oldroyd-Bingham framework is that it only predicts linear
elastic behavior when the stress in the material is below the critical stress or yield stress σy .
Many real yield stress fluids such as soft gels do not behave in this manner. Their behavior is
usually elastically-dominated pre-yield, but some important aspects of viscous or dissipative
behavior are still present. For example, most yield stress fluids will exhibit a finite (albeit
small) value of the loss modulus G00 (ω) over a wide range of oscillatory frequencies, indicating
that the material experiences dissipative losses and some irreversible deformation below the
yield stress. For example, Carbopol microgels, which are a commonly-studied ideal yield
stress fluid [22], exhibit transient elastoplastic creeping flow below their yield stress [23, 24].
This type of behavior is ubiquitous among yield stress fluids [2], and the Oldroyd-Bingham
framework cannot account for it.
An important issue pertaining to numerical simulations of the Oldroyd-Bingham model
is the determination of the correct regularization protocol of the equations given in Eq. 3-
2 [25, 15, 26]. The discontinuous nature of the equations for Bingham-like models require
solving for the a priori unknown spatial location of a yield surface, across which the material
changes from its yielded to its unyielded state [25, 15, 26, 27]; this can be computationally
expensive. To avoid these numerical issues, the form of the equations given in Eqs. 3-2 can
be regularized by specifying a very large but finite zero shear rate viscosity at low values of
|D|. The Papanastasiou model is one frequently used regularization scheme [28]. With the
specification of a large zero shear rate viscosity, the model then predicts a very slow, but
non-zero, creeping flow which will occur at all stresses below the yield stress σy . However,
this is only a first order approximation of the behavior of real materials. Measurements in
real yield stress fluids show a time-varying apparent viscosity at low stresses below the yield
4
stress [24]. Creeping flow below the yield stress is therefore a transient phenomenon, and
not a steady state one which can be characterized by a single viscosity parameter [29] .
In a recent publication, Dimitriou et al. introduced a model based on the plasticity
mechanism of kinematic hardening (KH) to capture the response of an “ideal” yield stress
fluid to a wide variety of rheological deformation histories [30]. Among the test protocols
considered were large amplitude oscillatory shear (LAOS) tests, which have proven difficult
to predict accurately with a simple constitutive law, due to their inherently nonlinear and
time-varying nature as the stress amplitude periodically varies across the yield envelope of
the material. The “KH model” that was introduced in [30] and [31] showed several quanti-
tative improvements over a simpler elastic-Herschel-Bulkley model for describing Carbopol
gels undergoing transient deformation histories. The goal of the present work is to provide
a thorough discussion of the rheological predictions of this KH model and its extensions,
thereby elucidating the benefits that it can provide as a future canonical framework for use
in numerical simulations or theoretical applications.
We first discuss a simple, one-dimensional version of the model and derive analytical
expressions for the model predictions in steady shear, creep and small amplitude oscillatory
shear. Then we formulate a fully frame-invariant, thermodynamically admissible, three-
dimensional version of the model. We will discuss how to convert the model from its natural
Lagrangian framework to an Eulerian framework, which is more appropriate for numerical
simulations of steady flow scenarios. Model predictions for several different common rheo-
metric flow histories (shear, extension) are provided, and we identify the similarities between
this model and other tensorial constitutive models for complex fluids.
As a canonical framework, this KH model possesses several improvements over the
Oldroyd-Bingham framework. It is able to predict the slow irreversible transient creeping
deformation that many yield-stress fluids exhibit. It has also been successful at predict-
ing experimental data of EVP materials under large amplitude oscillatory shear [30, 31].
Furthermore, it eliminates the stress discontinuity and conditional statement required for
identifying if/when yielding behavior is present in a material. Finally, it is a flexible frame-
work, and allows for other features of a real material’s response to be implemented (e.g.
5
nonlinear elasticity, or even thixotropy [4, 32]).

2. Results and Discussion

We discuss two different formulations of the KH model. This model was introduced in
previous work [30], and was successful at predicting the response of a Carbopol microgel to
a number of different steady and oscillatory shearing protocols. We reproduce the model
formulation here in two parts: First, a simplified one-dimensional version of the model is
introduced. We then show how it can be extended to a generalized three-dimensional, frame-
invariant, thermodynamically admissible version of the KH model, which is also provided in
the appendix of Dimitriou et al. [30].

2.1. 1D formulation of model

A fundamental mathematical assumption that is used for the KH model is the decompo-
sition of the deformation within the material into two components - an elastic contribution
and a plastic contribution. In the case of the three-dimensional version of the KH model,
the material’s deformation gradient F is multiplicatively decomposed into plastic and elastic
parts using the Kroner decomposition (see Sec. 2.2 for additional details). For the simpler
1D (shear) version of the KH model however, we additively decompose the total shear strain
γ as
γ = γe + γp , (4)

where γ e is the elastic, reversible part of the shear strain, and γ p is the irreversible plastic
part of the shear strain. The elastic strain is related linearly to the shear stress σ through
a shear modulus G
σ
γe = . (5)
G
The magnitude of the rate of change of the plastic strain, |γ̇ p |, is given by a power law “flow
rule” of the form  (1/m)
p |σ − σback |
|γ̇ | = , (6)
k

6
where σback is a quantity known as the back stress in the material (discussed in further detail
below), k is the material consistency, and m is a power law exponent. The stress difference
σ − σback is also termed the effective stress which is responsible for driving the plastic flow
in the material. The direction of plastic strain, np , must also be specified:
σ − σback
np = . (7)
|σ − σback |
The rate of change of plastic strain is then given as the product of its direction and magni-
tude, i.e. γ̇ p = np |γ̇ p |. The plastic strain rate is therefore assumed to be co-directional with
the effective stress in the material.
A critical component of the KH framework is its ability to account for evolution in the
value of the back stress, σback , as a function of flow history. The back stress represents the
center of the yield surface of the material in stress space. It can be related to an internal
strain-like variable A, through a back stress modulus C,

σback = CA . (8)

This internal or “hidden” variable A can be related to variations in the defect energy within
the deforming material; this will be discussed further in the following section. For the specific
model discussed in this section, the variable A evolves according to the Armstrong-Frederick
[33] kinematic hardening equation,

Ȧ = γ̇ p − qA|γ̇ p | , (9)

where q is a dimensionless constitutive parameter which governs the change of the material
structure due to shearing. As we show below, 1/q is also a measure of the yield strain of the
material. In what follows we discuss the predictions of the KH model under several different
types of simple one-dimensional flows.

Steady simple shear. Under steady shearing conditions, i.e. a nonzero constant value of γ̇ p ,
with Ȧ = 0, the KH model given by Eqs. 6-9 reduces to the canonical Herschel-Bulkley
relationship between the magnitude of the stress and the shear rate,

|σ| = C/q + k|γ̇ p |m , (10)

7
where the ratio of the KH model parameters C/q = σy is the steady state yield stress in the
material. For almost all actual yield stress materials we expect m < 1. While the Herschel-
Bulkley model only defines the stress for a nonzero shear rate, in what follows we show
how the KH model can provide predictions for deformations occurring at stresses below the
critical yield stress C/q.

Creep. For applied stresses below the steady state yield stress, σ0 < σy , the KH model
predicts a slow transient creeping flow. Under this creeping deformation, it can be shown
(derivation provided in the appendix) that the instantaneous value of the apparent viscosity
σ0
defined as η + (t) ≡ γ̇ p (t)
is given by the following expression

 1/(1−m)
+ t
η ' ηc , (11)
tc
where tc is a characteristic time scale and ηc is a characteristic viscosity scale. These scales
are found to be:
 1/m
k
tc ≡ , (12)
C
 1/m   1/(1−m)
k qσ0  1 − m
ηc ≡ σ0 1− . (13)
C C m
The expression in Eq. 11 holds for all values of m < 1 (and in the singular limit when m = 1,
it can be shown that η + grows exponentially [34]). The power law growth of the apparent
viscosity over time during creep has been observed previously in EVP materials [23]. For
carbopol microgels, several authors have measured the exponent which characterizes the
growth of viscosity over time. A range of values for this exponent have been experimentally
observed across several different carbopol formulations [30, 23, 24, 35, 29]. One subtle
aspect of the KH model’s predicted creep behavior is that for values of m < 1, the quantity
1/(1 − m) is greater than 1, resulting in the plastic strain in the material increasing for
all times, but approaching an asymptotic upper limit. This behavior follows from Eq. 11.
With γ̇ p ∼ t1/(m−1) , and m < 1, the integral of the plastic strain rate in the limit of t → ∞
converges, so the plastic strain approaches a finite value at long times. Thus, the model
does not predict an unbounded power law growth in the total material strain over time.
8
Some recent experimental work indicates that the strain can grow in an unbounded fashion,
at least for one particular carbopol formulation, and within a range of moderate stresses
which are below the yield stress [35]. However this behavior appears to vary from material
to material, and is not a ubiquitous feature observed across all EVP materials. In the work
by Møller et al. [24], some EVP materials exhibited sufficiently fast viscosity growth for the
strain to approach an asymptotic limit.

Small amplitude oscillatory shear (SAOS). For the special case where m = 1, analytical
expressions for G0 (ω) and G00 (ω) for the KH model under small amplitude oscillatory stress
(σ = σ0 cos ωt) can be found (see appendix for derivation):

G0 (ω) (1 + G/C) + (kω/C)2

= , (14)
G (1 + G/C)2 + (kω/C)2
G00 (ω) (G/C)(kω/C)
= . (15)
G (1 + G/C)2 + (kω/C)2

The expressions for G0 (ω) and G00 (ω) above are the same as those for a standard 3-parameter
viscoelastic model, with a spring of modulus G in series with a Kelvin element. The Kelvin
element consists of a spring with modulus C and a dashpot with viscosity coefficient k,
corresponding to a retardation time λ2 = k/C.
For the case where m < 1, the intrinsically nonlinear form of the plastic flow rule in
Eq. 6 results in very strict constraints on the limits of linearity in the imposed deformation.
The expressions for the linear viscoelastic moduli reduce to:

G0 (ω) = G, (16)

G00 (ω) = 0 . (17)

While the expressions above indicate that in the limit of very small strains the KH model
predicts zero energy dissipation, we will show in the next section that for practical values of
the material coefficients, small but finite values of G00 are predicted at small strains (order of
1%) because of additional small but non-negligible contributions from the inherently nonlin-
ear nature of the Armstrong-Frederick kinematic hardening expression in Eq. 9. However,
these values asymptotically tend to zero as the strain amplitude approaches zero. It is also
9
possible to modify the form of the viscoelastic model below the yielding point, in order to
accurately capture the experimentally observed frequency-dependent variations on the linear
viscoelastic behavior; this has been demonstrated previously in [31]. Although not discussed
in detail here, Section 2.4.1 shows various mechanical analogs for simplified versions of the
KH model.

