J. Non-Newtonian Fluid Mech.

108 (2002) 301326

Viscoelastic analysis of complex polymer melt ows using
the eXtended PomPom model
Wilco M.H. Verbeeten, Gerrit W.M. Peters

, Frank P.T. Baaijens

Materials Technology, Faculty of Mechanical Engineering, Eindhoven University of Technology,
P.O. Box 513, 5600 MB Eindhoven, The Netherlands
Abstract
The ability of the multi-mode eXtended PomPom model to predict inhomogeneous ows of a polyethylene melt
is investigated. Two benchmark problems are examined: the conned ow around a cylinder, and the ow through
a cross-slot device. Numerical results for the eXtended PomPom model are compared to experimental data, and
predictions of the Giesekus and exponential Phan-Thien Tanner (PTT-a) models. Characteristic features observed
experimentally in the benchmark ows are described well by all three models. The eXtended PomPom model
performs most satisfactorily, both with respect to the rheological data and the inhomogeneous ow data.
2002 Elsevier Science B.V. All rights reserved.
Keywords: Constitutive models; Differential models; eXtended PomPom model; PTT-a model; Giesekus model; Polymer
melt; Flow around a cylinder; Cross-slot ow; Experimental results; Quantitative comparison
1. Introduction
A quantitative description of the rheological behaviour of polymer melts is crucial in understanding
the relation between processing and product properties. As an intermediate step between the well-dened
rheometrical ows and complicated industrial processing ows, simplied, experimentally accessible,
inhomogeneous ows that exhibit a combination of transient shear and elongational deformation are
investigated. The viscoelastic analysis of these ows allows the assessment of constitutive models and
numerical predictions for prototype industrial ows.
One of the main problems in constitutive modeling is to obtain a correct description of the transient
nonlinear behaviour in elongational and shear ows simultaneously. Well-known and widely used models,
such as the multi-mode exponential Phan-Thien Tanner (PTT-a) and the Giesekus model yield unsatisfac-
tory results in the prediction of inhomogeneous ows [2]. Recently, a new class of constitutive equations
has been introduced based on the tube model of Doi and Edwards [9], which is able to simultaneously

Corresponding author. Fax: +31-40-244-7355.

E-mail address: gerrit@wfw.wtb.tue.nl (G.W.M. Peters).
URL: http://www.mate.tue.nl/
0377-0257/02/$ see front matter 2002 Elsevier Science B.V. All rights reserved.
PII: S0377- 0257( 02) 00136- 2
302 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
describe the transient shear and elongational behaviour of polyethylene melts. Some of them start off
in an integral version, e.g. the PomPom model [22,14,6], but an approximation of the differential type
is also given for computational convenience. Others are only available as an integral model, e.g. the
Molecular Stress Function (MSF) model [29], and no differential approximation exists. Although the
newly developed deformation elds method by Peters et al. [24] is an attractive way to calculate models
of the integral type, numerical methods using differential constitutive equations are still more efcient.
Verbeeten et al. [33] introduced the eXtended PomPommodel, a modication of the original differential
PomPom model of McLeish and Larson [22], which is able to describe the behaviour of polyethylene
melts over a wide range of rheometric data, including reversed ow. This last rheological ow is still
rather troublesome for most integral models. In this work, we will restrict ourselves to the investigation
of differential constitutive equations.
Bishko et al. [5] already showed results for the original differential PomPom model in a complex
4:1 contraction ow. Although only a single-mode version of the PomPom model was used, good
qualitative predictions were obtained for highly branched polymers. Their computational results are
promising and predicted all the specic features observed in experiments for LDPE melts. However,
a full comparison with experimental complex ow data was not possible, since a multi-mode version
is necessary for an accurate description of the rheology of polymer melts. Furthermore, due to the ex-
cessive shear-thinning behaviour of the original differential PomPom model in fast shear ows, the
shear stress versus shear rate curve shows a maximum. This induces a non physical constitutive in-
stability. A multi-mode version will avoid this problem on a sufciently wide range of shear rates,
because modes with higher relaxation times will dominate after the shear-thinning behaviour sets in
for the faster modes, resulting in a less shear-thinning behaviour. Moreover, as the shear-thinning be-
haviour of the eXtended PomPom model is less extreme, this particular model effectively avoids this
instability.
Like in Bishko et al. [5], most studies of polymer melts are conducted on the benchmark contraction or
contraction/expansion problem, e.g. Xue et al. [35], Wapperom and Keunings [34], and Alves et al. [1]
for purely numerical research, Martyn et al. [1820] for a mostly experimental investigation in a three
dimensional geometry, andBraudoet al. [4] for a combinednumerical/experimental study. Unfortunately,
material elements over the centerline of contraction problems are extended only moderately, because the
elongational rate is witnessed during a too short time interval. Much less research has been performed on
the ow around a conned cylinder, e.g. Peters et al. [25] and Renardy [28] for numerical studies, and
Schoonen [30], Baaijens et al. [2], Hartt and Baird [13] and Baaijens [3] for a combined experimental
and numerical investigation. A viscoelastic analysis of a polymer melt in a cross-slot device has only
been published once to our knowledge [27]. These latter two geometries impose large strain rates on the
material over an extended period of time inducing nite strains and are therefore discriminating towards
constitutive models with different steady elongational properties.
The objective of this study is to assess the performance of the multi-mode eXtended PomPom model
[33] for two benchmark ows, i.e. the conned ow around a cylinder and the cross-slot ow, and
compare it with experimental results taken from Schoonen [30]. Both geometries were designed to have
nearly two-dimensional ow kinematics and birefringence, i.e. a depth-to-height aspect ratio of at least
eight in the whole ow domain, such that comparison with two-dimensional calculations is allowed
[30,31]. Furthermore, the results will be compared against the performance of the exponential PTT and
the Giesekus model. The differences and similarities between the two prototype industrial ows will be
examined with respect to strain rates, stresses and their history in a Lagrangian sense. We will discuss
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 303
the advantages and drawbacks of these ows when used for testing constitutive models by comparing
experimental and numerical results.
For the numerical part of this study, the Discrete Elastic Viscous Stress Splitting technique in combina-
tion with the Discontinuous Galerkin method (DEVSS/DG) is used. In Section 2, the problem denition
and the constitutive equations are outlined. The computational method is briey described in Section 3
and can be found in more detail in Bogaerds et al. [7]. The material characterization and the performance
of the three constitutive models in simple rheometric ows is given in Section 4. Results for the two
prototype industrial ow geometries are presented in Section 5, followed by a discussion and conclusions
on the different ows. It is shown that all three models are able to qualitatively, and to a large extent also
quantitatively, predict the velocities and stresses in the different complex ow geometries. In general, the
eXtended PomPom model shows the best quantitative comparison.
2. Problem denition
Isothermal and incompressible uid ows, neglecting inertia, are described by the equations for con-
servation of momentum (1) and mass (2):

