Available online at www.sciencedirect.

com

Applied Mathematical Modelling 32 (2008) 2753–2767

www.elsevier.com/locate/apm

A molecular based model for polymer viscoelasticity:

Intra- and inter-molecular variability
H.T. Banks *, J.B. Hood 1, N.G. Medhin
Center for Research in Scientific Computation, North Carolina State University, Raleigh, NC 27695-8205, United States

Received 1 February 2005; received in revised form 1 June 2007; accepted 11 September 2007
Available online 29 September 2007

Abstract

We develop dynamic equations for rubber viscoelasticity based on a stick-slip continuum molecular-based model. The
model developed is a continuum tube reptation model in which a chemically cross-linked (CC) system of molecules act as
constraint box per unit volume for a physically constrained (PC) system of molecules. The CC-system carries along the PC-
system during instantaneous step deformations. The subsequent relaxation of the PC-system is determined by the config-
uration of the CC-system, its own configuration and confirmation, and external force fields. Conversely, the deformation
of the PC-system acts as an internal variable affecting the deformations of the constraining CC-system. We model the rela-
tionship between these processes to derive a model of viscoelasticity in rubber deformation. In developing a relaxation pro-
cess for the PC-system, we start from the fact that the PC-system is composed of long molecular chains. The dynamics of
these molecular chains are developed by modelling them as chains of beads connected by springs, which represent inter-
molecular potentials. Various segments of the molecular chains relax at different rates. In addition, variability in relaxation
times across molecular chains is permitted.
Ó 2007 Elsevier Inc. All rights reserved.

Keywords: Molecular chain bead models; Polymer viscoelasticity; Variable and distributed relaxation times

1. Introduction

Various molecular and phenomenological models have been developed to model both small and large
deformations in rubber [1–7]. Many of these efforts involve uniform molecular relaxation times even though
there is substantial experimental evidence [8–12] to suggest that the assumption of uniform relaxation time is
not valid. The focus here is to develop models accounting for the multiple relaxation times observed in exper-
iments dealing with the relaxation of elastomers.
In our effort to account for multiple relaxation times, we integrate and extend ideas from molecular models
of Doi and Edwards [13] and Johnson and Stacer [14] and derive a class of nonlinear distributed parameter

*
Corresponding author. Tel.: +1 919 515 3968; fax: +1 919 515 1636.
E-mail address: htbanks@ncsu.edu (H.T. Banks).
1
Current address: Department of Mathematics, Midwestern State University, Wichita Falls, TX, United States.

0307-904X/$ - see front matter Ó 2007 Elsevier Inc. All rights reserved.
doi:10.1016/j.apm.2007.09.018
2754 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767

systems (partial differential equations) with internal strain dynamics that are related to the pseudo-phenom-
enological models of [4,5]. These latter models have provided good agreement with both quasi-static and
dynamic data for rubber in uniaxial tension and in shear.
The dynamic model for rubber viscoelasticity presented here is based on the continuum version [14] of a
tube reptation model considered in [13]. In [13], step-strain relaxation of polymers is modeled with a constraint
(stick-slip) theory in which PC-molecules deform with CC-molecules during a large step-strain. Then the PC-
molecules contract and creep to return to a lower energy and higher entropy state. As a result, the total energy
density at a constraint strain dissipates in time and a viscoelastic theory results. These models are based on
cross-linking of rubber network theories for rubbers and other polymers [7,15,16]. Here we refine our previous
model [1] by accounting for the presence of multiple relaxation times and including them in the relaxation pro-
cess consistent with the reptation mechanism for the relaxation of the PC-molecule. This is accomplished by a
closer study of the dynamics of the PC-molecules.
The continuum molecular model of rubber viscoelasticity proposed in [14] is based on considering the
chemically cross-linked molecules as providing cells or boxes with entrapped PC-molecular segments. The
model involves placing a unit cell or box at each point of the rubber continuum and deriving subsequent equa-
tions for the associated principal stretches. The model developed here employs this fundamental idea.