LAOS and other transient experiments. Predictions of the model for LAOS (large ampli-
tude oscillatory shear) experiments are discussed in detail in our previous articles [30, 31].
For completeness we briefly review the results from these experiments, because they were
the original basis for introducing the mechanism of kinematic hardening to model EVP be-
havior of yield stress fluids. Specifically, the kinematic hardening mechanism captures the
“Bauschinger effect” [36], which is made apparent when data from LAOS experiments are
plotted in cyclic stress-strain (Lissajous) curves. The Bauschinger effect results in a distinc-
tive rhomboidal shape of the stress-strain curves [31]. To predict this behavior, a constitutive
model must predict a reduction of the yield stress in a direction opposite to the direction
the material is being deformed in. For example, when the material is being deformed under
shear in a given direction, reversal of the deformation will cause the material to yield at a
lower stress in the opposite direction. This requires an understanding of the yield stress as
a directional quantity, even in the case where the model is used to predict one-dimensional
deformations. It is for this reason that the back stress can have either positive or negative
values.
The specific form of the kinematic hardening equation which is used here to predict this
behavior (Eq. 9) is taken from Armstrong and Frederick [33]. This form is used widely
in the plasticity literature, however to our knowledge, prior to [30] it had not been previ-
ously utilized to predict yielding behavior in complex fluids. One additional benefit of the
Armstrong-Frederick form is the ability of the model to predict transient plastic deformation
below the steady state yield stress. Other more sophisticated EVP models can also predict
this type of transient plastic deformation, however slow creeping flow will not necessarily
result in the correct shape of the predicted Lissajous curves under LAOS (see for example

10
the LAOS predictions of the Saramito model in [20], which differ distinctly from the model
predictions and experimental observations shown in [30] and [31]). Additional extensions to
this framework to incorporate multiple relaxation modes have recently been considered by
Wei et al. [37].

E H B M o d e l K H M o d e l
(a ) 6 0
(b ) 6 0

5 0 5 0

4 0 S tr a in r a te s 4 0 S tr a in r a te s
S tre s s (P a )

S tre s s (P a )
.0 0 5 s -1
.0 0 5 s -1

3 0 .0 1 s -1 3 0 .0 1 s -1
.0 2 s -1 .0 2 s -1
.0 5 s -1 .0 5
2 0 2 0

1 0 1 0

0 0
0 .0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8 0 .9 1 .0 0 .0 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8 0 .9 1 .0

S tr a in S tr a in

(c ) (d )
6 0 6 0

4 0 4 0

2 0 2 0
S tre s s (P a )

S tre s s (P a )

0 0

-2 0 -2 0

-4 0 -4 0

-6 0 -6 0

-0 .4 -0 .3 -0 .2 -0 .1 0 .0 0 .1 0 .2 0 .3 0 .4 -0 .4 -0 .3 -0 .2 -0 .1 0 .0 0 .1 0 .2 0 .3 0 .4

S tr a in S tr a in

Figure 1: Simulated predictions for the 1D versions of the EHB model and the KH model for startup of
steady shear flow ((a) and (b)), and large amplitude oscillatory shear ((c) and (d)). Steady shear startup
simulations are carried out at a range of shear rates, while LAOS simulations are carried out at γ0 = 0.1
(inner curves) and γ0 = 0.4 (outer curves). Values of the model parameters used in these simulations are
similar to those used in [30] to model a Carbopol microgel under LAOS: G = 300 Pa, σy = C/q = 45 Pa,
C = 540 Pa, q = 12, k = 23 Pa.sm , m = 0.43.

The changes that arise due to implementation of the KH mechanism in the generic be-
havior predicted by EVP models for transient rheological flows is shown in Fig. 1. Model
predictions are given for startup of steady shear flow, in addition to large amplitude oscil-
latory shear. As can be seen in the comparison of Fig. 1 (c) and (d), the introduction of
kinematic hardening results in a more gradual transition of the material behavior from a
linear elastic solid below yield, to viscoplastic flow above yield. The KH mechanism also
11
predicts energy dissipation and a continuous frequency and strain dependent evolution of
the loss modulus G00 (ω, γ0 ) as the material deformation approaches the yield point. As
discussed by Fraggedakis et al. [38] LAOS can be used as a key test to elucidate this dis-
sipative behavior. In Fig. 1 (d) this is manifested by the nonzero area enclosed within the
Lissajous curves plotted for γ0 = 0.1. This is contrary to the predictions of the EHB model,
where behavior at these strain levels is purely elastic. The comparison of Fig. 1 (a) and (b)
(startup of steady shear) further illustrate how implementation of the KH mechanism results
in a more gradual (and physically more realistic [30, 39]) transition from an un-yielded to
a yielded state in the material. Capturing the correct material response across a range of
strains under reversing flows is a key ingredient to correctly predicting the kinematics in
complex flows (e.g. sedimentation of a sphere) that involve EVP materials [40, 38].

Summary. The KH framework provides considerable versatility, and offers numerous im-
provements over the simpler Herschel-Bulkley model. Among these are the ability of the
KH model to predict the slow viscoplastic creeping flow (with a power law growth in appar-
ent viscosity) for stresses below the yield stress, σ < σy . Real EVP materials are typically
observed to exhibit this power-law creeping behavior [23, 24, 30]. The KH model can also
accurately predict the response of EVP materials under large amplitude oscillatory shear
(LAOS) [30]. The simpler elastic-Bingham model cannot capture these rheological features
of EVP materials.

2.1.1. Fitting of KH model to experimental data for a real yield stress fluid
Rheological measurements were carried out on a Carbopol microgel (a minimally thixotropic,
“ideal” yield stress fluid [23]) to fit the KH model parameters, as well as to demonstrate the
model’s versatility in predicting rheological behavior. The two tests that are selected are
(1) measurement of the material’s steady shearing flowcurve, and (2) measurement of the
viscoelastic moduli G0 and G00 as a function of strain amplitude γ0 .
The particular Carbopol microgel that is used is a 0.5% wt. 901 variant (manufactured by
Lubrizol). Measurements are made with a TA instruments AR-G2 stress-controlled rheome-
ter, with a roughened cone-plate geometry fixture in order to suppress wall slip. The mea-
12
surement of the steady shearing flowcurve is carried out by imposing a constant shear rate on
the material γ̇0 , and waiting for 3 minutes for the stress to equilibrate. After equilibration,
the imposed rate is then increased. For the oscillatory shear measurements of the storage
and loss moduli G0 (ω, γ0 ) and G00 (ω, γ0 ), tests are carried out at a constant frequency of
ω = 2 rad/s and the strain amplitude was steadily increased.
The experimental data, together with a best fit for each data set to the KH model, are
given in Fig. 2. The best fit curves are given by solid lines in each of the graphs. As we
have shown in Eq 10, The KH model predicts a steady flowcurve identical to the Herschel-
Bulkley relation, which has been shown by a number of previous workers to agree well
with the measured flowcurves for Carbopol microgels [41]. The measurements of G0 (γ0 ) and
G00 (γ0 ) in Fig. 2 (b) show a general trend that is ubiquitous among yield stress fluids. For
small values of γ0 , the material exhibits a linear viscoelastic behavior that is dominated by
elasticity, with G0  G00 . As the imposed strain amplitude is steadily increased towards the
point of yielding, the material response will no longer be linearly viscoelastic. This coincides
with a sharp monotonic decrease in the elastic modulus G0 as well as an increase in the loss
modulus G00 . These changes are indicative of a yielding transition in the material, where
the behavior changes from solid-like to liquid-like. The loss modulus G00 eventually reaches
a maximum, roughly at the point where G0 and G00 cross over. The two moduli both then
decrease monotonically with γ0 and G00 continues to exceed G0 as the strain amplitude is
continuously increased.
The simulated best fit response of the KH model, given by the solid lines in Fig. 2
(b), correctly predicts the changes in G0 (γ0 ) and G00 (γ0 ) of the Carbopol microgel as the
strain amplitude is increased. A similar level of “goodness of fit” was also demonstrated
by Fraggedakis et al. [40, 38] using the mechanism of kinematic hardening. As has been
reported previously [42], the local maximum in G00 at intermediate strains coincides with the
crossover point of G0 and G00 , which the KH model is also able to capture. Furthermore, the
KH model predicts a non-zero value of the loss modulus G00 for small but non-zero strain
amplitudes, and this value is significantly less than G0 (with tan δ ∼ 0.2). Finite values of G00
are present at strain amplitudes on the order of 1%, even though the loss modulus G00 can
13
 (a) 1000 (b) 1000

Carbopol Data
KH model - fit to flowcurve data
KH model - fit to SAOS data
100
]aP[ ssertS

]aP[ ''G ,'G

100 10

1 G'
G''
KH model - fit to SAOS data, q=4
KH model - fit to flowcurve data,q =12
10
0.1
0.001 0.01 0.1 1 10 100 1000
0.01 0.1 1 10 100

-1
Shear rate [s ] Strain amplitude

Figure 2: Fit of the KH model discussed in Sec. 2.1 to rheological data for a Carbopol microgel. The
flowcurve (i.e. a plot of |σ| vs. |γ̇|) is given in (a), and in (b) the dependency of the storage and loss moduli
G0 and G00 on strain amplitude γ0 is plotted. Solid lines are best fits, while dashed lines are fits using model
parameters determined from the other data set.

be shown to rigorously approach zero if the abscissa was extended to much smaller values
of γ0 . This is in contrast to the elastic-Bingham model, which predicts a constant value of
G0 = G, and an identically zero value of the loss modulus for all strain amplitudes below
the yielding point. The only disagreement between the KH model and the experimental
data occurs at low strain amplitudes, where the KH model overpredicts the experimentally
measured values of G00 . However, it still predicts values of G00 which are a factor of four to
five lower than G0 , indicative of a material response that is primarily elastic at small strains.
In addition to the best fit provided by the solid line in Figs. 2 (a) and (b), we also
illustrate sensitivity of the fit to slight variations in fitting coefficients, as well as sensitivity
to experimental variability between data sets, by showing additional fits to the experimental
data using dashed lines. For each data set (flowcurve or oscillatory shear data), the dashed
line corresponds to parameters obtained by fitting the model to the other data set. The
values of the fitting parameters are thus: flowcurve, solid line: C/q = σy = 43 Pa, k = 19
Pa.sm , m = 0.43; flowcurve, dashed line: C/q = σy = 43 Pa, k = 23 Pa.s m
, m = 0.43;
oscillatory shear, solid line: C/q = σy = 43 Pa, k = 23 Pa.sm , m = 0.43, G = 200 Pa, q = 4;
oscillatory shear, dashed line: C/q = σy = 43 Pa, k = 19 Pa.sm , m = 0.43, G = 200 Pa,
q = 12. In general, we observe that the lower value of k used to fit the flowcurve data, results
in a slight underprediction of the values of G00 and G0 at large strains. This slight discrepancy

14
in this fitting coefficient is likely a result of experimental variability. The increase in q from
a value of 4 to 12 (which appears to fit the Lissajous curves for this system best [30], and is
hence used in Fig. 1) does not have any impact to the oscillatory shear data at large strains.
However the increase from q = 4 to q = 12 does result in some changes to G0 at moderate
strains and to G00 at moderate to low strains. Despite these small differences, we see that
in general, both sets of fitting coefficients capture the key trends observed in the Carbopol
microgel

2.2. 3D version

The one-dimensional version of the KH model is useful for predicting material behavior
under simple shearing conditions. For more complex flow scenarios, a more general constitu-
tive framework is required. This framework must provide a constitutive law that is tensorial
in nature, frame-invariant, and thermodynamically admissible. The relationship between
the 3D version and 1D version of the KH model will therefore be analogous to that between
the Upper Convected Maxwell (UCM) model and the simple scalar form of the Maxwell
model (or that between the Oldroyd-B model and the Jeffreys model) [6].
A 3D version of the KH model has already been discussed in the appendix of [30], but
we reproduce the formulation here because it forms the basis of the results which will follow.
Throughout this section we will adopt the tensor notation of Gurtin et al. [5]. Because the
KH model used here is a simplified version of the general elastoplastic formulations utilized
by Anand et al. [43, 44, 45], it is fully frame invariant and thermodynamically admissible.
For simplicity, we will omit some of the details presented in these earlier works that pertain
to the issues of frame invariance and thermodynamic admissibility, and only discuss the key
features of the KH model.