=

0, (1)

u = 0, (2)
where

is the gradient operator, denotes the Cauchy stress tensor and u the velocity eld. The Cauchy
stress tensor is dened for polymer melts by:
= pI +2
s
D +
M

i=1

i
. (3)
Here, p is the pressure, I the unit tensor,
s
denotes the viscosity of the purely viscous (or solvent) contri-
bution, D = (1/2)(L+L
T
) the rate of deformation tensor, in which L = (

u)
T
is the velocity gradient
tensor and ()
T
denotes the transpose of a tensor. The visco-elastic contribution of the ith relaxation mode
is denoted by
i
and M is the total number of different modes. The multi-mode approach is necessary for
most polymeric uids to give a realistic description of stresses over a broad range of deformation rates.
The visco-elastic contribution
i
still has to be dened by a constitutive model.
2.1. Constitutive models
Within the scope of this work, a sufciently general way to describe the constitutive behaviour of a
single-mode is obtained by using a differential equation based on the generalized Maxwell-type equation:

i
+f
GS
(
i
, D) +(
i
)
1

i
= 2G
i
D, (4)
with G
i
the plateau modulus of the ith mode obtained from the linear relaxation spectrum determined by
dynamic measurements. The upper convected time derivative of the stress

i
is dened as:

i
=
i
L
i

i
L
T
=

i
t
+ u

i
L
i

i
L
T
, (5)
304 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
where t denotes time. Both tensorial functions f
GS
(
i
, D) and (
i
) depend upon the chosen constitutive
equation. Notice, that by choice of f
GS
(
i
, D) = 0 and (
i
) =
i
, with
i
the ith linear relaxation
time, the Upper Convected Maxwell (UCM) model is retrieved.
For the Giesekus model we have:
f
GS
(
i
, D) = 0, (
i
)
1
=
1

i
_

i
G
i

i
+I
_
, (6)
with
i
a material parameter to be tted on the rheological data. For the exponential PTT-a model it holds
that:
f
GS
(
i
, D) =
i
(D
i
+
i
D), (
i
)
1
=
1

i
[e
(
i
/G
i
)Tr(
i
)
I], (7)
with
i
the slip parameter from the Gordon-Schowalter derivative and
i
a material parameter, both to be
tted on the rheological data.
A new class of constitutive relations based on the tube model and a simplied topology of branched
molecules was recently proposed by McLeish and Larson. The basic idea of the model is that the rheo-
logical properties of entangled polymer melts mainly depend on the topological structure of the polymer
molecules. The simplied topology consists of a backbone with a number of dangling arms at both
ends and is called a PomPom molecule. The backbone is conned by a tube formed by other back-
bones. A key feature is the separation of relaxation times for the stretch and orientation. For more
details we refer to [22,33]. Verbeeten et al. [33] incorporated local branch-point displacement as intro-
duced by Blackwell et al. [6] and modied the orientation equation of the original differential version
to the eXtended PomPom model to overcome three problems: solutions in steady state elongation
show discontinuities; the equation for orientation is unbounded for high strain rates; the model does
not have a second normal stress difference in shear. A branched molecule can be represented by sev-
eral equivalent PomPom modes. The extra stress equation for the eXtended PomPom (XPP) model is
dened as:

i
= G
i
I = 3G
i

2
i
S
i
G
i
I, (8)
with
i
the backbone tube stretch, dened as the length of the backbone tube divided by the length at
equilibrium, and S
i
denoting the orientation tensor. The evolution equations for the orientation, based on
the Giesekus model as given in Eq. (6), and backbone tube stretch are given by:

S
i
+2[D : S
i
]S
i
+
1

b,i

2
i
_
3
i

4
i
S
i
S
i
+(1
i
3
i

4
i
Tr(S
i
S
i
))S
i

(1
i
)
3
I
_
=0, (9)

i
=
i
[D : S
i
]
e
v
i
(
i
1)

s,i
(
i
1), v
i
=
2
q
i

b,i
=
i
. (10)
Here,
b,i
is the relaxation time of the backbone tube orientation equal to the linear relaxation time
i
,
s,i
is the relaxation time for the stretch, v
i
a parameter denoting the inuence of the surrounding polymer
chains on the backbone tube stretch, and q
i
is the amount of arms at the end of a backbone.
The XPP model can also be written in a fully equivalent single-equation fashion [33], which has the
same structure as the Giesekus and PTT-a models. This facilitates implementation in existing software
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 305
packages. In that case the functions from Eq. (4) are dened as:
f
GS
(
i
, D) = 0, (
i
)
1
=
1

b,i
_

i
G
i

i
+F(
i
)I +G
i
(F(
i
) 1)
1
i
_
, (11)
with
F(
i
) = 2r
i
e
v
i
(
i
1)
_
1
1

i
_
+
1

2
i
_
1

i
Tr(
i

i
)
3G
2
i
_
(12)
and

i
=
_
1 +
Tr(
i
)
3G
i
, r
i
=

b,i

s,i
, v
i
=
2
q
i

b,i
=
i
. (13)
Notice, that for this denition of the backbone stretch, we may run into numerical problems if
1 + (Tr(
i
)/3G
i
) < 0. Physically, Tr(
i
) can not become smaller than 0. However, numerically we
have encountered these unrealistic values at the front and back stagnation points in the ow around a
cylinder at sufciently high Weissenberg numbers. Unphysical negative backbone stretch values in the
double-equation XPP model were also encountered at the exact same locations for higher Weissenberg
numbers than presented in this work. This is a numerical artifact, similar to the negative values for the trace
of the extra stress, Tr(
i
), encountered in the Giesekus and PTT-a models for the Weissenberg number
shown in this paper, independent of the meshes used. In this respect, the cross-slot ow is a smooth ow
since no a-physical stretches are predicted, either for the single-equation and the double-equation XPP
model. We would like to remark, that these unrealistic negative backbone stretch values have not been
encountered while computing homogeneous rheometrical ows.
A disadvantage for a two-dimensional numerical implementation, concerning both versions, is the
non-zero third stress component
zz
, contrary to the Giesekus and PTT-a models. Compared to the
single-equation form, the double-equation formulation has an extra equation for the stretch, resulting
in a larger system matrix in FE codes. However, the equation for the third orientation component S
zz
can
be eliminated, since it is known that Tr(S
i
) = 1. As we eliminate the extra stresses, or, equivalently, the
orientation tensor and the stretch, at the element level (see Section 3), the computational cost of these
two extra equations is relatively low. Therefore, elimination of S
zz
is not performed.
From the computational point of view the double-equation XPP model is preferred because computa-
tions do not fail if during the iterative process negative values of the backbone stretch are computed. The
single-equation XPP model causes the computations to stop if (Tr(
i
)/3G
i
) < 1, which is a-physical
since Tr(
i
) should be positive, but may occur during the iterative process.
3. Computational method
The DEVSS/DG method, Baaijens et al. [2], which is a combination of the Discrete Elastic Viscous
Stress Splitting (DEVSS) technique of Gunette and Fortin [10] and the Discontinuous Galerkin (DG)
method, developed by Lesaint and Raviart [16], is used. Application to the double-equation XPP model is
discussed in this section. The weak formulation and solution strategy for the Giesekus and PTT-a models
is similar and can be found in more detail in Bogaerds et al. [7] and Baaijens et al. [2].
306 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
3.1. DEVSS/DG method
Consider the problem described by the Eqs. (1), (2), (9) and (10). Following common FE techniques,
these equations are converted into a mixed weak formulation:
Problem DEVSS/DG. Find ( u, p,

D, S
i
,
i
) for any time t such that for all admissible test functions
( v, q, E, s
i
, l
i
),
_
D
v
, 2
s
D +2 (D