2. Variability of relaxation rates across the length of a constrained molecule

We model a typical PC-molecule by a chain of N-beads connected by springs. Let the position of the ith
bead be given by
X
N
Ri ¼ Aik qk ; ð1Þ
k¼1
2
with, for example, Aik  expða2 ji  kjb Þ, where throughout we use the notation f  g to indicate that f be-
haves like or can be approximated by g (see Fig. 1).
Thus, the most dominant contribution to Ri comes from qi . In (1) we have
X
3
qk ¼ qkl el ; ð2Þ
l¼1

where el is a unit vector in the direction of the lth coordinate axis.

Let
S ¼fðm; nÞjm ¼ 1; 2; . . . ; N ; n ¼ 1; 2; 3g;
T ¼f1; 2; . . . ; 3N g

and ~h a bijection from S onto T.

Let
XN
oR‘ oR‘
gðai; bjÞ ¼ ; 1 6 a; b 6 N ; 1 6 i; j 6 3:
‘¼1
oqai oqbj

Fig. 1. Vector representation of nodes in a polymer molecule.

 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767 2755

Define the 3N  3N matrices

Gð~hða; iÞ; ~
hðb; jÞÞ ¼ gðai; bjÞ;
1 ~ ~ jÞÞ:
haibj ¼ G ðhða; iÞ; hðb;

We are interested in the dynamics of the bead-spring chain model. For that we seek the relevant configura-
tional distribution of the chain. We consider a configurational distribution for a chain that is in a fairly general
medium possibly subjected to an external force field.
In the case when the mobility matrix is H nm ¼ 1f Idnm , where I is the identity matrix and f is the friction con-
stant [13], the configurational distribution w of the chain can be shown to obey the equation ([13], p. 79, see
Eq. (3.162))
! !
ow o ow o oU o ðV Þ
¼ fhaibj k B T þ fhaibj  ðV wÞ; ð3Þ
ot oqai oqbj oqai oqbj oqai ai

where k B is the Boltzmann’s constant, T the temperature, and U is the potential that includes inter-molecular
potential as well as possible contributions due to external driving fields. Finally, the medium viscosity is given
by

ðV Þ oRn
V ai ¼ fhaibk  ðH 1 Þnm  ðj  Rm Þ; ð4Þ
oqbk

([13], p. 78), where j is velocity gradient tensor. In (3) and (4) we sum over repeated indices (Einstein
notation).
In the case when the inter-molecular potential is modified by adding a Lennard–Jones type potential and
there is an external electric field, E ¼ ðE1 ; E2 ; E3 Þ, which induces a dipole going from one molecule (bead) to
the next, oriented in the same direction, we have
d
hq i ¼ fk  C lrdk hqdk i þ fl2 Ek hlrbk ðAmþ1;b  Am;b ÞðAmþ1;j  Am;j ÞEm hqjm i
dt lr
X
N 1
þ fl1 Ek hlrbk ðAnþ1;b  An;b Þ þ fhlrbk Anb Anl0 jkm hql0 m i ð5Þ
n¼1
X
N 1 X
 fhlrbk K n ðAnb  Anþ1;b Þ ½ðAn;m  Anþ1;m Þhqmk i:
n¼1 m
R
Here hq‘r i represents the average of q‘r , i.e., hq‘r i ¼ wqlr dq,
N X
X N
C lrdk ¼ hlrbk ðAn1;d  An;d ÞðAn1;b  An;b Þ
n¼2 b¼1

and
6 b 8  6 b 6
K n ¼ 12~ðr Þ R n ð1  ðr Þ R n Þ

is the Lennard–Jones potential, where the constant ~ is the depth of the well and r is the separation at which
b n is a minimum where R
R b n ¼ jRn  Rnþ1 j.
Because we have 2the formulation of our Aik weighted so that its largest value occurs when i ¼ k, e.g.,
b
Aik  expða2 ji  kj Þ, the system presented in Eq. (5) can be thought of as almost diagonal.
Let Uðt; sÞ be the fundamental solution of (5). The time dependent off-diagonal elements of the coefficient
matrix in (5) are assumed to be small quantities compared to the diagonal elements. In fact, we take these
quantities to be zero in subsequent discussions. Since the off-diagonal elements are assumed to be much smal-
ler quantities than the diagonal elements, we have
2756 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767

hqkl ðtÞi  eðtsÞ=skl Ukl ðt; sÞhqkl ðsÞi;

Ukl ðt; t0 ÞUkl ðt0 ; t0 Þ  Ukl ðt; t0 Þ; Ukl ðt; tÞ ¼ 1;
Ukl ðt; t0 Þ P 0; Reðskl Þ > 0:

The parameters skl ; k ¼ 1; . . . ; N ; l ¼ 1; 2; 3 are the relaxation parameters of interest. In the case where we
ignore the hydrodynamic effect (the velocity field in the surrounding medium caused by the motion of one par-
ticle which could effect the motion of the other particles) and time dependent external fields in the dynamics of
the molecular chain, we will have
hqkl ðtÞi  eðtsÞ=skl hqkl ðsÞi;

which amounts to setting

Ukl  1:

3. Extending the continuum model of Johnson and Stacer

We recall that the PC-molecule is modeled by N beads connected by springs, and the position of the ith
bead is given by
X
N
Ri ¼ Aik qk ; ð6Þ
k¼1

or
X
3
Ri ¼ Ril el : ð7Þ
l¼1

We also recall from Section 2 that

hqkl ðtÞi  eðtsÞ=skl Ukl ðt; sÞhqkl ðsÞi: ð8Þ

We note that
XX
hRi ðtÞi ¼ Aik hqkl iel : ð9Þ
l k

We would like our generalization of the continuum model of Johnson and Stacer to reflect the presence of
multiple relaxation parameters skl (see (8)). On the basis of (7) and (8) we expect the extension at a given point
of the PC-molecule to be a superposition of extensions across the length of the PC-molecule.
Let the CC-constraint-tube have length LðtÞ and the trapped PC-molecule PN have length ‘ðtÞ. We may con-
sider ‘ðtÞ as composed of N segments ðak ‘ÞðtÞ, k ¼ 1; . . . ; N , ak P 0, k¼1 ak ¼ 1, where ðak ‘ÞðtÞ relaxes in
the direction of el with relaxation parameter skl . We relate the stretches in the constraining CC-molecule
and the PC-molecule by the linear relationship (for a discussion of nonlinear analogues, see [2])
ðak ‘Þ
Dðak ‘Þ ¼ Dðak LÞ
ðak LÞ
ð10Þ
‘
¼ Dðak LÞ:
L
In what follows we will, without loss of generality, drop the ‘‘l’’ in skl and Ukl , and consider sk 2 R.
In the time interval t0 6 t 6 t1 , we use (8) for segments of the CC-tube to obtain
‘ðt0 Þ
ðak ‘ÞðtÞ ¼ ðak ‘Þðt0 Þ þ Dðak LÞðt0 Þeðtt0 Þ=sk Uk ðt; t0 Þ; k ¼ 1; . . . ; N :
Lðt0 Þ
 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767 2757

Here we follow [14] and consider a sequence of instantaneous step tensile deformations at times fti g in the
constraint tube. We write ‘i for ‘ðti Þ, Li for Lðti Þ, ðak ‘Þi for ðak ‘Þðti Þ, and ðak LÞi for ak Lðti Þ for i ¼ 0; 1; . . . Then
we have
‘0
ðak ‘Þ1 ¼ ðak ‘Þ0 þ Dðak LÞ0 eðt1 t0 Þ=sk Uk ðt1 ; t0 Þ:
L0
Set
L1 ¼ L0 þ DL0 :
In the time interval t1 6 t 6 t2 , we obtain
ðak ‘ÞðtÞ ¼ ðak ‘Þ0 þ ½ðak ‘Þ1 þ Dðak ‘Þ1  ðak ‘Þ0 eðtt1 Þ=sk Uk ðt; t1 Þ
 
‘0 ðt1 t0 Þ=sk
¼ ðak ‘Þ0 þ Dðak LÞ0 e Uk ðt1 ; t0 Þ þ Dðak ‘Þ1 eðtt1 Þ=sk Uk ðt; t1 Þ
L0
‘0 ‘1
¼ ðak ‘Þ0 þ Dðak LÞ0 eðtt0 Þ=sk Uk ðt; t0 Þ þ Dðak LÞ1 eðtt1 Þ=sk Uk ðt; t1 Þ:
L0 L1
In the continuum limit (again see [14]), we arrive at an expression for the length of the kth segment of the PC-
molecule
Z t
‘ðsÞ d
ðak ‘ÞðtÞ ¼ ðak ‘Þ0 þ ðak LÞðsÞeðtsÞ=sk Uk ðs; t0 Þds:
0 LðsÞ ds
Summing from k ¼ 1; . . . ; N and writing the resulting formula in differential form, we obtain for tensile
deformations
! !
d‘ X N
‘ðtÞ dLðtÞ XN
ak XN
ak
¼ ak Uk ðt; t0 Þ  ‘ðtÞ þ ‘0 : ð11Þ
dt k¼1
LðtÞ dt s
k¼1 k
s
k¼1 k