Kinematics. The 3D version of the KH model begins with the definition of the deformation
gradient F. The components of the deformation gradient are given by:

∂xi
Fij = , (18)
∂Xj

15
where (as discussed in Sec. 1) Xj are the cartesian components of the vector X, which
describes the position of material points in a body before a given deformation, and xi are the
cartesian components of the vector x, which describes the position of the material points X
after a given deformation.4 Following the definition of the deformation gradient, the Kroner
decomposition [46] can be used to decompose F into elastic and plastic contributions,

F = Fe Fp . (19)

The elastic part of the deformation gradient is further decomposed into the form

Fe = Re Ue , (20)

where Re is a rotation and Ue is a stretch. The elastic stretch Ue has a spectral represen-
tation
3
X
e
U = λei rei ⊗ rei . (21)
i=1
In the spectral representation of the elastic stretch Ue , λei are the principal values and rei are
the principal directions of the elastic stretch tensor. The operator ⊗ indicates the dyadic
product. Using Ue , it is possible to define an elastic strain tensor which will be of use in
specifying the elastic behavior of the material
1 e e | 1  e2 
Ee ≡ (F (F ) − 1) = U −1 . (22)
2 2
Following [5], we name the above tensor Ee the Green-Saint Venant strain tensor.5
From a kinematical point of view, the time-differentiation of the Kroner decomposition
in Eq. 19 allows for the velocity gradient, L, to be written in terms of an elastic velocity
gradient Le and a plastic velocity gradient Lp ,
−1
L = Le + Fe Lp Fe , (23)

4
The deformation gradient F used here is identical to the “displacement gradient tensor” denoted as E
in [6]. Here we reserve the use of the symbol E to describe the infinitesimal strain tensor defined in Sec. 1.
Furthermore, in [6] the notation r is used in place of x and r0 is used in place of X
5
According to [6], the tensor Ee would be called the “relative finite strain tensor” for the elastic part of
the deformation gradient Fe , and would be denoted as − 12 γ [0] (see Eq. 8.1-9 in [6]). The tensor Fe (Fe )| is
also the Finger tensor of Fe (denoted as B in Eq. 8.1-7 in [6]).
16
where Le and Lp are defined as follows:

−1
Lp ≡ Ḟp Fp , (24)
−1
Le ≡ Ḟe Fe . (25)

The plastic velocity gradient Lp can be split into its symmetric and skew components such
that Lp = Dp + Wp . We call Dp the plastic stretching tensor (it is also sometimes referred
to as the plastic deformation rate tensor) and Wp the plastic spin tensor. One of the
assumptions in this model is that of plastic irrotationality, which takes Wp = 0. This
assumption is general and is justified in [5]. The plastic stretching Dp can also be written
as a product of its magnitude, dp = |Dp |, and its direction Np = Dp /|Dp |.

Dp = dp Np . (26)

Free energy. An important component of the KH model is the form of the free energy Ψ.
We assume that the free energy consists of two separate terms: an elastic free energy Ψe ,
and a defect energy Ψp .

1
Ψ = G|Ee |2 + Λ|trEe |2 + Ψp (A) . (27)
| {z2 | {z }
} defect energy
elastic energy Ψe

In Eq. 27 above, the parameter G is a shear modulus, and the parameter Λ is related to the
bulk modulus K through K = Λ + 2G/3. The form of the elastic free energy results in the
following relationship between the second Piola elastic stress Te and the Green-St. Venant
tensor Ee :
Te = 2GEe + Λ (trEe ) 1 , (28)

where 1 is the identity tensor. The second Piola elastic stress Te is defined (and related to
the Cauchy stress T) as follows [5]:

Te ≡ J(Fe )−1 T(Fe )−| , (29)

where J is the determinant of the deformation gradient F.

17
 To define the defect energy Ψp (A), we must first introduce the concept of the Mandel
stress, which is defined as

Me ≡ J(Fe )| Fe Te = J(Fe )| T(Fe )−| . (30)

The Mandel stress is typically associated with an intermediate structural space in the ma-
terial [5], and its deviatoric part is Me0 = Me − 31 (trMe ) 1. Figure 3 (adapted from [5])
illustrates how the decomposition of the deformation gradient F results in three distinct
“spaces” in which various tensor fields (e.g. stresses, deformation gradients) reside. These
three spaces are the reference space, which is where the undeformed reference body resides,
the observed space, which is where the deformed body resides, and an intermediate struc-
tural space. From the definition in Eq 30, it follows that the Mandel stress is a “structural
tensor field” (according to the nomenclature of [5]) and therefore maps vectors in the struc-
tural space to other vectors in the structural space. It is used to define the defect energy
in what follows, because it is dependent on changes to the material structure. The Mandel
stress is related to, but distinct from the Cauchy stress (or true stress), which resides in the
observed space of the deformed body.
The elastic energy Ψe is used to model purely reversible, elastic deformations within the
material. The defect energy Ψp on the other hand, is introduced to model the structural
rearrangement of the material’s microconstituents. In the three dimensional KH formula-
tion, Ψp depends on a strain-like tensor A, which is analogous to the one-dimensional scalar
internal variable A that was discussed in Sec. 2.1. The structural rearrangement that oc-
curs when the material undergoes plastic flow can result from different underlying physical
processes. In the case of metals, movement of dislocations cause a change in the defect
energy. For foams and dense suspensions, topological rearrangement of bubbles or particles
will cause the defect energy to change. Different elastoplastic materials will have different
dependencies of the defect energy on the strain-like tensor A. However, the basic form of
Ψ given in Eq. 27, in which free energy is additively decomposed into a defect contribution
and an elastic part, is general enough to describe a wide range of EVP materials.
For our purposes, we specify a very simple form of the defect energy Ψp . We first consider
18
 structural space
material structure

Fp (X) Fe (X)

X• F(X) •x
reference body deformed body

infinitesimal neighborhoods
reference space observed space

Figure 3: Diagram adapted from Gurtin et. al. [5] illustrating how the decomposition of the deformation
gradient F results in three distinct spaces.

the spectral representation of the tensor A:

3
X
A= ai li ⊗ li . (31)
i=1

In the above equation, li are the principal directions of A, and ai are the principal values.
The following form is employed for Ψp in the KH model [44, 45]:
1 
Ψp = C (log a1 )2 + (log a2 )2 + (log a3 )2 .

(32)
4
In Eq. 32 above the back stress modulus C has been introduced. Eq 32 above results in an
expression for the back stress, which can be obtained by taking the derivative of Ψp with
respect to A [44],
Mback = C log A . (33)

The tensor A is defined through an evolution equation [45]

Ȧ = Dp A + ADp − qA(log A)γ̄˙ p , (34)

which is a generalized version of the Armstrong-Frederick kinematic hardening rule. Eq. 34

is typically evaluated with an initial condition of A equal to 1, the identity matrix, which
19
corresponds to a “virgin state” in the material [44, 45]. In Eq. 34 we have introduced the
new material constant q (equivalent to the dimensionless q parameter discussed for the 1D
model in Sec. 2.1). The parameter q governs the dynamic evolution of A.
As was the case for the 1D KH model, we can write an effective stress

Meeff = Me0 − Mback , (35)

and constrain the plastic stretching to be codirectional to the effective stress

Meeff
Np = . (36)
|Meeff |
What now remains is to determine the magnitude of the equivalent plastic strain rate,
dp . We define an equivalent plastic strain rate and equivalent shear stress:
√
γ̄˙ p ≡ 2dp Equivalent plastic strain rate (37)
1
σ̄ ≡ √ |Meeff | Equivalent shear stress (38)
2
√
The 2 factors arise in the definitions of the equivalent quantities above so that under simple
shearing flows, these quantities agree with the plastic shear rate γ̇ p and the stress σ used
for the 1D (shear) version of the model introduced in Sec. 2.1. This will be made apparent
in Sec. 2.3.1, where we obtain analytical expressions for the tensorial quantities of the 3D
model under simple shearing flow, and relate them to predictions from the 1D version of the
model.
After defining the equivalent shear stress and plastic strain rate, the power-law rate-
dependent flow rule, which relates σ̄ and γ̄˙ p , is specified:
 σ̄ 1/m
˙γ̄ p = . (39)
k
This above flow rule has eliminated the conditionality that appears in many 3D generaliza-
tions of the Bingham or Herschel-Bulkley models [3, 13, 14]. The discontinuity in the plastic
flow behavior has therefore been “regularized” through the introduction of the evolving
tensor A. Regularization of the yield criterion plays an important role in the numerical sim-
ulations of these types of constitutive laws [27, 25] - here it has been accomplished without
explicitly specifying a finite, large constant viscosity below the yield stress.
20
2.3. 3D KH Model under steady flows

A viscometric flow commonly employed by rheologists is a steady simple shearing defor-

mation, where the velocity gradient L is given as follows:
 
0 γ̇ 0
 
L = 0 (40)
 
0 0
 
0 0 0

Rheologists will typically use such flows to determine the shear and normal stress components
of the Cauchy stress tensor T for a complex fluid or soft solid, and then compare these
measurements to predictions from constitutive laws. The 1-dimensional version of the KH
model described and simulated in Sec. 2.1 only accounts for a single scalar shear stress.
With the 3-dimensional form of the KH model outlined in the previous section, it is now
possible to predict the full 3-dimensional form of the Cauchy stress tensor T for a number
of different rheological flows including steady simple shear flow.
A rheologist will usually measure components of the Cauchy stress T, in terms of the
imposed stretching (or rate of deformation), D. It is therefore necessary to obtain an
analytical relationship of the form T = f (D) rather than of the form T = f (Dp ). To
assist in obtaining such an expression, an assumption of small elastic deformations can be
utilized. This assumption holds for a wide range of materials, and is valid when the yield
stress σy = C/q is much smaller than the elastic shear modulus G, i.e. the yield strain in
shear is γy ≡ C/(qG)  1. For a real EVP material, one can experimentally verify this
0 0
criterion of small elastic strains by checking that the ratio JM /σy  1. JM is a nonlinear
LAOStress measure, called the tangent compliance, and is defined as

0 dγ
JM ≡ , (41)
dσ σ=0

0
for an oscillatory shearing deformation where σ = σ0 cos ωt [30]. JM is generally represen-
0
tative of the elastic modulus of EVP materials, so if the constraint JM /σy  1 is valid,
then considerable plastic flow will be occurring at stresses which only cause small elastic
deformations. From such an assumption, it follows that the elastic stretch tensor is very
21
close to the identity tensor, i.e. Ue ' 1, and therefore from Eq. 20 Fe ' Re . The elastic
part of the deformation gradient Fe is now simply a rotation, for which its inverse is its
transpose, (Re )−1 = (Re )|
Following this assumption, one can then obtain a simplified relation between the stretch-
ing tensor D = 12 (L + L| ) and the plastic stretching tensor Dp . The derivation is detailed
in Appendix B.1, but we summarize it here:

−1
D = Re Dp Re . (42)

The above equation can be used to obtain the following expression providing the rate of
deformation tensor D in terms of the deviatoric Cauchy stress T0 ≡ T − 13 (trT) 1 and a
new internal parameter Ā (the derivation for this step is also provided in Appendix B.1).
 1/m !
1 |T0 − C log Ā| T0 − C log Ā
D= √ √ . (43)
2 2k |T0 − C log Ā|
−1
The tensor Ā ≡ Re ARe in the equation above is an evolving internal parameter which
lies in the space of the deformed body (illustrated in Fig. 3). A, on the other hand, lies in
the structural space of the body. The evolution equation for Ā is now different from that
of A.
◦ √
Ā = DĀ + ĀD − q 2Ā log Ā|D| . (44)
◦
˙ + ĀW − WĀ is the corotational derivative of Ā [5] (and W = Ṙe Re−1 is
Where Ā = Ā
the spin tensor). Alternatively Eq. 44 can be expressed in terms of the upper convected

˙ − ĀL| − LĀ 6 :
derivative of Ā which is Ā = Ā
 √
Ā = −q 2Ā log Ā|D| (45)

6
According to [6] the notation that would be used for the upper-convected (or contravariant convected)
derivative would be A(1) , while the notation used for the corotational (or Jaumann) derivative would be
D
Dt A. Note that the definition of the velocity gradient tensor L used here (in accordance with [5]) is the
same tensor as that denoted by (∇v)† in [6]. See pg. 81 in [5] and pg. 296 in [6] for the definitions of the
two tensors.