D) +
M

i=1
G
i
(3
2
i
S
i
I)
_
( v, p) = 0, (14)
(q, u) = 0, (15)
(E,

D D) = 0, (16)
_
s
i
,

S
i
+2[D : S
i
]S
i
+
1

b,i

2
i
_
3
i

4
i
S
i
S
i
+(1
i
3
i

4
i
Tr(S
i
S
i
))S
i

(1
i
)
3
I
__

e=1
_

e
in
s
i
: u n(S
i
S
ext
i
) d = 0 i {1, 2, . . . , M}, (17)
_
l
i
,

i

i
[D : S
i
] +
e
v
i
(
i
1)

s,i
(
i
1)
_

e=1
_

e
in
l
i
u n(
i

ext
i
) d = 0 i {1, 2, . . . , M}. (18)
Here, (, ) denotes the L
2
-inner product on the domain , D
v
= (1/2)(

v + (

v)
T
), an auxiliary
viscosity,
e
in
is the inow boundary of element
e
, n the unit vector pointing outward normal on the
boundary of the element (
e
) and S
ext
i
and
ext
i
denote the orientation tensor and backbone stretch of the
neighboring, upstream element.
The stabilization term 2 (D

D) has been added to the momentum Eq. (14), where the discrete
approximation

D is obtained from an L
2
projection of the rate of deformation tensor (Eq. (16)). The
auxiliary viscosity =

M
i=1
G
i

i
is found to give satisfactory results.
Time discretisation of the constitutive equation is obtained using an semi-implicit Euler scheme. In
this scheme most of the variables are updated implicitly with the exception of the external components
S
ext
i
and
ext
i
which are taken explicitly (i.e. S
ext
i
= S
ext
i
(t
n
) and
ext
i
=
ext
i
(t
n
)). Hence, the terms
_
s
i
: u n(S
ext
i
) d and
_
l
i
u n(
ext
i
) d have no contributions to the Jacobian which allows for
elimination of the orientation tensor and backbone stretch at element level. Furthermore, the equation for
the discrete rate of deformation

D is decoupled from the Stokes problem and thus updated explicitly.
3.2. Solution strategy
In order to obtain an approximation of ProblemDEVSS/DG, a 2Ddomain is divided into quadrilateral
elements. A bi-quadratic interpolation for the velocity u, bi-linear for the pressure p and discrete rate
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 307
of deformation

D, and a discontinuous bi-linear interpolations for the orientation tensor S
i
and stretch

i
are known to give stable results (see [7] and [2]). Integration of Eqs. (14)(18) over an element is
performed using a 3 3-Gauss quadrature rule common in FE analysis.
To obtain the solution of the nonlinear equations, a one step Newton-Raphson iteration process is
carried out. Consider the iterative change of the nodal degrees of freedom(
u
,
p
,

D
,
S
,

) as variables
of the algebraic set of linearized equations. This linearized set is given by:
_
_
_
_
_
_
_
_
_
Q
uu
Q
up
Q
u

D
Q
uS
Q
u
Q
pu
0 0 0 0
Q

Du
0 Q

D

D
0 0
Q
Su
0 0 Q
SS
Q
S
Q
u
0 0 Q
S
Q

_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_
_

_
_
_
_
_
_
_
_
_
=
_
_
_
_
_
_
_
_
_
f
u
f
p
f

D
f
S
f

_
_
_
_
_
_
_
_
_
, (19)
with f

( = u, p,

D, S
i
,
i
) correspond to the residuals of equations Eqs. (14)(18), while Q

follow
from linearisation of these equations. Due to the fact that S
ext
i
and
ext
i
have been taken explicitly in
Eqs. (17) and (18), the matrices Q
SS
, Q
S
, Q
S
, and Q

form a block diagonal structure which allows

for calculation of
_
Q
SS
Q
S
Q
S
Q

_
1
on element level. This matrix is recomputed at each Newton iteration.
Consequently, this enables the reduction of the global DOFs by static condensation of the orientation
tensor and backbone stretch block.
Application of a iterative solvers, such as Bi-CGSTAB, to the resulting set of equations proved unsuc-
cessful. Therefore, a decoupling procedure is adopted. First, the Stokes problem ( u, p) is solved using

D of the previous solution, after which the updated solution for velocities and pressure is used to nd a
new approximation for

D. The following two separate problems now emerge:
Problem DEVSS/DG
a
. Given ( u, p,

D, S
i
,
i
) at t = t
n
, nd a solution at t = t
n+1
of the algebraic set:
_
_
_
_
Q
uu
(Q
uS
Q
u
)
_
Q
SS
Q
S
Q
S
Q

_
1
_
Q
Su
Q
u
_
Q
up
Q
pu
0
_
_
_
_
_

p
_
=
_
_
_
_
f
u
(Q
uS
Q
u
)
_
Q
SS
Q
S
Q
S
Q

_
1
_
f
S
f

_
f
p
_
_
_
_
, (20)
and
Problem DEVSS/DG
b
. Given

D at t = t
n
and u at t = t
n+1
, nd

D
from:
(Q

D

D
)(

D
) = (f

D
+Q

Du

u
). (21)
Notice that the right hand side is now taken with respect to the new velocity approximation. The no-
dal increments of the orientation tensor and backbone stretch are retrieved element by element
308 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
following:
_

_
=
_
Q
SS
Q
S
Q
S
Q

_
1
_
f
S
f

_
, (22)
with f
S
and f

also taken with respect to the new velocity approximation (f

S
(S
n
i
, u
n+1
), f

(
n
i
, u
n+1
)).
To solve the non-symmetrical system of Problem DEVSS/DG
a
an iterative solver is used based on
the Bi-CGSTAB method of Van der Vorst [32]. The symmetrical set of algebraic equations of Problem
DEVSS/DG
b
is solved using a Conjugate-Gradient solver. Incomplete LU preconditioning is applied to
both solvers. In order to enhance the computational efciency of the Bi-CGSTAB solver, the velocity
variables of the center node are eliminated at element level by static condensation. Besides a reduction
of the number of degrees of freedom, the key result of this is that the zero lower block diagonal matrix
related to the pressure variables becomes non-zero, substantially improving the rate of convergence of
the iterative solver.
Finally, to solve the above sets of algebraic equations, both essential and natural boundary conditions
must be imposed on the boundaries of the ow channels. At the entrance and the exit of the ow channels
the fully developed velocity proles are prescribed. At the entrance, the values of the orientation tensor
and backbone stretch of a few elements downstream the inow channel are prescribed along the inow
boundary, imposing a periodic boundary condition for these stress variables. In this way, the orientation
tensor and backbone stretch at the inow boundary develop similarly as the stress variables in the rest of
the ow domain.
4. Material characterization
The polymer melt that is investigated in this work is a commercial grade low density polyethylene
(DSM, Stamylan LD 2008 XC43), further referred to as LDPE. This long-chain branched material has
been extensively characterized by Schoonen [30] and was also used in Peters et al. [27].
In the linear viscoelastic regime, all three constitutive models investigated in this work reduce to the
linear Maxwell model. In all cases, four modes have used to describe the linear viscoelastic data, obtained
from dynamic measurements at a temperature of T = 170

C. The non-linearity parameters q and the

ratio
b
/
s
for the XPP model and for the Giesekus model are determined using the transient uniaxial
elongational data only. For the PTT-a model and are identied by using transient uniaxial elongational
and shear data. Since second planar elongational or second normal stress difference data is not available for
this material, the anisotropy parameter in the XPP model is chosen as 0.3/q. Anisotropy is decreasing
from the free ends inwards, and by choosing as an inverse function of the number of arms q, this is
indeed accomplished (see [33]). All parameters are given in Table 1.
Fig. 1 shows the rheological behaviour of the three constitutive models and measured data in uniaxial
and simple shear behaviour. The steady state data points in the uniaxial elongational viscosity plot are
the end points of the transient curves. Since the samples break at the end of the transient measurements,
it is hard to say if steady state has truly been reached. However, sometimes it is claimed that steady state
coincides with break-up of the samples [21].
Within the experimental range, all three models give a reasonable to good prediction of major portions of
the transient uniaxial elongational viscosity curves. Signicant differences are observed in the steady-state
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 309
Fig. 1. Transient (a) and quasi-steady state (b) uniaxial viscosity
u
, transient (c) and steady state (d) shear viscosity
s
, and
transient (e) and steady state (f) rst normal stress coefcient
1
at T = 170