4. Strain energy density

We recall here the discussion on strain energy density and stick-slip model found in [1] for tensile deforma-
tions in the context of segmented PC-molecules.
To use the stick-slip model in continuum simulation of the reptation model of rubber elasticity, one con-
siders a network of cells or boxes in the rubber continuum with sides k1c , k2c , and k3c . The CC-box will have
positive strain energy density W cc ðk1c ; k2c ; k3c Þ for all stretches except when k1c ¼ k2c ¼ k3c ¼ 1. Moreover, a
box for the PC-system with sides parallel to those of the CC-box is defined by sides k1p , k2p , and k3p along with
an energy density W pc ðk1p ; k2p ; k3p Þ.
Stresses are calculated from the strain energy density [17] by determining how the energy density function
changes with respect to changes in the stretches or displacements. In the model considered, the strain energy
density of the rubber continuum is assumed to have the form
W ¼ W cc þ W pc
ð12Þ
¼ W cc ðk1c ; k2c ; k3c Þ þ W pc ðk1p ; k2p ; k3p Þ:
To find the stresses at a generic point of the rubber continuum, one must determine how W changes with re-
spect to stretches of the CC-system. The stretches of the PC-boxes are then treated as internal variables
depending on the stretches of the CC-system.
The Cauchy stress in the principal direction ej is given by
oW
Rj ¼ kjc  P; ð13Þ
okjc
where P is the hydrostatic stress. If we consider (12) with the kip as internal variables depending on the kic , then
(13) becomes
2758 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767

oW cc X3
oW pc okip
Rj ¼ kjc þ kjc  P: ð14Þ
okjc i¼1
okip okjc

In obtaining (11), we have related the instantaneous tensile deformations Dðak ‘Þðtm Þ ¼ Dðak ‘Þm and
Dðak LÞðtm Þ ¼ Dðak LÞm according to the formula
Dðak ‘Þm ðak ‘Þm
¼ : ð15Þ
Dðak LÞm ðak LÞm
Using this observation in the analogy of Dðak ‘Þm , Dðak LÞm with principal stretches kip , kic , respectively, we may
write
okjp kjp
¼ dji : ð16Þ
okic kic
In the case of tensile deformations, choosing j ¼ 1 for the direction of loading and using (12) and (16) in (14),
we have
oW cc oW cc
R2 ¼ R3 ¼ 0 ¼ k2c þ k2p  P: ð17Þ
ok2c ok2p
With the hydrostatic stress determined by (17), the tensile Cauchy stress is thus given by
oW cc oW cc oW pc oW pc
R1 ¼ k1c  k2c þ k1p  k2p : ð18Þ
ok1c ok2c ok1p ok2p

5. A dynamic continuum model

To develop a dynamic continuum model for the analysis of rubber/elastomer undergoing large deforma-
tions, we use specific forms of the strain energy functions W cc , W pc introduced in Section 4.
We will use parameterized expressions (for motivation, see [5]) for W cc andW pc in terms of invariants of the
deformations. In particular, we assume
2 2
W pc ¼ A1 ðI pc;1  3Þ þ A2 ðI pc;2  3Þ þ A3 ðI pc;1  3Þ þ A4 ðI pc;2  3Þ ; Ai P 0; i ¼ 1; 2; 3; 4; ð19Þ
2 2
W cc ¼ B1 ðI cc;1  3Þ þ B2 ðI cc;2  3Þ þ B3 ðI cc;1  3Þ þ B4 ðI cc;2  3Þ ; Bi P 0; i ¼ 1; 2; 3; 4; ð20Þ
where we use the invariants
I pc;1 ¼ k21p þ k22p þ k23p ; ð21Þ
I cc;1 ¼ k21c þ k22c þ k23c ; ð22Þ
I pc;2 ¼ k21p k22p þ k21p k23p þ k22p k23p ; ð23Þ
I cc;2 ¼ k21c k22c þ k21c k23c þ k22c k23c : ð24Þ