22
This evolution equation now involves a frame-invariant, corotational (or alternatively upper
convected) derivative which accounts for the fact that Ā lies in the deformed space of the
body.
The two simple expressions, Eq. 45 and 43, analytically describe how the Cauchy stress
T evolves in the material given a particular imposed deformation rate D. These types of
expressions can be used to understand the response of the material in an Eulerian reference
frame, which is the basis for most rheological experiments. The only simplification that was
made to arrive at these two equations was that of small elastic strains, i.e. C/(qG)  1 (a
reasonable assumption for many yielding materials), resulting in Ue ' 1.

2.3.1. Steady shear

With Eqs. 45 and 43 derived, it is possible to obtain analytical expressions for the
components of the Cauchy stress tensor T under steady shearing conditions as given in
Eq. 40. For ease of representation, we rewrite Eq. 43 as follows
√ m+1
T0 = 2 k|D|m N + C log Ā, (46)

where N is the direction of stretching, defined as N = D/|D|. To arrive at an analytical

solution for the steady state values of the components of T under steady shear, we set
˙ = 0 in Eq. 45, and then solve for the steady state value of the tensor Ā from the following
Ā
equation:
√
ĀL| + LĀ − q 2Ā log Ā|D| = 0 . (47)

Due to the presence of the Ā log Ā term in this equation, the individual components of Ā
are related through a set of nonlinear coupled equations. The equations can be linearized
and simplified considerably for the case where q  1. The initial conditions for imposition
of steady shear are Ā(t = 0) = 1. For large values of q, the recovery or destruction term
√
−q 2Ā log Ā|D| grows quickly, resulting in Ā being very close to 1 for all times. We can
therefore use the substitution Ā = 1 + B, with |B|  1. The Mercator series for the natural
logarithm of Ā is as follows:
B2 B3
log Ā = log (1 + B) = B − + − ... (48)
2 3
23
Eq. 48 above can then be combined with Eq. 47. This results in the following expression:
√
2D + BL| + LB − q 2|D|B = 0 + O(B2 ) . (49)

Inserting a steady shear velocity gradient into the equation above (from Eq. 40), and keeping
only terms that are linear in B, gives the following relationship between the components of
B:      
0 γ̇ 0 2γ̇B12 γ̇B22 0 B B12 0
     11 
γ̇ 0 0 +  γ̇B22 0 − q γ̇ B12 B22 0 ' 0 . (50)
     
0
     
0 0 0 0 0 0 0 0 0
Solving this linear set of equations for the components of B, gives the following values:
 
2
2/q 1/q 0
 
B '  1/q 0 0 . (51)
 
 
0 0 0

Next, using the fact that for q  1, log Ā ' B, we can obtain the components of the
deviatoric Cauchy stress tensor:
 
2σy /q σy + k γ̇ m 0
 
T0 ' σy + k γ̇ m 0 . (52)
 
0
 
0 0 0

We therefore find that the 1-2 shear component of the deviatoric Cauchy stress is given by
T12 ' σy + k γ̇ m , which is the expected result for a Herschel-Bulkley type material with a
shear yield stress σy = C/q. Note, however, that the expression for T0 in Eq. 52 is not
strictly deviatoric, except in the limit where q → ∞. This is due to the fact that the
approximate solution above does not account for order 1/q 2 terms in the 2-2 entry of log Ā.
This can be amended by taking higher order terms in the expansion of log Ā. The resulting
expressions obtained for the terms in log Ā are rather complex (and not included here).
However, these expressions can be simplified by expansion through a Taylor series, resulting
in terms up to order 1/q in the 1-2 component, and terms up to order 1/q 2 for the 1-1 and

24
2-2 components. This yields the following expression for log Ā:
 
2
1/q 1/q 0
 
log Ā '  1/q −1/q 2 0 . (53)
 
 
0 0 0

From which follows a better approximation for the now truly deviatoric Cauchy stress T0 :
 
σy /q σy + k γ̇ m 0
 
T0 ' σy + k γ̇ m −σy /q 0 . (54)
 
 
0 0 0

This KH model therefore predicts a first normal stress difference given by N1 ' 2C/q 2 =
2σy /q, which is independent of the applied shear rate γ̇. These normal stress differences
arise in the absence of elastic stretches, because we have set C/(qG)  1 to arrive at these
expressions.
To confirm the validity of the expression in Eq. 54, and also investigate the range of q
values over which it holds, we carried out numerical simulations in which Eqs. 45 and 43 were
evolved under startup of a very small steady shear rate. The purpose of the small steady
shear rate was to verify that the Cauchy stress evolves towards a point where T12 ' C/q
and where N1 ' 2C/q 2 (as is given in Eq. 54). The shear rate γ̇ is therefore chosen such
that (C/q)  k γ̇ m . Constitutive parameters identical to those used for the fitting of the
Carbopol microgel in Sec. 2.1 were used (i.e. C/q = σy = 45 Pa, m = 0.43, k = 23
Pa.sm ) with the exception that G was set to infinity. The particular shear rate used was
 1/m
γ̇ = 1 × 10−3 s−1 , or in non-dimensional form, Γ̇ ≡ γ̇ qk
C
= 4.8 × 10−3  1. These
simulations were carried out for a large range of q parameters, and the steady state values
of both the dimensionless normal stress and shear stress (N1 /σy and T12 /σy respectively)
are plotted below as a function of the dimensionless parameter q. As the results in Fig. 4
show, the approximate solution given by Eq. 54 predicts the values of T12 and N1 for values
of q & 4. To improve the agreement at smaller values of q, higher order terms could be
included in each component of T0 .

25
 Shear stress T12/s y
10
Normal stress difference N1/ s y

y
Analytical expression for N 1/

s/
y

Analytical expression for T12 /s y

1
N dna
1
y
s/
T21

0.1

0.1 1 10 100
q

Figure 4: Plot of the simulated values (circles) of T12 /σy and N1 /σy for the 3D version of the KH model for
a range of values for the q parameter. The solid lines show the prediction of these stresses obtained using
the approximate solution in Eq. 54.

The relative magnitude of the normal stress difference observed in an elastoplastic yield
stress fluid thus varies with the magnitude of the parameter q in the KH formulation.
Increasing the q parameter, while keeping the shear yield stress σy constant, causes a decrease
in the magnitude of the first normal stress difference N1 . However, q also affects the dynamics
of the evolution equation of Ā. Specifically, larger values of q result in a quicker increase in
the recovery term in the evolution equation for Ā. As a result, one would expect that the
shear and normal stresses saturate to their steady values more quickly for large values of q.
To illustrate this, we plot below the evolution in N1 /σy and T12 /σy vs. accumulated strain
γ for the simplified 3D KH model (as given in Eq. 43 and 45) simulated under startup of
steady shear at a shear rate of Γ̇ = 4.8 × 10−3 . The same model parameters are used as in
Fig. 4, and curves are plotted for seven different q values logarithmically spaced from 5 to
100. From Fig. 5 it is evident that larger values of q result in smaller strains accumulating
before the stresses saturate. The factor 1/q therefore represents a critical strain required
for the KH model to fully yield and evolve towards its steady state. Thus, this factor is
effectively the yield strain of the KH framework. Since many yielding materials show yield
at strains much smaller than unity, these derivations for the case when q  1 are the most

26
practically relevant. The small values of the normal stress difference predicted in the limit
q  1 also help rationalize why robust experimental measurements of N1 in yield stress
fluids are difficult to make [47].

(a) (b)
0.4

1.0

0.3
0.8
y

q increasing

y
s/21T

s/1N
0.6
0.2

0.4
q increasing
0.1
0.2

0.0 0.0
0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0

Strain, g Strain, g

Figure 5: Plots of N1 /σy and T12 /σy vs. time for the 3D KH model simulated under startup of a steady
shear rate of γ̇ = 1 × 10−3 s−1 . The same model parameters as in Fig. 4 are utilized, and curves are plotted
for q values logarithmically spaced from q = 5 to q = 100 (q = 5, 8, 14, 37, 61, 100).

In the limiting case where q → ∞, the first normal stress difference approaches zero and
the shear stress T12 responds immediately to the imposition of a steady shear rate. In this
limiting case, the behavior predicted by the KH model is identical to that of a Herschel-
Bulkley (m < 1) or Bingham like (m = 1) model, with a von Mises yielding criterion to
determine if plastic flow will occur. The case for m = 1 is identical to the inelastic version
of the model introduced by Oldroyd [3], while the case of m < 1 and q → ∞ is identical to
the model used by Ovarlez et al. [13]).
The kinematic-hardening behavior therefore regularizes the flow/no flow condition in
the Herschel-Bulkley model through the introduction of transient viscoplastic flow which
can occur at all levels of stress. This is in contrast to regularization schemes which involve
specifying Newtonian viscous flow with a very large viscosity for stresses below the yield
stress σy [27]. This is an unrealistic assumption, because for many materials the apparent
viscosity η + (t, σ0 ) of the fluid varies in time when a stress is applied below σy , i.e. the
material creeps before arresting. The limiting zero shear-rate viscosity of such materials is
therefore not constant in time (or with increasing strain). The kinematic-hardening behavior
27
can also allow for partial elastic recoil to occur when the stress is stepped to zero after the
material has undergone a creep test. This type of behavior cannot be accounted for in the
3 dimensional Herschel Bulkley type models used by Ovarlez et al. [13] or Martinie et al.
[14].