C of the XPP, Giesekus and PTT-a models for

DSM Stamylan LD2008 XC43 LDPE melt.
310 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
Table 1
Linear and non-linear parameters for tting of the DSM Stamylan LD 2008 XC43 LDPE melt
i Maxwell parameters XPP Giesekus Exponential PTT
G
i
(Pa)
i
(s) q
i

b,i
/
s,i

i

i

i

i
1 7.2006 10
4
3.8946 10
3
1 7.0 0.30 0.30 0.30 0.08
2 1.5770 10
4
5.1390 10
2
1 5.0 0.30 0.30 0.20 0.08
3 3.3340 10
3
5.0349 10
1
2 3.0 0.15 0.25 0.02 0.08
4 3.0080 10
2
4.5911 10
0
10 1.1 0.03 0.04 0.02 0.08
T
r
= 170

C; v
i
= 2/q
i
; activation energy: E
0
= 48.2 kJ/mol; WLF-shift parameters: C
1
= 14.3 K, C
2
= 480.8 K.
curves. The XPP model captures the available data points best, while the PTT-a model demonstrates a
non-smooth behaviour, reecting the individual relaxation times. This could be improved by choosing a
larger number of relaxation times. Based on measurements of other low density polyethylene melts (see
[23,11]), the uniaxial elongational viscosity at high strain rates is expected to be elongational thinning.
Therefore, the plateau value predicted by the Giesekus model is not realistic.
For simple shear, all models show a rather good agreement with the measured data (Fig. 1c and d).
The Giesekus model somewhat over predicts the measurements. At shear rates exceeding 10
2
(s
1
), the
curves for all models drop rapidly (see Fig. 1d), corresponding to the longest relaxation time. Since
these high shear rates do not occur in the experiments, this does not cause difculties in the numerical
simulation of these ows. Although difcult to observe in Fig. 1c, the PTT-a model predicts oscillations
in transient start-up of simple shear. These oscillations occur because the non-linearity parameter had
to be chosen larger than to obtain a good t of the shear data. These oscillations are a drawback of
the PTT-a model. Appendix A elaborates more on this issue for a different low density polyethylene
melt.
All three models slightly over predict the rst normal stress coefcient. In the start-up region of the
rst normal stress coefcient curve, the different relaxation times of the XPP model can be detected (see
Fig. 1e). This could be avoided by choosing more Maxwell modes. As mentioned before, this would also
result in a smoother curve for the steady state uniaxial viscosity predicted by the PTT-a model. However,
as computational time increases signicantly with every mode added, four Maxwell modes is considered
as a satisfactory optimum between a good description of the rheological behaviour and computational
cost.
5. Complex ows
Acomparison is made between numerical and experimental results for two complex ows: owaround
a cylinder between two parallel plates and owthrough a cross-slot device. These benchmark owgeome-
tries are known to have regions with pure simple shear, pure planar elongation and combinations thereof.
Velocities have been measured using particle tracking velocimetry, while ow induced birefringence in
combination with the stress-optical rule is used for comparison with numerical stress computations. For
each geometry, we will present results for one ow rate only. These ow rates differ for the cylinder
and cross-slot ow geometry, and have been chosen such that the the maximum planar elongational rate
achieved in both geometries is in the same order of magnitude, about 2 s
1
for the cylinder problem and
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 311
about 1.5 s
1
for the cross-slot geometry. For more details on the experimental aspects and extended
experimental results, we refer to Schoonen [30].
Both geometries are designed to have near two-dimensionality, i.e. a depth-to-height aspect ratio of
at least 8:1 throughout the whole ow domain. Schoonen [30] demonstrated that with an aspect ratio of
8:1 the inuence of the conning front and back walls is about 6% maximum on the isochromatic stress
patterns within the ow rate range used in this investigation. This was conrmed in numerical studies
by Schoonen et al. [31] and Bogaerds et al. [7]. This is a sufciently small deviation to be nominally
two-dimensional and compare experimental results to two-dimensional calculations. The general rule for
deviations due to three-dimensional effects is that the integrated effect of experimental stresses (which
represents the birefringence patterns) is lower near walls and higher over centerlines in comparison to
two-dimensional calculations.
To characterize the strength of the different ows, the Weissenberg number is dened as:
Wi =

u
2D
h
. (23)
Here,

denotes the viscosityaveragedrelaxationtime for the material

_

=
_

M
i=1

2
i
G
i
_
/
_

M
i=1

i
G
i
__
,
u
2D
is the two-dimensional mean velocity and h a characteristic length of the ow geometry.
To link the calculated stresses to the measured isochromatic fringe patterns, the semi-empirical stress-
optical rule is used. It states that the deviatoric part of the refractive index tensor is proportional to the
deviatoric part of the stress tensor:
n
d
= C
d
, (24)
in which C is the stress-optical coefcient, identied to equal (see [27]):
C = 1.47 10
9
Pa
1
. (25)
The stress-optical coefcient is slightly dependent on temperature [17].
For two-dimensional ows, the stress-optical rule can be simplied to:

FRG
=
_
4
2
xy
+N
2
1
=
k
dC
, (26)
with
FRG
dened as the isochromatic fringe stress,
xy
the shear stress in the xy-plane, N
1
=
xx

yy
the
rst normal stress difference, the wave length of the light used in the measurements, d the light path
length in the birefringent medium, which in these cases are the depths of the ow cells, and k the fringe
order of the observed dark fringe bands where extinction of the light occurs.
5.1. Flow around a cylinder
A schematic representation of the planar cylinder ow geometry is shown in Fig. 2(a). Experiments
were performed at a temperature of 170

C, resulting in a viscosity averaged relaxation time for the

material of

= 1.7415 s. The radius of the cylinder is R = 1.1875 mm, the channel height equals
H = 4.95 mm, and the depth of the ow cell d = 40 mm. The depth-to-height ratio results in 8.08,
creating a nominally two-dimensional ow. The height-to-diameter ratio equals 2.08. The characteristic
312 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
Fig. 2. Schematic (a) and detail of FE meshes (b, c, d) of the planar ow around a cylinder. The origin of the coordinate
system is at the center of the cylinder. T = 170

C,

= 1.7415 s,
max
= 4.5910 s, u
2D
= 1.975 mm/s, R = 1.1875 mm,
H = 4.95 mm 4R, d = 40 mm.
mean two-dimensional velocity at the inow is u = 1.975 mm/s. As a characteristic length, the radius of
the cylinder R is chosen, giving the ow averaged Weissenberg number:
Wi = 2.9. (27)
Notice, that the averaged Weissenberg number of the longest relaxation time used in the calculations is
Wi = 7.6, and may even be much higher locally.
Only half of the geometry needs to be analyzed due to symmetry. The center of the cylinder is placed
at x/R = 0, y/R = 0, while the inlet is at x/R = 8 and the oulet at x/R = 15. The periodic boundary
for the inlet stresses is situated at x/R = 5. At the entrance and exit, a fully developed velocity prole
is prescribed. No slip boundry conditions are imposed on the top conrning wall the cylinder, and the
velocity in y-direction is suppressed for the symmetry line y/R = 0. The time step equals t = 0.001 s
for all three constitutive models.
Mesh renement has been investigated for the XPP model using three different meshes, as depicted in
Fig. 3. The mesh characteristics are given in Table 2. Due to the thin stress layer along the cylinder wall,
Fig. 3. Calculated isochromatic fringe stresses in the planar ow around a cylinder along the cylinder wall for the meshes C1,
C2 and C3 using the XPP model at Wi = 2.9. T = 170