We also impose incompressibility conditions k1c k2c k3c ¼ 1, k1p k2p k3p ¼ 1. In what follows, we use the relation-
ships (for the tensile deformation case)
1
k2c ¼ k3c ¼ pffiffiffiffiffiffi ; ð25Þ
k1c
1
k2p ¼ k3p ¼ pffiffiffiffiffiffi ; ð26Þ
k1p
which satisfy these incompressibility conditions. From (18), the engineering stress R ¼ R1 =k1c is given by
oW cc k2c oW cc k1p oW pc k2p oW pc
R¼  þ  : ð27Þ
ok1c k1c ok2c k1c ok1p k1c ok2p
 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767 2759

Thus, to approximate the engineering stress, we set

‘ ¼ 1 þ ox up ; L ¼ 1 þ ox uc ;
 1 ¼ ox up ;  ¼ ox uc :

Then using (27), we make the approximation

Rð; 1 Þ ¼ re ðÞ þ rve ð; 1 Þ;
¼ 6ðB1 þ B2 Þox uc þ 6ðA1 þ A2 Þðox up  ox uc ox up Þ þ    ð28Þ
 6ðB1 þ B2 Þox uc þ 6ðA1 þ A2 Þox up

where we have divided the stress into elastic re and viscoelastic rve components (see [5] for detailed
discussions).
From (11), we have
!
XN
ak 1 þ 1 X N
_ 1 þ 1 ¼ _ ak Uk ðt; t0 Þ
s
k¼1 k
1 þ  k¼1

and we note that

1 þ 1
 ð1 þ 1 Þð1  Þ ¼ 1 þ 1    1 :
1þ
Thus,
!
XN
ak XN
_ 1 þ 1  _ ak Uk ðt; t0 Þ: ð29Þ
s
k¼1 k k¼1

From (29), under the assumption that 1 ðt0 Þ ¼ 0, we have

X
N
1 ðtÞ ¼ ðtÞ ak Uk ðt; t0 Þ
k¼1
2 N  3
Z P ak ð30Þ
t
6
 sk ðtsÞ X 7
 ðsÞos 4e k¼1 ak Uk ðs; t0 Þ5ds:
t0 k

In terms of the deformations ox up , ox uc (30) is written as

X
N
ox up ¼ ox uc ak Uk ðt; t0 Þ
k¼1
2 N  3
Z P ak ð31Þ
t
6
 sk ðtsÞ X 7
 ox uc ðs; xÞos 4e k¼1 ak Uk ðs; t0 Þ5ds:
t0 k

We next consider the longitudinal vibration of a rubber rod with (undeformed) cross-sectional area Ac and
mass density q. Using (28) we write the equation of motion (recall that S ¼ Ac R is the engineering stress resul-
tant – see (2.1) of [5])
qAc o2t uc  Ac ox ½6ðB1 þ B2 Þox uc þ 6ðA1 þ A2 Þox up  ¼ F ; ð32Þ
where F is the applied load (in force/unit length) and this equation must be solved with (31).
In summary, the full nonlinear approximate model we derive for the CC–PC molecular system in tensile
deformations for a rod has the form
2760 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767

qo2t uc  ox Rð; 1 Þ ¼ F =Ac ; ð33Þ

Rð; 1 Þ ¼ 6ðB1 þ B2 Þox uc þ 6ðA1 þ A2 Þðox up  ox uc ox up Þ þ    ð34Þ
!
X N
ak 1 þ 1 X N
_ 1 þ 1 ¼ _ ak Uk ðt; t0 Þ: ð35Þ
s
k¼1 k
1 þ  k¼1