2.3.2. Shear free flows

Uniaxial elongational flow. For elongational flow [6], the velocity gradient L takes the fol-
lowing form:  
˙ 0 0
 
L = 0 −/2 0  . (55)
 
˙
 
0 0 −/2
˙
Following the same procedure outlined in Sec. 2.3.1 (i.e. neglecting order B2 terms and
higher), a linear relationship between the components of B can be obtained:
    √ 
2˙ 0 0 2B ˙ 0 0 3q B
˙ 11 0 0
   11   √ 
 0 −˙ 0  +  0 −B22 ˙ 0 −  ' 0 . (56)
     
0 3q B
˙ 22 0
     √ 
0 0 −˙ 0 0 −B33 ˙ 0 0 3q B
˙ 33

These components are solved and inserted into Eq. 46 (recall that log Ā ' B) to obtain an
expression for the deviatoric Cauchy stress, T0 .
 √ 
2C
2( 3)m−1 k ˙m + √3q−2 0 0
 √ m−1 m 
T0 '  −( 3) k ˙ − C
 . (57)
 
0 √
1+ 3q
0
 √ m−1 m C

0 0 −( 3) k ˙ − √
1+ 3q

As was also the case in Eq. 52, the expression for T0 above is only deviatoric in the limit
of large values of q. The diagonal terms in the equation for T0 above have power law terms
proportional to ˙m , in addition to constant “yield terms” which contain the coefficients C
and q. Using the expressions for the 1-1 and 2-2 components of T0 , we write an approximate
analytical expression for the extensional viscosity:
√
T11 − T22 √ m+1 m−1 3 3σy
ηE ≡ ' 3 k ˙ + √ , (58)
˙ ˙ − 2/q 2 − 3/q)
(3
28
where we have substituted σy = C/q to result in the order 1/q 2 terms in the denominator.
Considering the limit where ˙ → 0, and q  1, then the T11 stress component approaches
√ √
a value of 2C/ 3q, while the T22 and T33 stress components approach values of −C/ 3q.
Under these conditions, the material will have just yielded. We insert the components of
the stress tensor in this case into the von Mises yield criterion in order to determine an
equivalent stress required for yielding
r
(T11 − T22 )2 + (T22 − T33 )2 + (T11 − T33 )2 + 6(T12 2 + T13 2 + T23 2 )
σe = , (59)
2
√ √
and obtain a value of σe = 3C/q = 3σy . This agrees with the equivalent stress σe
determined for the simple shear case when q  1. The material therefore yields under
both shear and extensional deformations when the equivalent stress reaches the same value.
This verifies that the KH model satisfies the von Mises yield criterion for large values of q.
The von Mises yielding criterion has been experimentally verified for a wide range of EVP
materials, including emulsions, physical gels and colloidal gels, among others [15, 13, 14].
To verify the validity of the approximation in Eq. 57, we plot in Fig. 6 the steady
state, dimensionless normal stress difference (N1 /σy = (T11 − T22 )/σy ) against q for a very
 1/m
C
small dimensionless extensional strain rates (Ė ≡ ˙ qk = 4.8 × 10−3 ) under uniaxial
elongational flow. In Fig. 6 we include the analytical approximation from Eq. 57 (solid line)
as well as the direct numerical results obtained through numerical simulation of Eqs. 45 and
43 (hollow circles). There is good agreement between the analytical approximation and the
√
numerical results for q & 8. The numerical results generally predict a value of N1 ' 3σy
in uniaxial extension for all values of q, so this appears to be a good approximation that
holds even for smaller values of q. The analytical approximation, however, diverges at a
√
value of q = 2/ 3, and then becomes unphysical and negative for small values of q. While
higher order terms of B can be used to obtain a more accurate approximation for N1 (),
˙ the
results in Fig. 6 and the form of Eq. 58 suggest that the following expression for uniaxial
extensional flow of the KH model holds over a large range of strain rates and q values:
√ m+1 m √
N1 () ˙ ˙ '
˙ = ηE () 3 k ˙ + 3σy . (60)
29
 10

y
s/
N1
1

0.1
Normal stress difference N1/ s y

Analytical expression for N 1/ s y

Ã3

0.1 1 10 100
q

Figure 6: Plot of the simulated values (circles) of N1 /σy for the 3D version of the KH model for a range of
values for the q parameter, under elongational flow at low deformation rates Ė  1 . The solid lines show
the prediction of these stresses obtained using the approximate solution in Eq. 57.

Planar elongational flow. For planar elongational flow [6] the velocity gradient L takes the
following form:
 
˙ 0 0
 
L = 0 −˙ 0 . (61)
 
 
0 0 0
Combining the above equation with Eq. 49 gives the following relationship between the
components of the tensor B
     
2˙ 0 0 2B11 ˙ 0 0 2q B
˙ 11 0 0
     
 0 −2˙ 0 +  0 −2B22 ˙ 0 − ˙ 22 0 ' 0 . (62)
     
  0 2q B
     
0 0 0 0 0 0 0 0 0

These can be solved for, and then inserted into Eq. 54 in order to obtain the components of
the deviatoric Cauchy stress T0 :
 
C
˙ m+
k(2) q−1
0 0
 
T0 '  ˙ m−
−k(2) C (63)
 
0 q+1
0
 
0 0 0
30
The above result allows us to write the following approximation for the planar elongational
viscosity (analogous to Eq. 58 above for the uniaxial elongational case)

T11 − T22 2σy

ηP ≡ ' 2m+1 k ˙m−1 + . (64)
˙ ˙ − 1/q 2 )
(1

For shear flow, elongational flow and planar elongational flow, the results giving the Cauchy
stress tensor reduce to that of a Newtonian fluid with shear viscosity µ = k for the limits
of m = 1, and C/q = σy = 0. As a result, the Trouton ratio for elongational flow will also
approach ηE /k = 3 in this limit, and for planar elongational flow the Trouton ratio will
approach ηE /k = 4.

2.3.3. Accounting for elastic deformations - large and small

Small elastic deformations. The simplifications so far of the equations in Sec. 2.2 have
resulted in only the plastic part of the deformation gradient, Fp , contributing to the total
stretching deformation D of the material (this has resulted in the elastic modulus G not
playing a role in the analytical results). It is possible to analytically account for small
contributions to the deformation D from the elastic part of the deformation gradient, Fe .
In Eq. B.3, the assumption U̇e = 0 was made. However if we account for a nonzero value of
U̇e , then the elastic part of the deformation gradient Le can be rewritten:

−1 −1 −1
Le = Ṙ e e
| {zR } +R e e e e
| U̇ U{z R } . (65)
We , a spin De , a stretching

−1 −1
Referring to the term Re U̇e Ue Re as a stretching term only applies for small elastic
−1
deformations, when the skew part of U̇e Ue is close to zero (see pg. 90 of [5]). The
elastic part of the deformation gradient now also has a stretching De in addition to the spin
We (which was the only part of Le accounted for in our assumptions thus far). Our total
deformation Dt can now be written as a sum of two stretching deformations, D and De

Dt = D + De , (66)

−1
where D = Re Dp Re , as given previously in Sec. 2.3, and D follows the same behavior
as given earlier in Eqs. 45 and 43. Small elastic deformations can now be accounted for
31
by adding an additional elastic stretching De onto the plastic part of the stretching, D =
−1
Re Dp Re . The additional elastic stretching De is related to a time derivative of the Cauchy
stress T. We take the time derivative of Eq. 28 to obtain the following:

˙ e )1 .
Ṫe = 2GĖe + Λ(trE (67)
◦ −1
By noting that T = Re Ṫe Re (this again only holds for small elastic stretches), and also
−1
using the identity Ėe = Fe| De Fe ' Re De Re (see pg. 90 of [5]), we arrive at the following
relation for the Cauchy stress:
◦
T = 2GDe + Λ (trDe ) . (68)

The relation above can also be inverted to give the elastic stretching De in terms of the
◦
corotational derivative of the Cauchy stress T. The equation above allows one to account
for the additional presence of small elastic stretches in the total deformation of the material.

Large elastic deformations. Obtaining an analytical expression for cases of large elastic
−1
stretches (i.e. Ue  1) is difficult, because the approximate relation D = Re Dp Re
no longer holds. As a result, converting the tensorial form of the equations in 2.2 to an
Eulerian reference frame involving D (or Dt ) is, to the authors’ knowledge, not possible.
However, numerical simulations on the set of equations given in Sec. 2.2 can be carried out.
These are used to examine what the full 3D form of the KH model predicts when elastic
stretches are large.
Startup of steady shear flow simulations were carried out using Matlab with the form of
the equations given in Sec. 2.2 at a number of different imposed shear rates. The evolution
on the components of the stress tensor T were monitored and the simulations were stopped
when the tensor components approached steady state values. Plots of the dimensionless
shear stress T12 /σy and the dimensionless first normal stress difference N1 /σy against applied
 1/m
C
dimensionless shear rate Γ̇ ≡ γ̇ qk are given below in Fig. 7. The particular fitting
coefficients used for this 3D simulation of the full form of the KH model are consistent with
those used for a previous study of Carbopol microgels [30]. Specifically: C/q = σy = 45 Pa,

32
m = 0.43, k = 23 Pa.sm , q = 12, and G = 350 Pa. The full 3D form of the KH model in
Sec. 2.2 also requires specifying the parameter Λ. We set Λ = 5000 Pa, so that C/(qΛ) =
0.01  1 and the material’s elastic behavior approaches that of an incompressible solid .
Overlaid on the values corresponding to this simulated flowcurve in Fig. 7 are the analytical
prediction for the shear stress from Eq. 54 (which assumed q  1 and C/(qG)  1).
s/y
1

1
N dna
s/y

N1/ s y from 3D KH model

21

0.1 N1 / s y = 2/q
T

e p
(N1 +N1 )/ s y

T12/ s y from 3D KH model

T12 /s y analytical predition

0.1 1 10 100

Figure 7: Prediction of the steady flowcurve for the 3D form of the KH model given in Sec. 2.2. Circles give
the predicted dimensionless shear stress and triangles indicate the computed first normal stress difference.
The solid lines are predictions using the analytical approximations in Eq. 54.

The analytical expression for the shear stress T12 is the same as the 1-D Herschel-Bulkley
equation, and it agrees with the predictions of the simulation. This shows that the 3D form of
the KH model is a more general form of the 1-D version discussed in Sec. 2.1. It also implies
an equivalency between the parameters fitted using the 1-D version of the model, and the
parameters that must be provided to the 3D formulation. This makes fitting experimental
flowcurve data (such as that in Fig. 2 (a)) to the 3D form of the KH model very simple.
At the lowest shear rates Γ̇, one can make an a posteriori estimate of the order of
magnitude of elastic stretches (γ e ). A rough estimate gives γ e ∼ T12 /G ∼ C/(qG) = 0.13,
which is still well below unity. At the highest shear rates Γ̇ ≥ 100, the analytical expression
for T12 slightly over-predicts the values obtained from simulations of the full 3D form of the
KH model. At these highest shear rates, we estimate the elastic stretches to be γ e ∼ 0.58,
33
which are nearing a value of 1. However, there is only a 7.5% disagreement between the
analytical expression for T12 from Eq. 54 and the simulated value of T12 at these high shear
rates.
From Fig. 7, it is clear that the values of N1 are under-predicted by the analytical
rate independent approximation in Eq. 54. The analytical expression in Eq. 54, which was
derived assuming small elastic deformations, offers a lower bound on the value of N1 . When
there are large elastic deformations (C/(qG)  1), a large additional contribution towards
the total first normal stress difference N1 will come from nonlinearities arising in the elastic
stress-strain relation specified in Eq. 28. For shear rates that approach zero, one can estimate
the contribution to N1 from the elastic-stress strain relation (we call this contribution N1e ).
This can be done by noting that for the case of simple shear with magnitude γ in the 1-2
entry in the elastic part of the deformation gradient Fe , the form of the elastic free energy
in Eq. 28 predicts a normal stress difference T11 − T22 = Gγ 2 . Under steady shear where
the material has yielded, the extent of elastic shear strain accumulated γ can be estimated
as σy /G, which is essentially the “yield strain” of the material. This contribution to N1 is
therefore N1e = σy2 /G = C 2 /(q 2 G). This estimate also underpredicts the simulated value of
N1 provided in Fig. 7, suggesting that the actual value of N1 consists of contributions from
both the defect energy Ψp and the elastic free energy Ψe . When these contributions are
added as per the following equation:

2σy σy2
N1 ' N1p + N1e = + , (69)
q G

the prediction approaches the simulated value of N1 at low shear rates. This demonstrates
that the functional form of N1 can be tuned by altering the expression which gives the elastic
free energy, e.g. introducing a linear elastic, neo-Hookean or Mooney-Rivlin solid [7, 5]. This
provides additional flexibility within the framework of the KH model to describe the elastic
response of real elastoviscoplastic (EVP) materials, however, the predictions for N1 (γ̇) must
then be obtained numerically.