C.
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 313
Table 2
Mesh characteristics of the ow around a cylinder
Mesh C1 Mesh C2 Mesh C3
#Elements 920 1830 2,600
#Nodes 3,847 7,547 10,661
#DOF( u, p) 6,858 13,378 18,853
#DOF(

D) 3,012 5,832 8,193
#DOF(S, ) = 4 #elements 20 73,600 146,400 208,000
Smallest element size at front/back stagnation point (10
4
R
2
) 9.0 4.0 1.1
Drag force on cylinder 293,547.5 292,694.8 292,693.8
this is a numerically difcult region. Therefore, to be condent that mesh convergence has been achieved,
the isochromatic fringe stresses along the cylinder wall for the three meshes are plotted in Fig. 3. Taking
the nest mesh C3 as a reference, the largest relative difference of the fringe stress for mesh C1 is 3.31%,
while for mesh C2 this equals 0.38%. Furthermore, the drag force on the cylinder is calculated given in
Table 2. The relative difference between the drag forces is 0.29% in case of mesh C1 and 0.00034% for
mesh C2. We feel that results for mesh C2 are sufciently converged with respect to mesh renement.
For convenience, mesh C2 is used for further investigations.
Withthe current parameter setting, the PTT-a model encounteredconvergence problems andsteady-state
could not be reached. Computations for this model diverged after t = 9.1 s, which is less than twice the
maximum relaxation time. Calculations using ner meshes did extend the calculations to larger times,
but could not avoid divergence. By using a less strain hardening parameter set convergence problems
were not encountered within the range of Weissenberg numbers investigated. We therefore suspect that
the high strain hardening behaviour of the material and the associated parameter set, is one of the key
reasons for the convergence problems.
Fig. 4 shows the normalized velocity proles at ve different cross-sections (x/R = 4, 2.5, 1.5,
1.5, 2.5) and the velocities over the centerline (y/R = 0). The measured data points show a slight
asymmetry at cross-sections closest to the cylinder, indicating a minor misplacement of the cylinder. In
general, the predicted velocities are in good agreement with the measured data for the XPP and Giesekus
models. Over the centerline, the models underpredict the measured velocity prole downstream of the
cylinder, although the centerline exit velocity is predicted correctly.
The measured and calculated isochromatic stress patterns for the XPP and Giesekus models are shown
in Fig. 5. Both models predict very similar stress patterns, and especially for the XPP model all details
are predicted correctly in this qualitative picture. The Giesekus model predicts one fringe too many in
the region along the cylinder and in the downstream section due to the overprediction of the rst normal
stress coefcient as shown in Fig. 1f.
Fig. 6 shows a more quantitative comparison of the stresses over the centerline y/R = 0 and along
the line x/R = 0. Fig. 5a indicates that both models accurately predict the fringes over the upstream
centerline. Along the downstream centerline, the models overpredict the experimental data.
The curved wall of the cylinder induces a stress boundary layer that is difcult to resolve numer-
ically. It is also difcult to access experimentally. By using a laser in the optical set-up, fringe tran-
sitions are observable within 0.02 mm of the cylinder, which was not possible using a conventional
mercury lamp as a light source. Nevertheless, it is quite difcult to count the fringes very close to the
314 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
Fig. 4. Calculated and measured velocity proles in the planar ow around a cylinder over ve cross-sections (a)
(x/R = 4, 2.5, 1.5, 1.5, 2.5), and over the centerline (b) (y/R = 0) at Wi = 2.9. T = 170

C.
Fig. 5. Calculated and measured isochromatic fringe patterns of the planar ow around a cylinder for the XPP model (a) and the
Giesekus model (b) at Wi = 2.9. T = 170

C.
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 315
Fig. 6. Calculated and measured stresses in the planar ow around a cylinder over the centerline (y/R = 0) (a) and over the
cross-section (x/R = 0) (b) for the XPP and Giesekus molels at Wi = 2.9. T = 170

C.
cylinder due to a small misalignment of the optical set-up (as was already indicated by the velocity
proles).
Notice, that it was impossible to count fringes up to the downstream stagnation point. Here, the fringes
showvague transitions, probably accounted for by viscous heating or beamdeections due to temperature
effects [26,30]. Peters and Baaijens [26] numerically investigated the inuence of viscous heating in a
ow around a cylinder for a one-mode PTT model with typical polystyrene melt parameters. They found
an increase in temperature at the back of the cylinder and in its wake. Due to this rise in temperature
the viscosity decreased and thus higher velocities and lower stresses were predicted. Already at Wi = 4
a reduction of the stress up to 40% and differences in downstream peak velocities of 8% was found
in their calculations. The temperature dependency of the viscosity of a polystyrene melt, however, is
higher than for a long-chain branched melt. Therefore, the inuence of viscous heating on the viscosity,
velocity prole and stresses decreases for an LDPE melt. However, the presence of temperature effects
is experimentally clearly shown in Fig. 7, which is taken from Schoonen [30]. To come to strong conclu-
sions that the deviations between the calculations and the experimental data is truly due to temperature
effects, more investigations are needed. A full analysis of the light path through a polymer melt with
refractive index gradients should be performed, as is done for an axisymmetric owcell by Harrison et al.
[12].
Along the cross-section at x/R = 0 (see Fig. 5b) the prediction of the XPP model is in good agreement
withthe experimental data for y/R > 1.15. Close tothe cylinder wall the XPPmodel predicts a stress peak,
that is not observed experimentally. The Giesekus model behaves in a similar fashion, but overpredicts
the experimental data. This is expected because in this part of the domain the ow is shear dominated
and for simple shear ows the current parameter set used for the Giesekus model is known to overpredict
normal stress in shear. Like the XPP model, the Giesekus model predicts a stress peak near the cylinder
wall. Because of experimental difculties, like the misalignment of the cylinder, it is impossible to detect
if such a distorted pattern is also present in the experiments. Since the shear rate at the cylinder wall
1015 s
1
, the peaks are not caused by a constitutive instability. Nor can it be accounted for by an
316 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
Fig. 7. Recorded deection of a laser beam in y-direction (
ys
) along the centerline of the planar ow around a cylinder of an
LDPE melt. The recording screen is placed at 1.86 m from the ow cell. The dimensions for the laser beam spot is scaled down
in x-direction by a factor of 20.
insufcient renement of the mesh, but rather as an artifact of the high strain hardening materials, since
it is predicted by both models.
To see if uid elements in this particular complex ow can reach their elongational planar steady state,
a particle is followed along the downstream centerline starting just downstream the rear stagnation point
at x/R = 1.01. The rst normal stress difference as a function of the Lagrangian time for the XPP model
is plotted as the solid curve in Fig. 8. This curve, showing a typical peak value, is the same as plotted in
Fig. 5a, however now as a function of time instead of position. The dashed line is the build-up in stress if
Fig. 8. Calculated rst normal stress difference in the planar ow around a cylinder over the downstream centerline (y/R = 0,
x/R 1.01) as a function of Lagrangian time () and rst normal stress difference at constant maximum strain rate as
exprienced over the downstream centerline (- - -) for the XPP model at Wi = 2.9. T = 170