We remark that the system (33)–(35) can be viewed as a member of a class of systems similar to those treated
in [18] and the methods presented there can be used to guarantee well-posedness (existence and uniqueness) of
the corresponding variational forms of this system. Indeed, the models treated in [18] can be properly viewed
as generalizations of (33)–(35) that permit (see [2]) nonlinear instantaneous elastic responses as well as other
refinements.
To this point in our development for models with intra-molecular variability, we have treated a rubber sam-
ple as consisting of identical CC-molecules and identical PC-molecules and have ignored the possibility of any
inter-molecular variation. The CC-molecules, which are forced, are not part of the relaxation phenomena and
have a contribution to the elastic part re of the stress that depends only on the infinitesimal strain . However,
the PC-molecules contribute in a significant way to the relaxation and they may have variable (across the pop-
ulation of molecules) relaxation characteristics. If, for example, one has M distinct classes of PC-molecules
with strains j for the jth class contributing to the viscoelastic stress component, then the stress Rð; 1 Þ in
(33) should be replaced by
Rð; 1 ; . . . ; M Þ ¼ re ðÞ þ rve ð; 1 ; . . . ; M Þ ¼ re ðÞ þ RM
j¼1 nj rj ð; j Þ; ð36Þ
where rj ð; j Þ is the j dependent contribution to the overall stress in the CC-tube and nj is the frequency or
proportion of the jth class in the PC-molecule population.
There are several ways in which the PC-molecules might inherently differ, even if they have similar molec-
ular composition; they might differ in the number of nodes N and in the relaxation parameters themselves. We
introduce additional notation which can be used to more easily formulate this variation.
We allow the values of N, the number of nodes in the molecule, and the parameters a, s and U to vary from
one molecule to another. We further assume that there are j ¼ 1; . . . ; M such types or classes of molecules,
each with its own set of N j nodes and parameters ajk , sjk and Ujk , for k ¼ 1; . . . ; N j . With this in mind, we
can extend the nonlinear model for the CC–PC molecular system by
qo2t uc  ox Rð; 1 ; . . . ; M Þ ¼ F =Ac ;
Rð; 1 ; . . . ; M Þ ¼ re ðÞ þ RM M
j¼1 nj rj ð; j Þ  6ðB1 þ B2 Þ þ 6ðA1 þ A2 ÞRj¼1 nj j ;
!
X Nj Nj
ajk 1 þ j X
_ j þ j j ¼ _ ajk Ujk ðt; t0 Þ:
k¼1 sk
1 þ  k¼1

We can simplify the above system of equations by making a few substitutions. First we define
j
1 XN
ajk
¼
Tj j
k¼1 sk

and next we define

Nj
X
U j
ðt; t0 ; fajk g; fUjk gÞ ¼ ajk Ujk ðt; t0 Þ:
k¼1

Thus, we have the following model, based on these discrete variabilities

qo2t uc  ox Rð; 1 ; . . . ; M Þ ¼ F =Ac ; ð37Þ
Rð; 1 ; . . . ; M Þ  6ðB1 þ B2 Þ þ 6ðA1 þ A2 ÞRM
j¼1 nj j ; ð38Þ
1 1 þ j j
_ j þ j _ U ðt; t0 ; fajk g; fUjk gÞ; ð39Þ
Tj 1 þ 
PM
for j ¼ 1; . . . ; M; where nj P 0 and j¼1 nj ¼ 1:
 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767 2761

The above generalization permits only a finite number of different possibilities of relaxation parameters
(corresponding to the M types of molecules we are considering). We may further generalize our system to
a continuum of types of molecules with arbitrary but finite length by using a probability distribution formu-
lation. To do this, we must first determine the space over which such a distribution would be defined.
We first observe that we would like to be able to vary the number, N j , of nodes in a molecule. These num-
bers are positive integers and we assume there exists an upper bound for all N j dependent on the particular
type of elastomer considered. We shall call this upper bound N  and thus we define the space.
N ¼ fN j N 2 Z and 0 < N 6 N  g

to be the space of all possible chain lengths.

We next need to specify the spaces in which the variable parameters reside. Although we have simplified our
model to be in terms of Tj and Uj , we recall that these new parameters are based on the parameters ajk ,
sjk and Ujk . For each term, we define an admissible set of parameters as follows:
( )
N
X
N
A¼ ~ a ¼ fak gk¼1 jak 2 R; ak P 0 for all k; ak ¼ 1 ;
k¼1

D ¼ f~
s¼ fsk gNk¼1 j sk 2 R; sk P 0 for all kg;
FT ¼ ff jf ðt; t Þf ðt ; t0 Þ ¼ f ðt; t0 Þ; f ðt; tÞ ¼ 1; f ðt; t0 Þ 2 Rþ g;
0 0