34
2.4. Further discussion
2.4.1. Oldroyd-B model vs. KH model
The form of the KH model discussed in Sec. 2.3 is very similar to the Oldroyd-B constitu-
tive model that is often used to describe polymer melts and solutions. The Oldroyd-B model
is equivalent to an upper convected Jeffreys model [6], in which the total stress in the ma-
terial is additively decomposed into a polymeric contribution Tp and a solvent contribution
Ts , as follows:
T = 2ηs D + Gp (Â − 1) . (70)
| {z } | {z }
Ts Tp

With Gp a modulus and ηs a viscosity coefficient. The tensor variable Â, much like the Ā
tensor used by the KH model in Sec. 2.3, is an internal strain-like microstructural variable
which evolves according to the following differential equation:
˙ Gp
 = L + ÂL| − ( − 1) . (71)
ηp
Eq. 71 above can be derived from considering a dilute suspension of linearly elastic dumbbells
[48]. The tensor  = hQQi/Q0 2 represents the second moment tensor of the dumbbell
distribution, and the relaxation time scale for the dumbbells is given by τp = ηp /Gp
Revisiting the form of the KH model discussed in Sec. 2.3, we can decompose the de-
viatoric Cauchy stress for our elastoviscoplastic material in a manner similar to Eq. 70.
However for the KH model, the viscous solvent term Ts becomes replaced by an inelastic or
generalized Newtonian fluid term, while the polymeric contribution to stress (which results
from deformation of the material microstructure) becomes the back stress, Tback :
√ m+1 D
T0 = 2 k|D|m + C log Ā , (72)
|D| | {z }
| {z } Tback
Ts

where the tensor Ā evolves according to the following differential equation:

˙ = LĀ + ĀL| − √2q|D|Ā log Ā .

Ā (73)

Eqs. 72-73 are similar to the Oldroyd-B model if we set m = 1 and consider large values
of q. When q is large, we can approximate both the log Ā terms and the Ā log Ā terms as
35
(Ā − 1). This results in the following expressions for the Cauchy stress, and the evolution
equation for Ā for the KH model in the limit q  1:

2kD + C(Ā − 1) ,
T0 = |{z} (74)
| {z }
Ts Tback

˙ = LĀ + ĀL| − √2q|D|(Ā − 1) .

Ā (75)

Eqs. 74 and 75 now appear identical to the equations for the Oldroyd-B constitutive law,
with the one exception that the relaxation rate 1/τp = Gp /ηp in front of the recovery term in
√
Eq. 71 has been replaced by 2q|D|. This is equivalent to a White-Metzner approximation
in which the polymeric relaxation time τp in the Oldroyd-B model is a function of the
magnitude of the strain rate tensor |D|. Due to the specific functional form chosen for this
√
relaxation time (τKH = 1/( 2q|D|)) the relaxation time diverges in the limit of slow flows.
This simple modification causes a yielding behavior to arise.
This comparison of the Oldroyd-B model and the KH model is best illustrated visually
through a diagram of mechanical analog elements as shown in Figs. 8 (a) and (b). The
mechanical analog element for the KH model is drawn for the limiting case of m = 1, q  1
and C/(qG) → 0. In Fig. 8, the elements for the two models are arranged in a similar
fashion, with two branches resulting in the stress being additively decomposed into two
components. For the case of the KH model, however, the polymer contribution to the stress
Tp is replaced by the back stress, Tback . Table 1 shows the parameter equivalency between
the KH model and the Oldroyd-B model. This equivalency only holds for the KH model
framework when m = 1, q  1 and C/(qG)  1.
In order to account for elastic deformations in the KH model, it is necessary to include
another elastic element with modulus G in series with this Jeffreys type element. When
such an element is added, the mechanical analog then becomes identical to that which is
discussed in Sec. 2.1, where predictions of the linear viscoelastic moduli were given for the
one dimensional version of the model when m = 1. This is due to the fact that at small
deformations, |D| approaches zero, which results in an infinite viscosity for the dissipative
element in the back stress branch.
36
Figure 8: Mechanical analog element for the Oldroyd-B model (a) and the KH model (b). The mechanical
analog element for the KH model is drawn for the limiting cases of m = 1, q  1 and C/(qG)  1.

Table 1: Table showing the parameter equivalency between the KH model framework and the Oldroyd-B
model (here the KH model is simplified for the limiting case where m = 1, q  1 and C/(qG)  1).

3DKH Oldroyd-B

(for m = 1, q  1 and C/(qG) → 0)

Solvent Viscosity k ηs

Modulus C Gp

Polymer Viscosity √ C ηp
2q|D|
ηp
Relaxation Time τKH = √ 1 τp =
2q|D| Gp

Internal Variable Ā Â

Back/Polymer Stress Tback Tp

37
 The use of modified Jeffreys-like constitutive laws to describe yielding behaviors is com-
mon in the rheology literature. The models used by Coussot et al. [49], Quemada [50],
Dullaert and Mewis [51] and Souza Mendez [52] are variants of the Jeffreys model (and
most of these are cast into one dimensional forms). The model introduced by Saramito
[20, 21] is not built up using the Jeffreys model as a starting point, but can still be rep-
resented using simple mechanical analog elements, and with the correct modifications, can
be converted into a representation such as the model discussed here. Indeed, recent work
by Fraggedakis [38] implemented the kinematic hardening mechanism discussed here into
the Saramito model. This implementation improved the agreement of the model predictions
to experimental measurements of G0 and G00 for a carbopol microgel over a wide range of
frequencies and strains.
In summary, the unique aspect of the KH model framework is that with the modification
of one model parameter in the Jeffreys model, ηp , the KH constitutive law can predict a
yield-like behavior, slow viscoplastic creepeing flow, and the correct material behavior as
it transitions from an unyielded to a yielded state under large amplitude oscillatory flows
[30, 31, 38]. This is perhaps simpler and more intuitive than the modifications made in
the other models mentioned here, however these more complex models are also designed
to account for thixotropic effects. Thixotropy is often prevalent in EVP materials, so in
the following section we discuss how to implement thixotropic behavior into the KH model
framework.

2.4.2. Implementing thixotropy

The KH model framework as presently formulated does not account for thixotropic be-
havior. Thixotropic effects are typically manifested as an aging and shear rejuvenation
behavior [53]. Rheological aging behavior is usually associated with Brownian motion and
thermally attracted rearrangement of a material’s microconstituents, even in the absence of
shear. The internal parameter A in the KH model only evolves under the application of a
non-zero shear rate γ̄˙ p , so aging behavior will not be predicted by the model.
There are a number of phenomenological approaches to modify the KH model to account

38
for the thixotropic behavior which is frequently observed in EVP materials [53, 54]. Most of
these methods involve introducing a positive scalar internal variable, λ, which is a dimen-
sionless, scaled measure of the internal state of the material order. This parameter then
evolves with deformation according to a differential equation. This equation must account
for both aging and shear rejuvenation. The KH model parameters can then be specified as
a function of λ. For example, the viscosity k can be related to λ so that it decreases when
the material shear rejuvenates, and increases when the material ages. This is the approach
taken in the works by Coussot et al. [49] Quemada [50], Dullaert and Mewis [51] and Souza
Mendez [52]. Another approach would modify the scalar flow rule in Eq. 39 in order to
account for the presence of a scalar yield stress. This scalar yield stress is dependent on
the level of structure in the material, which again is captured by the parameter λ. The
appropriate flow rule thus becomes:
 1/m
p σ̄ − σy
γ̄˙ = if σ̄ > σy , γ̄˙ p = 0 otherwise (76)
k
σy = σy (λ) (77)

These equations allow the material to isotropically harden and soften. This is in contrast to
the process of kinematic hardening, in which the material may strengthen along the direction
of deformation, i.e. in a non-isotropic manner. Constitutive laws which account for both
isotropic and kinematic hardening allow for both the size and center of the yield surface (in
stress space) to vary, and may thus be referred to as isotropic-kinematic hardening (IKH)
models [31, 32].

2.4.3. The importance of thermodynamic admissibility

Although the KH model framework offers flexibility in implementing thixotropic behav-
ior, we emphasize that there are thermodynamic restrictions to the types of modifications
that are allowable. Many workers have utilized models which specify a dependency of a
modulus (e.g. G) on the structural parameter (e.g. λ) - see for example, Quemada [50],
Dullaert and Mewis [51], Souza Mendez [52] and Mujumdar et. al. [55]. Without careful
treatment, this type of dependency, combined with rheological aging in the material, may
39
violate dissipation inequalities that are derived through the first and second laws of ther-
modynamics. Gurtin et. al. [5] give the following mechanical free energy imbalance for an
isothermal material:
T : D − Ψ̇ = δ ≥ 0 , (78)

where Ψ is the free energy per unit volume in the material, and δ is the overall dissipation
in the material per unit volume, which is always non-negative. Consider a Jeffreys type
viscoelastic model, where the solvent viscosity is set to ηs = 0 and the polymer viscosity
is set to ηp → ∞ , leaving a single elastic element with modulus G(λ). For simple 1-D
deformations, Ψ = G(λ)γ 2 /2, with G an increasing function of λ. The aging process causes
an increase in λ under a zero shear rate, so it is possible for λ̇ (and Ψ̇) to be positive when
D or γ̇ are zero. Therefore, the left hand side of Eq. 78 will be negative during this aging
process, violating the dissipation inequality.
To illustrate how such models give unphysical predictions, we consider the following
sequence of step deformations (which have been discussed previously in [34, 56]) for this
type of “thixoelastic” element with σ = G(λ)γ : First, a step in strain of amplitude γ0
is imposed at t = t1 . At this initial time, the material is not in its fully aged state with
G(λ) = G1 , a low value. The stress will increase instantaneously to a value of σ(t1 ) = G1 γ0 .
The material is then allowed to age (i.e. the parameter λ increases) while the strain is held
constant, resulting in the modulus increasing to a new, larger value G(λ) = G2 . This results
in the stress in the material increasing to preserve the equality σ = G(λ)γ0 . At some time
t2 , the strain is stepped down to zero, which also results in the stress stepping down to zero.
The sequence of steps in strain and stress are shown graphically in the figure below:
The mechanical work done on the material for each step in strain, W1 and W2 can be
evaluated through the following integrals:
Z t+1
G1 γ02
W1 = σ(t)γ̇(t)dt = , (79)
t−
1
2
Z t+2
G2 γ02
W2 = σ(t)γ̇(t)dt = − . (80)
t−
2
2
In this thixoelastic element G2 > G1 , so the material returns more energy than what was
40
 G2g0
g0

ssertS
niartS

G1g0

t1 t2 t1 t2

Time Time

Figure 9: Sequence of steps in strain on a purely “thixoelastic” element with σ = G(λ)γ. The aging results
in an increase in stress while the strain is held constant. This allows the material to increase the energy it
has stored, while the extent of its deformation remains constant.

initially input and stored elastically at t = t1 . This is unphysical, and a more sophisticated
constitutive formulation is required in order to correctly capture this thixoelastic behavior.
Such a model must begin from specifying a form of the free energy in the material Ψ, so as
to not violate forms of the dissipation inequality such as that given by Eq. 78. This form of
the free energy must correctly account for the effect that aging has on the material, and may
necessarily need to be cast as a thermomechanically coupled model due to the underlying
thermal/Brownian mechanisms which are responsible for the aging process. Representative
formulations which have explicitly ensured the correct free energy imbalance include those
used by Anand et al. [43] from the plasticity literature, and Wei et al. [37], which are more
recent additions from the rheology literature.