C.
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 317
a constant strain rate equal to the maximum strain rate as experienced over the downstream centerline is
imposed on the material, so that it is able to reach its steady state value. Fig. 8 shows that for this complex
ow at this particular ow rate the planar elongational steady state stress is not reached.
5.2. Cross-slot ow
The planar cross-slot ow is a complex ow geometry with a stagnation point that is easily accessible
experimentally. Material ows in from two opposite sides, and ows out perpendicular to the inow
again at two opposite sides (Fig. 8a) . Experiments were performed at a temperature of T = 150

C.
Using the given activation energy from Table 1 and the temperature dependent Arrhenius function, linear
parameters can be transformed to this temperature. This results in a viscosity averaged relaxation time for
the material of

= 3.4294 s. Half the height of the channels is h = 2.5 mm and the depth is d = 40 mm,
giving a depth-to-height ratio of 8. The corners are rounded with a radius of h/2 = 1.25 mm. The mean
two-dimensional in- and outow velocities are u = 3.10 mm/s, which results from taking the same pump
ow rate as for the ow around a cylinder. If half the channel height is chosen as a characteristic length,
the ow averaged Weissenberg number results:
Wi = 4.3, (28)
while the averaged Weissenberg number for the longest relaxation time equals Wi = 11.2.
A detail of the FE mesh used for the calculations of all three models is given in Fig. 8b. Only a quarter
of the geometry needs to be analyzed due to symmetry. The stagnation point is located at the origin of the
coordinate system. The inlet is placed at y/h = 10 and the outlet at x/h = 20. The periodic boundary
for the inlet stresses is situated at y/h = 6. At the entrance and exit, a fully developed velocity prole
is prescribed. No slip boundary conditions are imposed on the top conning wall, and the velocity in x-
Fig. 9. Schematic (a) and detail FE mesh (b) of a quarter of the planar cross-slot ow. The origin of the coordinate system
is in the stagnation point. T = 150

C,

= 3.4294 s, u
2D
= 3.10 mm/s, h = 2.5 mm, R = 1.25 mm = h/2, d = 40 mm,
#elements = 1904, #nodes = 7875, #DOF( u, p) = 13976, #DOF(

D) = 6102, #DOF(S, ) = 4 #elements 20 = 152320,
smallest element size located at stagnation point: 7.7 10
4
h
2
.
318 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
Fig. 10. Calculated and measured velocity proles in the planar cross-slot ow over ve cross-sections in y-direction (a)
(y/h = 3.2, 2.5, 2.0, 1.5, 0.5), over six cross-sections in x-direction (b) (x/h = 0.6, 1.0, 1.4, 1.8, 2.4, 3.2), and over the
symmetry lines (c) (y/h = 0, x/h = 0) at Wi = 4.3. T = 150

C.
and y-directions is suppressed for the symmetry lines x/h = 0 and y/h = 0, respectively. The time
step equals t = 0.002 s for all three models. No convergence problems were detected for either of the
constitutive models.
Fig. 9a shows the measured and calculated velocity proles at several cross-sections of the inowchan-
nel (y/h = 3.2, 2.5, 2.0, 1.5, 0.5). All three models show a perfect match with the measurements.
In the outow channel, velocities at different cross-sections (x/h = 0.6, 1.0, 1.4, 1.8, 2.4, 3.2) are
measured and compared to the calculations, as given in Fig. 9b. Again, all three models showan excellent
agreement with measured data. The centerline velocities, in Fig. 9c, show the same tendency. The largest
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 319
Fig. 11. Calculated and measured isochromatic fringe patterns of the planar cross-slot ow for the XPP model (a), the Giesekus
model (b) and the PTT-a model (c) at Wi = 4.3. T = 150

C.
difference can be detected over the downstream centerline, showing the highest overshoot for the PTT-a
model, although the difference is rather small.
The isochromatic light intensities for the experiments and three models are shown in Fig. 11. All three
models show a good agreement with the experiments, except for the fully developed in- and outow
regions. Here, all models overpredict the stresses by almost one fringe, being largest for the Giesekus
320 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
Fig. 12. Calculated and measured stresses in the planar cross-slot ow over the centerline (y/h = 0, x/h = 0) for the XPP,
Giesekus, and PTT-a models at Wi = 4.3. T = 150

C.
model and lowest for the XPP model. This is consistent with results for the rst normal stress coefcient
as given in Fig. 1f.
The centerline stress build-up and relaxation is quantitatively shown in Fig. 12. From this gure, it
is clear that all three models correctly describe the rst part of the stress build-up. The maximum is
overpredicted by all three models, and is largest for the PTT-a model, followed by the Giesekus model.
The stress relaxation for the PTT-a model is a little too fast and therefore under predicts the experimental
data downstream of the stagnation point. The XPP and Giesekus models follow the experimental data
excellently a little away from the stagnation point. It was impossible to distinguish all individual fringes
experimentally very close to the stagnation point due to the high stress gradient. Furthermore, due to the
high extension at the stagnation point, it might be that the stress-optical rule fails in this region. This may
account for the deviation between experiments and calculations.
To see if planar extensional steady state is reached, a particle is tracked along the centerline, starting
at x/h = 1 10
6
, y/h = 5 and passing very closely the stagnation point at x/h 1 10
3
, y/h
110
3
. The stress that is build up during owover the centerline as a function of Lagrangian time for the
XPP model is plotted as a solid curve in Fig. 13. This curve is equal to the curve in Fig. 12. The plateau
corresponds to the stagnation region, where residence time is long. The dashed curve corresponds to
the stress that is build up if the maximum strain rate as experienced over the centerline is constantly
imposed on the material. It is obvious from the gure, that in the stagnation region, stresses do reach
their planar elongational steady state values. The model even predicts a higher peak stress, meaning that
the location of the highest strain rate does not necessarily have to be the location of the highest stress.
From the last graph, we can conclude, that the cross-slot device can be used as a rheological tool and
the planar elongational steady state values can be extracted from this complex ow. It is, however, more
cumbersome to determine experimentally the strain rate that the particle experiences in the stagnation
region. The strain rate could be determined by performing calculations, but logically it depends on the
model that is used. From Fig. 9c, an experimental strain rate can be estimated:
exp
= 1.4 1.5 s
1
. The
XPP, Giesekus and PTT-a models predict
XPP
= 1.53 s
1
,
Gsk
= 1.49 s
1
, and
PTTa
= 1.60 s
1
.
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 321
Fig. 13. Calculated rst normal stress difference in the planar cross-slot owover the centerline (y/h = 0 x/h = 0) as a function
of Lagrangian time () and rst normal stress difference at constant maximum strain rate as experienced over the centerline
(- - -) for the XPP model at Wi = 4.3. T = 150