~ ¼ fUk gN  jUk 2 FT and Uk 2 C n ; for all kg:

F ¼ fU k¼1

We note that the time dependence of the functions Uk is related to that of the velocity gradient, j, and the
electric/external field, E (see (5)). Thus the functions Uk should all be as smooth as j and E. We may assume
that the velocity gradient and the electric/external field, if present, are at least continuous. Thus we may as-
sume that F is a subset of the C n functions on ½0; 1Þ  ½0; 1Þ for some n P 1.
We define T, which is the space of all possible sets of relaxation parameters, to be

T ¼ fð~
a;~ ~ ~
s; UÞj a 2 A;~ ~ 2 Fg
s 2 D; and U

with subspaces defined to be

TN ¼ fT 2 Tjak ¼ 0; k ¼ N þ 1    N  g:
Finally, we define the combined space, N  T, by
N  T ¼ fðN ; T ÞjN 2 N ; T 2 Tg:

Using this space we are able to represent the parameters for any molecule of admissible length without having
to introduce separate spaces for each possible number of nodes. This can be done by defining a distribution on
N  T, which we will refer to as PðN ; T Þ, for N 2 N and T 2 T, that must have the following condition:
PðN ; T Þ ¼ 0 for ðN ; T Þ such that T 62 TN :
Using this probability, we have the following new equation for the stress:
Z
Rð; PÞ ¼ 6ðB1 þ B2 Þ þ 6ðA1 þ A2 Þ ~ð; N ; T ÞdPðN ; T Þ; ð40Þ
NT

where ðN ; T Þ 2 N  T and ~ satisfies

!
XN
ak 1 þ ~ X N
_~ þ ~ ¼ _ ak Uk ðt; t0 Þ: ð41Þ
k¼1 k
s 1 þ  k¼1

This randomly distributed model incorporates both intra-molecular variability of relaxation parameters and
inter-molecular variability of relaxation parameters and composition of the molecules. It is thus a continuum
extension of the model given in (37)–(39) where there are only discrete possibilities.
2762 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767

6. Qualitative analysis of the model

In this section, we provide a brief analysis of the model derived in the previous section. In particular, we
consider Eq. (32) subject to (31). Similar analysis can be performed on the model given in Eqs. (37)–(39)
and those ideas are summarized at the end of the following analysis.
We set (see (31) and (32))

P 1 ¼ 6ðB1 þ B2 Þ; ð42Þ
P 2 ¼ 6ðA1 þ A2 Þ; ð43Þ
XN
R¼ ak Uk ðt; t0 Þ: ð44Þ
k¼1

In the case when we ignore hydrodynamic effect and external time dependent fields in the dynamics of the
molecular chain (PC-chain), we set Uk ðt; t0 Þ ¼ 1 and hence R ¼ 1 in (44).
In what follows, we set t0 ¼ 0. Thus, from (32) and (31) we have
8 N  9
Z t > P ak >
<  sk ðtsÞ =
F
qo2t uc  ½P 1 þ P 2 o2x uc þ P 2 os e k¼1 o2x uc ðsÞds ¼ : ð45Þ
0 >
: >
; A c

We define
8 N  9
Z > P ak >
t <  sk ðtsÞ =
v ¼ ½P 1 þ P 2 uc  P 2 os e k¼1
uc ðsÞds
0 >
: >
;
  ð46Þ
Z t PN
ak
!
 ðtsÞ X N
sk a k
¼ ½P 1 þ P 2 uc  P 2 e k¼1 uc ðsÞds:
0 k¼1 k
s

Then, setting
XN
ak
c¼ P 1 =ðP 1 þ P 2 Þ; ð47Þ
s
k¼1 k

bðtÞ ¼ ct; ð48Þ

we have
! ! !
X N
ak XN
ak 1
vtt þ vt  c vt þ v  ðP 1 þ P 2 Þvxx
k¼1 k
s s
k¼1 k
q
Z t ! !   ð49Þ
X N
ak F
ðbðtÞbðsÞÞ
þc 2
e vs þ v ds ¼ ðP 1 þ P 2 Þ  c2 ebðtÞ uc ð0Þ :
0 s
k¼1 k
Ac q