3. Conclusions

We have proposed a comprehensive framework for modeling elasto-viscoplastic behavior

in complex fluids. This framework is rooted in the approach taken in plasticity, where
deformations are decomposed systematically into reversible (elastic) and irreversible (plastic)
components.
Two different versions of a “kinematic hardening” model have been discussed here. A
simple one-dimensional version, and a three-dimensional, frame invariant, thermodynami-
cally consistent version. The 3D formulation is analogous to how the UCM or Oldroyd-B
models represent general frame-invariant formulations of the linear Maxwell and Jeffreys

41
models respectively [6]. We also demonstrated an equivalency between the model parame-
ters of the 3D frame invariant form of this KH model, and the 1D form.
Predictions of this KH model for several canonical rheological flows (steady shear, uni-
axial and planar extension) were considered, and analytical expressions of the tensor valued
Cauchy stress were derived under certain limiting conditions corresponding to small elastic
deformations. Predictions were also provided for different rheological experiments (steady
flowcurve, creep, SAOS). and the model was fitted to experimental data for a model yield
stress fluid (a Carbopol microgel).
The 3D KH formulation can be simplified to other constitutive laws under certain limits.
Specifically, for the limit σy → 0, the 3D KH model simplifies to a generalized Newtonian
fluid model with either a constant Newtonian viscosity, or power law rheology, depending
on the value of m. In the limit of q → ∞, the 3D KH model simplifies to a 3D Bingham
or Herschel Bulkley like viscoplastic fluid (again depending on the value of m) with a von
Mises yielding criterion which determines whether plastic flow will occur or not. In the limit
q  1, the expression for the yield stress σy = C/q was shown to represent a material with
a (back stress) modulus C and a yield strain γy ' 1/q.
The similarities between the frame-invariant Oldroyd-B model and the 3D KH model
were also illustrated. The KH framework can be likened to a White-Metzner generalization
of the upper-convected Jeffreys model in which the relaxation time has a specific functional
√
form τKH = 1/( 2q|D|). We ended with a discussion of two important current aspects of
modeling TEVP behavior. First, how to implement thixotropy into this framework, and
secondly, how to keep these models thermodynamically admissible as they are modified to
describe the response of real materials.
We conclude by reiterating the many benefits of this constitutive framework. First, the
framework discussed here is fully 3D, frame invariant, and thermodynamically admissible.
This is useful for conducting simulations of soft materials under complex flow scenarios.
Second, the framework is flexible because it can allow the introduction of more complex
behavior, such as thixotropy, different types of kinematic hardening, and variants of vis-
coelastic behavior below yield to predict normal stresses and other aspects of nonlinear
42
elasticity.
Compared to the commonly used generalization of the Bingham or Herschel-Bulkley
model [3] the KH model offers several improvements. First, the flow/no flow criterion is
regularized using the evolution equation for A. This lets the material naturally evolve to-
wards a yielded or unyielded state, depending on the level of the stress that develops in
the material under an imposed deformation. Consequently, there is no need to track yield
surfaces (in physical space) when simulating flow transients of the 3D KH model in more
complex geometries. Secondly, the KH model is able to predict transient creeping flow for
stresses below the yield stress σy . These types of transients are commonly observed in EVP
materials, and a single large zero shear rate viscosity parameter is not sufficient to charac-
terize this behavior. Third, tuning of the elastic material response in the 3D KH model can
be used to capture the range of normal stress difference responses observed in real in EVP
materials. Fourth, and finally, the KH model is generally able to capture material behavior
over a much larger range of rheological experiments (e.g. LAOS, creep behavior). All of this
is accomplished with the introduction of one additional fitting parameter, the coefficient q,
and an evolution equation for the internal variable A.

Acknowledgments

The authors would like to acknowledge L. Anand and K. Kamrin for help with developing
the constitutive model. They would also like to thank Dr. Will Hartt and Lori Bacca of
Procter and Gamble for providing the Carbopol samples. C.J.D. also acknowledges A. M.
Osorno for fruitful discussions, and Chevron Energy Technology Corporation for providing
funding.

Appendix A. Predictions of 1D KH model

Appendix A.1. Derivation of creep behavior of KH model

An expression for the asymptotic, long time dependency of strain on time for the KH
model undergoing creep can be obtained by assuming the following time-varying form for
43
the internal structure parameter A

σ0 b
A' − x , (A.1)
C t

where b is an unknown prefactor, and x is an unknown exponent. Taking the derivative of

Eq. A.1 results in an expression for Ȧ

bx
Ȧ ' . (A.2)
tx+1

Substituting Eq. A.1 into Eqs. 6 and 9, and then combining these two expressions yields
another expression for Ȧ
 1/m 
Cb qσ0 
Ȧ ' 1− . (A.3)
ktx C
When inserting Eq. A.1 into Eqn. 9, we approximate γ̇ p (1 − qA(t)) as γ̇ p (1 − qσ0 /C). Com-
paring the exponents and prefactors in Eq. A.3 with those of Eq. A.2 gives the following
expressions for x and b

m
x= , (A.4)
1−m
   m   1
qσ0  1 − m m−1 C m−1
b= 1− . (A.5)
C m k

The expressions above can be inserted into either Eq. A.3 or A.2, which can in turn be
inserted into Eq. 6. This results in an expression for the rate of plastic strain γ̇ p . This can
be used to determine the apparent viscosity η + ≡ σ0 /γ̇ p , which is as follows
 1/(1−m)
+ t
η ' ηc , (A.6)
tc

where tc is the characteristic time scale in Eq. 12 and ηc is the characteristic viscosity scale
in Eq. 13.

Appendix A.2. Derivation of SAOS behavior of KH model (m = 1)

For SAOS, an oscillatory shear stress σ = σ0 cos ωt is imposed with σ0  σy = C/q.

The magnitude of the terms on the right hand side of Eq. 9 can be compared, and since

44
qA  1 the second term can be ignored. As a result, A can be approximated by the following
differential equation:
Ȧ ' γ̇ p . (A.7)

For small oscillatory stresses, integration from a fully relaxed equilibrium gives A ' γ p .
Inserting this (and the expression σ = σ0 cos ωt) into Eq. 6 (while taking account of the
directional integer np ) then gives the following first order linear ODE:

C p σ0
γ̇ p + γ = cos ωt . (A.8)
k k

The above equation can be solved using an integrating factor, resulting in the following
expression for γ(t):
   
1 C kω
γ = σ0 + 2 cos ωt + σ0 sin ωt . (A.9)
G C + k2ω2 C + k2ω2
2

In Eq. A.9, the total (measurable) strain in the material γ = γ e + γ p is given, so a term
with the elastic modulus G enters directly into the resulting expression for the strain (and
this term is in phase with the driving sinusoidal stress).
The coefficients in front of the cosine and sine terms in the parentheses in Eq. A.9 above
are the linear viscoelastic compliances J 0 (ω) and J 00 (ω) respectively. These can be converted
to the linear viscoelastic moduli G0 (ω) and G00 (ω) [57], by using the following expressions
which relate J 0 (ω) to G0 (ω) and J 00 (ω) to G00 (ω)

1/G0 (ω)
J 0 (ω) = (A.10)
1 + tan 2 δ
1/G00 (ω)
J 00 (ω) = , (A.11)
1 + (tan 2 δ)−1

where δ is the phase difference between the stress and strain. When the equations above are
combined with the expressions for J 0 (ω) and J 00 (ω) from Eq. A.9, we obtain the following
expressions:

GC 2 + Gk 2 ω 2 + G2 C (1 + G/C) + (kω/C)2
G0 (ω) = = G , (A.12)
G2 + C 2 + 2GC + k 2 ω 2 (1 + G/C)2 + (kω/C)2
G2 kω (G/C)(kω/C)
G00 (ω) = 2 2 2 2
=G . (A.13)
C + k ω + 2GC + G (1 + G/C)2 + (kω/C)2
45
Appendix B. Derivation of the 3D frame-invariant form of KH model

Appendix B.1. Relating the stretching tensor to the plastic stretching tensor

One can obtain a simplified relation between the stretching tensor D = 12 (L+L| ) and the
plastic stretching tensor Dp associated with irreversible plastic deformations in the material.
To obtain such an expression, the assumption of small elastic deformations must hold. From
such an assumption, it follows that the elastic stretch tensor is very close to the identity
tensor, i.e. Ue ' 1, and therefore from the right polar decomposition of the elastic part of
the deformation gradient, Fe ' Re .
The first step towards obtaining this relation is to write the overall observed velocity
gradient L in terms of the plastic velocity gradient Lp and the elastic velocity gradient Le
[5]
−1
L = Le + Fe Lp Fe . (B.1)

We then show that the stretching tensor D is equal to the second term on the right hand
side in Eq. B.1. By combining Eq. B.1 with the expression D = 21 (L + L| ), the following is
obtained:
1 1 −1 −1
D = (L + L| ) = ((Le + Fe Lp Fe ) + (Le + Fe Lp Fe )| ) . (B.2)
2 2
For Eq. B.2, we first show that Le + Le| = 0. We note that:

−1 −1 −1 −1
Le = Ḟe Fe = (Ṙe Ue + Re U̇e )Ue Re = Ṙe Re . (B.3)

This follows from Ue ' 1 for all times, so the rate of change of elastic stretch U̇e is
zero. Eq. B.3 implies that Le is skew symmetric [5], so its transpose is its negative. Thus
Le + Le| = 0, so two of the terms in Eq. B.2 cancel out. We now show that the remaining
two terms simplify to the second term in Eq. B.1.

1 −1 −1 −1
D = ((Fe Lp Fe ) + (Fe Lp Fe )| ) = Re Lp Re , (B.4)
2

where the last equality follows from taking the transpose of the second term, noting that
Fe ' Re , and also noting that Lp = Dp (i.e. there is no plastic spin and Lp is symmetric

46
- this assumption is justified in [5]). We are left with the following simple equation, which
relates D to Dp :
−1
D = Re Dp Re . (B.5)

The expression in Eq. B.5 can be used to obtain a functional relationship of the form
T = f (D) in the observed space of the material. We combine Eq. 36 and Eq. 39 to obtain
a tensorial expression for D in terms of the deviatoric Mandel stress Me0 and the parameter
A:
1/m !
|Me0 − C log A| Me0 − C log A

e−1 1 −1
D = Re (dp Np )R = Re √ √ Re . (B.6)
2 2k |Me0 − C log A|
The expression above provides D in terms of the deviatoric Mandel stress Me0 . Recall from
the discussion around Fig. 3 that the Mandel stress is associated with the intermediate
structural space of the material. We desire to modify Eq. B.6 above so that it is written in
terms of the deviatoric Cauchy stress T0 , which is the true or observed stress in the material
that would be measured in a rheological experiment. To do this, we note that:
−1
Me0 = JRe T0 Re , (B.7)

where J is the determinant of the deformation gradient F. Eq. B.7 above follows from
Eqs. 29 and 30, and the fact that Me is an isotropic function of Ue - see pg. 563 of [5] for a
guided proof of this. We further restrict our material to be incompressible, so that J = 1.
We define a new tensorial variable, Ā, which is an evolving internal tensor field which lies in
the space of the deformed body (illustrated in Fig. 3). This tensor Ā is related to A (which
lies in the structural space of the body) as follows:
−1
A = Re ĀRe . (B.8)