C.
6. Conclusions
The performance of the eXtended PomPom (XPP) model is investigated in two complex ow geome-
tries, the planar owaround a conned cylinder and through a cross-slot device. The numerical predictions
using the XPP model are compared with experimental results and, where possible, calculations of the
Giesekus and exponential PTT-a models. Key features observed experimentally in these complex ows
are qualitatively, and to a large extent quantitatively, predicted by all three models. This includes the
stress build up due to planar extension and compression at or near the stagnation points followed by a
relatively slow stress relaxation.
The XPP model shows the best overall description of the available rheometrical data and, therefore, the
measured velocity and birefringence data in the two complex ows. This is not trivial, since for this model
parameter identication is based on linear viscoelastic data and transient uniaxial elongational data only,
while material particles witness complicated deformation histories. In particular at and near the stagnation
points particles are subject to planar in stead of uniaxial elongation as in the rheometrical characterization.
From the material particle point of view the inhomogeneous ows are transient, and therefore it is crucial
to capture the transient shear and elongational rheology of the polymer melt accurately.
A number of reasons can be pointed out that are responsible for the discrepancies between the ex-
perimental and numerical data. First of all, the non-linearity parameters of the constitutive models are
essentially tted on the uniaxial rheological data only, in case of the XPP and Giesekus model, or in
combination with steady shear rst normal stress coefcient data for the PTT-a model. Notice that all
data is represented on a logarithmic scale. Experiments to measure the uniaxial elongational properties are
cumbersome and relatively inaccurate. It is difcult to estimate the experimental error, but an inaccuracy
of about 10%would not be surprising. Furthermore, only four modes are used to t the rheological curves
with limited accuracy. It should, however, be realized that the number of relaxation times cannot be raised
arbitrarily, because their identication soon becomes ill-posed. This in particular imposes limits on the
achievable smoothness of the steady uniaxial viscosity curve of the PTT-a model. Within the current
322 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
PTT-a model the relaxation time is an exponential function of the trace of the extra stress tensor. Perhaps
other expressions would improve the behaviour of the model. Third, measurements in the complex ow
geometries contain errors. Clearly, although the geometries are designed to have near two-dimensionality,
three-dimensional end effects are present in the experiments. In case of the depth-to-height aspect ratio
used in the current experiment, about 8, Schoonen [30] and Bogaerds et al. [7] have shown that this may
induce errors of about 6% in the birefringence data. These effects cause the experimental results to
be lower near the walls and higher over the centerline in comparison to two-dimensional calculations.
Other errors are due to a misalignment of the cylinder while evidence exists that temperature effects,
causing viscous heating and beam deections [30], may not be neglected. Finally, of course, the validity
of the stress-optical rule may be questioned. In the present experiments, however, there is no immediate
indication that this rule fails.
Although, both the planar ow around a cylinder conned by two parallel plates and the cross-slot
device impose high strain rates on the material, important differences exist between these two complex
ows. The thin stress boundary layer around the cylinder causes a number of difculties. Experimentally,
it is problematic to detect all stress fringes close to the cylinder, since a small optical misalignment causes
reection of the light and distortion of the fringe pattern. In the wake of the cylinder, heat transport is
relatively low and therefore this region is susceptible for temperature effects which may cause the uid
to be less viscous and resulting in higher velocities, lower stresses and optical distortions, which is not
accounted for in the calculations. The ow around the cylinder is difcult to resolve numerically [8], and
indeed, convergence problems are encountered for the PTT-a model. For higher Weissenberg numbers
than shown in this work, divergence occurred for all three constitutive models. It is, at present, unclear
how to resolve this problem.
The cross-slot device has proven to be a rather friendly ow geometry. No severe experimental or
numerical problems were encountered. Compared to the cylinder ow it is easier to extract birefringence
and velocity data fromthe cross-slot device. Moreover, the cross-slot owis more discriminating, because,
unlike in other ow geometries such as the ow past a cylinder and contraction ows, steady state
(planar) elongation is reached. This is interesting, because it would allow this ow device to be used
as a rheometrical instrument, provided that the stress-optical rule applies. Values extracted from the
stagnation region could be used together with other rheometrical data to determine the non-linearity
parameters more accurately. A signicant advantage of the cross-slot device, compared to many other
elongational devices, is that it generates a conned ow. This is interesting, because less strain hardening,
or even strain softening, materials are difcult or impossible to handle in traditional uniaxial extensional
devices. Based on these consideration, the cross-slot ow geometry is the preferred geometry to test
constitutive models at non-viscometric ow conditions.
Appendix A. The rhological behaviour of the PTT-a model
To illustrate that the PTT-a model is only capable of satisfactory describing the rheological behaviour
of low density polyethylene melts up to a certain point, this model is presented for the well-characterized
Lupolen 1810H LDPE melt [11,15].
Two sets of parameters for the PTT-a model are given in Table 3. As can be clearly seen in Fig. 14,
set 1 correctly predicts the uniaxial elongation. Notice, that for 10
4
10
1
the different modes
can be detected in the steady state uniaxial viscosity plot. This is due to the excessive strain thinning
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 323
Fig. 14. Transient (a) and quasi-steady state (b) uniaxial viscosity
u
, transient (c) and steady state (d) shear viscosity
s
and
transient (e) and steady state (f) rst normal stress coefcient
1
of the PTT-a model with set 1 for Lupolen 1810H melt at
T = 150

C. = 0.0030, 0.0102, 0.0305, 0.103, 0.312, 1.04 s

1
. = 0.001, 0.01, 0.03, 0.1, 0.3, 1, 10 s
1
.
324 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
Fig. 15. Transient (a) and quasi-steady state (b) uniaxial viscosity
u
, transient (c) and steady state (d) shear viscosity
s
and
transient (e) and steady state (f) rst normal stress coefcient
1
of the PTT-a model with set 2 for Lupolen 1810H melt at
T = 150

C. = 0.0030, 0.0102, 0.0305, 0.103, 0.312, 1.04 s

1
. = 0.001, 0.01, 0.03, 0.1, 0.3, 1, 10 s
1
.
W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326 325
Table 3
PTT-a parameter sets for tting of the Lupolen 1810H melt at T = 150

C
i Maxwell parameters PTT-a 1 PTT-a 2
G
0,i
(Pa)
0,i
(s)
i

i

i

i
1 2.1662 10
4
1.0000 10
1
0.150 0.10 0.150 0.060
2 9.9545 10
3
6.3096 10
1
0.080 0.10 0.080 0.030
3 3.7775 10
3
3.9811 10
0
0.040 0.10 0.040 0.020
4 9.6955 10
2
2.5119 10
1
0.025 0.10 0.025 0.015
5 1.1834 10
2
1.5849 10
2
0.007 0.10 0.007 0.005
6 4.1614 10
0
1.0000 10
3
0.010 0.10 0.015 0.010
behaviour of the PTT-a model at high strain rates. The steady state shear viscosity and rst normal stress
coefcients are also predicted correctly. Since the slip parameter is larger than the elongation parameter
, oscillations occur in the start-up shear responses. These oscillations are not seen in the experiments.
For set 2, the parameter is reduced, such that oscillations in start-up shear responses disappear. The
uniaxial elongation is predicted correctly (see Fig. 15). Unfortunately, shear responses are overpredicted
again.
For such a strain hardening melt, the PTT-a model is unable to predict all features correctly. If elongation
is predicted correctly, shear is overpredicted, or oscillations occur in the start-up shear behaviour. If shear
is predicted correctly, elongation is underpredicted.
References
[1] M.A. Alves, F.T. Pinho, P.J. Oliveira, Effect of a high-resolution differencing scheme on nite-volume predictions of
viscoelastic ows, J. Non-Newtonian Fluid Mech. 93 (2000) 287314.
[2] F.P.T. Baaijens, S.H.A. Selen, H.P.W. Baaijens, G.W.M. Peters, H.E.H. Meijer, Viscoelastic ow past a conned cylinder
of a LDPE melt, J. Non-Newtonian Fluid Mech. 68 (1997) 173203.
[3] J.P.W. Baaijens, Evaluation of Constitutive Equations for Polymer Melts and Solutions in Complex Flows, Ph.D. Thesis,
Eindhoven University of Technology, Eindhoven, 1994.
[4] C. Braudo, A. Fortin, T. Coupez, Y. Demay, B. Vergnes, J.F. Agassant, A nite element method for computing the ow of
multi-mode viscoelastic uids: comparison with experiments, J. Non-Newtonian Fluid Mech. 75 (1998) 123.
[5] G.B. Bishko, O.G. Harlen, T.C.B. McLeish, T.M. Nicholson, Numerical simulation of the transient owof branched polymer
melts through a planar contraction using the PomPom model, J. Non-Newtonian Fluid Mech. 82 (1999) 255273.
[6] R.J. Blackwell, T.C.B. McLeish, O.G. Harlen, Molecular drag-strain coupling in branched polymer melts, J. Rheol. 44 (1)
(2000) 121136.
[7] A.C.B. Bogaerds, W.M.H. Verbeeten, G.W.M. Peters, F.P.T. Baaijens, 3D Viscoelastic analysis of a polymer solution in a
complex ow, Comp. Meth. Appl. Mech. Eng. 180 (3/4) (1999) 413430.
[8] A.E. Caola, Y.L. Joo, R.C. Armstrong, R.A. Brown, Highly parallel time integration of viscoelastic ows, J. Non-Newtonian
Fluid Mech. 100 (2001) 191216.
[9] M. Doi, S.F. Edwards, The Theory of Polymer Dynamics, Oxford University Press, Oxford, 1986.
[10] R. Gunette, M. Fortin, A new mixed nite element method for computing viscoelastic ows, J. Non-Newtonian Fluid
Mech. 60 (1995) 2752.
[11] P. Hachmann, Multiaxiale Dehnung von Polymerschmelzen, Ph.D. Thesis, Dissertation ETH Zrich No. 11890, 1996.
[12] P. Harrison, L.J.P. Janssen, V.P. Navez, G.W.M. Peters, F.P.T. Baaijens, Birefringence measurements on polymer melts in
an axisymmetric owcell, Rheol. Acta 41 (2002) 114133.
326 W.M.H. Verbeeten et al. / J. Non-Newtonian Fluid Mech. 108 (2002) 301326
[13] W.H. Hartt, D.G. Baird, The conned ow of polyethylene melts past a cylinder in a planar channel, J. Non-Newtonian
Fluid Mech. 65 (1996) 247268.
[14] N.J. Inkson, T.C.B. McLeish, O.G. Harlen, D.J. Grov, Predicting low density polyethylene melt rheology in elongational
and shear ows with PomPom constitutive equations, J. Rheol. 43 (4) (1999) 873896.
[15] M. Kraft, Untersuchungen zur scherinduzierten rheologischen Anisotropie von verschiedenen Polyethylen-Schmelzen,
Ph.D. Thesis, Dissertation ETH Zrich No. 11417, 1996.
[16] P. Lesaint, P.A. Raviart, On a Finite Element Method for Solving the Neutron Transport Equation, Academic Press, New
York, 1974.
[17] T.P. Lodge, Rheology: Principles, Measurements and Applications, Rheo-Optics: Flow Birefringence, 1st ed., VCH
Publishers, New York, 1994, Chapter 9, pp. 379421.
[18] M.T. Martyn, C. Nakason, P.D. Coates, Flow visualisation of polymer melts in abrupt contraction extrusion dies:
quantication of melt recirculation and ow patterns, J. Non-Newtonian Fluid Mech. 91 (2000) 109122.
[19] M.T. Martyn, C. Nakason, P.D. Coates, Measurement of apparent extensional viscosities of polyolen melts from process
contraction ows, J. Non-Newtonian Fluid Mech. 92 (2000) 203226.
[20] M.T. Martyn, C. Nakason, P.D. Coates, Stress measurements for contraction ows of viscoelastic polymer melts, J.
Non-Newtonian Fluid Mech. 91 (2000) 123142.
[21] G.H. McKinley, O. Hassager, The Considere condition and rapid stretching of linear and branched polymer melts, J. Rheol.
43 (5) (1999) 11951212.
[22] T.C.B. McLeish, R.G. Larson, Molecular constitutive equations for a class of branched polymers: the PomPom polymer,
J. Rheol. 42 (1) (1998) 81110.
[23] H.Mnstedt, H.M. Laun, Elongational behaviour of a low density polyethylene melt II, Rheol. Acta 18 (1979) 492504.
[24] E.A.J.F. Peters, M.A. Hulsen, B.H.A.A. van den Brule, Instationary Eulerian viscoelastic ow simulations using time
separable Rivlin-Sawyers constitutive equations, J. Non-Newtonian Fluid Mech. 89 (2000) 209228.
[25] E.A.J.F. Peters, A.P.G. van Heel, M.A. Hulsen, B.H.A.A. van den Brule, Generalization of the deformation eld method to
simulate advanced reptation models in complex ow, J. Rheol. 44 (4) (2000) 811829.
[26] G.W.M. Peters, F.P.T. Baaijens, Modelling of non-isothermal viscoelastic ows, J. Non-Newtonian Fluid Mech. 68 (1997)
205224.
[27] G.W.M. Peters, J.F.M. Schoonen, F.P.T. Baaijens, H.E.H. Meijer, On the performance of enhanced constitutive models for
polymer melts in a cross-slot ow, J. Non-Newtonian Fluid Mech. 82 (1999) 387427.
[28] M. Renardy, Asymptotic structure of the stress eld in ow past a cylinder at high weissenberg number, J. Non-Newtonian
Fluid Mech. 90 (2000) 1323.
[29] M.H. Wagner, H. Bastian, P. Rubio, The molecular stress function model for polydisperse polymer melts with dissipative
convective constraint release, J. Rheol. 45 (6) (2001) 13871412.
[30] J.F.M. Schoonen, Determination of Rheological Constitutive Equations using Complex Flows, Ph.D. Thesis, Eindhoven
University of Technology, Eindhoven, 1998.
[31] J.F.M. Schoonen, F.H.M. Swartjes, G.W.M. Peters, F.P.T. Baaijens, H.E.H. Meijer, A 3D numerical/experimental study on
a stagnation ow of a polyisobuthylene solution, J. Non-Newtonian Fluid Mech. 79 (1998) 529562.
[32] H.A. van der Vorst, Bi-CGSTAB: a fast and smoothly converging variant of Bi-CG for the solution of nonsymmetrical
linear systems, SIAM J. Sci. Stat. Comput. 13 (1992) 631644.
[33] W.M.H. Verbeeten, G.W.M. Peters, F.P.T. Baaijens, Differential constitutive equations for polymer melts: the extended
PomPom model, J. Rheol. 45 (4) (2001) 823844.
[34] P. Wapperom, R. Keunings, Simulation of linear polymer melts in transient complex ow, J. Non-Newtonian Fluid Mech.
95 (2000) 6783.
[35] S.C. Xue, N. Phan-Thien, R.I. Tanner, Fully three-dimensional, time-dependent numerical simulations of newtonian and
viscoelastic swirling ows in a conned cylinder. Part I. Method and steady ows, J. Non-Newtonian Fluid Mech. 87 (1999)
337367.