The above equation, together with Eq. B.7, can be substituted into Eq. B.6, in order to cancel
−1
out the Re and Re tensors outside the parentheses on the right hand side of Eq. B.6. These
steps result in the following expression, which provides the rate of deformation tensor D in
terms of the deviatoric Cauchy stress T0 and this new internal tensorial parameter Ā.
 1/m !
1 |T0 − C log Ā| T0 − C log Ā
D= √ √ . (B.9)
2 2k |T0 − C log Ā|
47
References

[1] D. Bonn, M. M. Denn, L. Berthier, T. Divoux, S. Manneville, Yield stress materials in soft condensed
matter, Rev. Mod. Phys. 89 (2017) 035005.
[2] H. A. Barnes, The yield stress–a review of παντ α ρει–everything flows?, Journal of Non-Newtonian
Fluid Mechanics 81 (1999) 133 – 178.
[3] J. G. Oldroyd, A rational formulation of the equations of plastic flow for a bingham solid, Mathematical
Proceedings of the Cambridge Philosophical Society 43 (1947) 100–105.
[4] J. Goddard, Dissipative materials as models of thixotropy and plasticity, Journal of Non-Newtonian
Fluid Mechanics 14 (1984) 141 – 160.
[5] M. Gurtin, E. Fried, L. Anand, The Mechanics and Thermodynamics of Continua, Cambridge, 2010.
[6] R. B. Bird, R. C. Armstrong, O. Hassager, Dynamics of Polymeric Liquids Vol. 1, John Wiley and
Sons, second edition, 1987.
[7] R. G. Larson, The Structure and Rheology of Complex Fluids, Oxford University Press, 1999.
[8] G. Lipscomb, M. Denn, Flow of bingham fluids in complex geometries, Journal of Non-Newtonian
Fluid Mechanics 14 (1984) 337 – 346.
[9] A. N. Beris, J. A. Tsamopoulos, R. C. Armstrong, R. Brown, Creeping motion of a sphere through a
bingham plastic, Journal of Fluid Mechanics 158 (1985) 219–244.
[10] N. J. Balmforth, R. V. Craster, R. Sassi, Shallow viscoplastic flow on an inclined plane, Journal of
Fluid Mechanics 470 (2002) 1–29.
[11] B. T. Liu, S. J. Muller, M. M. Denn, Convergence of a regularization method for creeping flow of a
bingham material about a rigid sphere, Journal of Non-Newtonian Fluid Mechanics 102 (2002) 179 –
191.
[12] G. Vinay, A. Wachs, I. Frigaard, Start-up transients and efficient computation of isothermal waxy
crude oil flows, Journal of Non-Newtonian Fluid Mechanics 143 (2007) 141–156.
[13] G. Ovarlez, Q. Barral, P. Coussot, Three-dimensional jamming and flows of soft glassy materials, Nat
Mater 9 (2010) 115–119.
[14] L. Martinie, H. Buggisch, N. Willenbacher, Apparent elongational yield stress of soft matter, Journal
of Rheology 57 (2013) 627–646.
[15] N. J. Balmforth, I. A. Frigaard, G. Ovarlez, Yielding to stress: Recent developments in viscoplastic
fluid mechanics, Annual Review of Fluid Mechanics 46 (2014) 121–146.
[16] A. S. Yoshimura, R. K. Prud’homme, Response of an elastic bingham fluid to oscillatory shear, Rheo-
logica Acta 26 (1987) 428–436.
[17] D. Doraiswamy, A. N. Mujumdar, I. Tsao, A. N. Beris, S. C. Danforth, A. B. Metzner, The Cox–Merz
rule extended: A rheological model for concentrated suspensions and other materials with a yield stress,

48
 Journal of Rheology 35 (1991) 647–685.
[18] P. Marmottant, F. Graner, An elastic, plastic, viscous model for slow shear of a liquid foam, European
Physical Journal E 23 (2007).
[19] S. Bénito, C. H. Bruneau, T. Colin, C. Gay, F. Molino, An elasto-visco-plastic model for immortal
foams or emulsions, European Physical Journal E 25 (2008).
[20] P. Saramito, A new constitutive equation for elastoviscoplastic fluid flows, Journal of Non-Newtonian
Fluid Mechanics 145 (2007) 1 – 14.
[21] P. Saramito, A new elastoviscoplastic model based on the Herschel–Bulkley viscoplastic model, Journal
of Non-Newtonian Fluid Mechanics 158 (2009) 154 – 161.
[22] J. M. Piau, Carbopol gels: Elastoviscoplastic and slippery glasses made of individual swollen sponges
Meso- and macroscopic properties, constitutive equations and scaling laws, Journal of Non-Newtonian
Fluid Mechanics 144 (2007) 1–29.
[23] P. Møller, A. Fall, V. Chikkadi, D. Derks, D. Bonn, An attempt to categorize yield stress fluid behavior,
Phil. Trans. R. Soc. A 367 (2009) 5139–5155.
[24] P. C. F. Møller, A. Fall, D. Bonn, Origin of apparent viscosity in yield stress fluids below yielding,
EPL (Europhysics Letters) 87 (2009) 38004.
[25] M. Denn, D. Bonn, Issues in the flow of yield-stress liquids, Rheologica Acta 50 (2011) 307–315.
[26] P. Saramito, A. Wachs, Progress in numerical simulation of yield stress fluid flows, Rheologica Acta
56 (2017) 211–230.
[27] R. Glowinski, A. Wachs, On the numerical simulation of viscoplastic fluid flow, Handbook of Numerical
Analysis 16 (2011) 483–717.
[28] T. C. Papanastasiou, Flows of materials with yield, Journal of Rheology 31 (1987) 385–404.
[29] M. Dinkgreve, M. M. Denn, D. Bonn, “Everything flows?”: elastic effects on startup flows of yield-stress
fluids, Rheologica Acta 56 (2017) 189–194.
[30] C. J. Dimitriou, R. H. Ewoldt, G. H. McKinley, Describing and prescribing the constitutive response
of yield stress fluids using large amplitude oscillatory shear stress (LAOStress), Journal of Rheology
57 (2013) 1–44.
[31] C. J. Dimitriou, G. H. McKinley, A comprehensive constitutive law for waxy crude oil: a thixotropic
yield stress fluid, Soft Matter 10 (2014) 6619–6644.
[32] M. Geri, R. Venkatesan, K. Sambath, G. H. McKinley, Thermokinematic memory and the thixotropic
elasto-viscoplasticity of waxy crude oils, Journal of Rheology 61 (2017) 427–454.
[33] P. J. Armstrong, C. O. Frederick, A mathematical representation of the multiaxial Bauschinger effect,
Materials at High Temperatures 24 (1966) 1–26.
[34] C. Dimitriou, The rheological complexity of waxy crude oils: Yielding, thixotropy and shear hetero-

49
 geneities, Ph.D. thesis, Massachusetts Institute of Technology, 2013.
[35] P. Lidon, L. Villa, S. Manneville, Power-law creep and residual stresses in a carbopol gel, Rheologica
Acta 56 (2017) 307–323.
[36] J. Bauschinger, Über die veränderung der position der elastizitätsgrenze des eisens und stahls
durch strecken und quetschen und durch erwärmen und abkühlen und durch oftmals wiederholte
beanspruchungen, Mitteilung aus dem Mechanisch-technischen Laboratorium der Königlichen poly-
technischen Hoschschule in München 13 (1886) 1–115.
[37] Y. Wei, M. J. Solomon, R. G. Larson, A multimode structural kinetics constitutive equation for the
transient rheology of thixotropic elasto-viscoplastic fluids, Journal of Rheology 62 (2018) 321–342.
[38] D. Fraggedakis, Y. Dimakopoulos, J. Tsamopoulos, Yielding the yield stress analysis: A thorough
comparison of recently proposed elasto-visco-plastic (EVP) fluid models, Journal of Non-Newtonian
Fluid Mechanics 238 (2016) 170 – 188.
[39] R. R. Fernandes, D. E. V. Andrade, A. T. Franco, C. O. R. Negro, The yielding and the linear-
to-nonlinear viscoelastic transition of an elastoviscoplastic material, Journal of Rheology 61 (2017)
893–903.
[40] D. Fraggedakis, Y. Dimakopoulos, J. Tsamopoulos, Yielding the yield-stress analysis: a study focused
on the effects of elasticity on the settling of a single spherical particle in simple yield-stress fluids, Soft
Matter 12 (2016) 5378–5401.
[41] P. Coussot, L. Tocquer, C. Lanos, G. Ovarlez, Macroscopic vs. local rheology of yield stress fluids,
Journal of Non-Newtonian Fluid Mechanics 158 (2009) 85–90.
[42] M. Dinkgreve, J. Paredes, M. M. Denn, D. Bonn, On different ways of measuring the yield stress,
Journal of Non-Newtonian Fluid Mechanics 238 (2016) 233 – 241.
[43] L. Anand, N. M. Ames, V. Srivastava, S. Chester, A thermo-mechanically coupled theory for large
deformations of amorphous polymers. Part I: Formulation, International Journal of Plasticity 25 (2008)
1474–1494.
[44] N. M. Ames, V. Srivastava, S. Chester, L. Anand, A thermo-mechanically coupled theory for large
deformations of amorphous polymers. Part II: Applications, International Journal of Plasticity 25
(2009) 1495–1539.
[45] D. L. Henann, L. Anand, A large deformation theory for rate-dependent elastic–plastic materials with
combined isotropic and kinematic hardening, International Journal of Plasticity 25 (2009) 1833 – 1878.
[46] E. Kroner, Allgemeine kontinuumstheorie der versetzungen und eigenspannungen, Archive for Rational
Mechanics and Analysis 4 (1960) 273–334.
[47] M. Habibi, M. Dinkgreve, J. Paredes, M. Denn, D. Bonn, Normal stress measurement in foams and
emulsions in the presence of slip, Journal of Non-Newtonian Fluid Mechanics 238 (2016) 33 – 43.

50
[48] R. B. Bird, R. C. Armstrong, O. Hassager, Dynamics of Polymeric Liquids Vol. 2, John Wiley and
Sons, second edition, 1987.
[49] P. Coussot, A. I. Leonov, J. M. Piau, Rheology of concentrated dispersed systems in a low molecular
weight matrix, Journal of Non-Newtonian Fluid Mechanics 46 (1993) 179–217.
[50] D. Quemada, Rheological modelling of complex fluids: IV: Thixotropic and “thixoelastic” behaviour.
start-up and stress relaxation, creep tests and hysteresis cycles, The European Physical Journal of
Applied Physics 5 (1999) 191–207.
[51] K. Dullaert, J. Mewis, A structural kinetics model for thixotropy, Journal of Non-Newtonian Fluid
Mechanics 139 (2006) 21–30.
[52] P. R. de Souza Mendes, R. L. Thompson, A critical overview of elasto-viscoplastic thixotropic modeling,
Journal of Non-Newtonian Fluid Mechanics 187–188 (2012) 8–15.
[53] J. Mewis, N. J. Wagner, Thixotropy, Advances in Colloid and Interface Science 147–148 (2009) 214–227.
[54] R. H. Ewoldt, G. H. McKinley, Mapping thixo-elasto-visco-plastic behavior, Rheologica Acta 56 (2017)
195–210.
[55] A. Mujumdar, A. N. Beris, A. B. Metzner, Transient phenomena in thixotropic systems, Journal of
Non-Newtonian Fluid Mechanics 102 (2002) 157 – 178.
[56] R. G. Larson, Constitutive equations for thixotropic fluids, Journal of Rheology 59 (2015) 595–611.
[57] F. A. Morrison, Understanding Rheology, Oxford University Press, first edition, 2001.

51