Rewriting (49), we find

! ! ! Z t ! !
XN
ak XN
ak 1 XN
ak
vtt þ vt  c vt þ v  ðP 1 þ P 2 Þvxx þ c 2
ecðtsÞ vs þ v ds ¼ f ; ð50Þ
s
k¼1 k
s
k¼1 k
q 0 s
k¼1 k

where
 
F
f ¼ ðP 1 þ P 2 Þ  c2 ect uc ð0Þ :
Ac q
 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767 2763

Next, we consider the standard Sturm–Liouville problem

 y 00  ky ¼ 0;
yðaÞ  h0 y 0 ðaÞ ¼ 0; h0 P 0; ð51Þ
yðbÞ  h1 y 0 ðbÞ ¼ 0; h1 P 0:
For this problem, it is known [19] that there is a sequence of eigenvalues 0 < k1 < k2 <    < kn <    % 1
and a corresponding complete family of orthonormal eigenfunctions u1 ; u2 ; . . . ; un ; . . .
We will explore (50) with an appropriate Sturm–Liouville problem such as (51). Thus, let us consider a solu-
tion of (50) in the form
X
1
vðt; xÞ ¼ vn ðtÞun ðxÞ: ð52Þ
n¼1

Also, expanding the function

X
1
f ðt; xÞ ¼ fn ðtÞun ðxÞ; ð53Þ
n¼1

we formally obtain
! ! !
XN
ak XN
ak 1
€vn þ v_ n  c v_ n þ vn þ ðP 1 þ P 2 Þkn vn
k¼1
s k k¼1
s k q
Z t " ! # ð54Þ
X N
ak
2 cðtsÞ
þc e v_ n þ vn ds ¼ fn :
0 s
k¼1 k

Let
Z " ! #
t XN
ak
2 cðtsÞ
wn ¼ c e v_ n þ vn ds ð55Þ
0 s
k¼1 k

and
8 n
< u1 ðtÞ ¼ vn ðtÞ
>
un2 ðtÞ ¼ v_ n ðtÞ ð56Þ
>
: n
u3 ðtÞ ¼ wn ðtÞ:
Then
2 3 2 n3 2 3
un1 u1 0
d 6 n7 6 n7 6 7
4 u2 5 ¼ An 4 u2 5 þ 4 fn 5; ð57Þ
dt
un3 un3 0
where
2 3
 N  0   1  0
6 7
6 c P ak  1 ðP 1 þ P 2 Þkn PN
6 sk q
c ak
sk
1 7
7
An ¼ 6 k¼1 k¼1 7:
6 N  7
4 P ak 5
c2 sk
c2 c
k¼1

We have
XN
ak 1 1
jfI  An j ¼ f3 þ f2 þ f ðP 1 þ P 2 Þkn þ ðP 1 þ P 2 Þkn c: ð58Þ
s
k¼1 k
q q
2764 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767

The Routh–Hurwitz theorem [20] guarantees that the roots of the polynomial equation
z 3 þ a2 z 2 þ a1 z þ a0 ¼ 0
have negative real parts if the following conditions are met:

(i) a0 > 0,
a a0
(ii) 2 > 0,
1 a1
(iii) a2 > 0.

Thus, the roots of jfI  An j ¼ 0 have negative real parts if

" ! #
1 X ak
ðP 1 þ P 2 Þkn  c > 0;
q k
sk

which is the case. It can further be shown that ReðfÞ < M; M > 0.
To study the stability property of the model presented in (37)–(39), we follow the recipe in Section 6, which
leads to (58). Corresponding to (58), we are led to examine the location of the roots of the system of charac-
teristic polynomials given by the recursive definition
1 ðP 1 þ P 2 Þkn kn P 1
D1 ðfÞ ¼ f3 þ f21
þf þ ;
T q qT1
  M  
1 kn P 2 nMþ1 Y 1
DMþ1 ðfÞ ¼ f þ Mþ1 DM þ fþ j :
T qTMþ1 j¼1 T

We can directly verify that the roots of the polynomials D1 and D2 all have negative real parts. Using the recur-
sion formula we can verify that there exist a positive number M and nonnegative   
P numbers n1 , n2 , n3 ; . . ., such
that, for any ni ; i ¼ 1; 2; 3; . . ., where 0 6 n1 6 n1 , 0 6 n2 6 n2 , 0 6 n3 6 n3 ; . . ., ni ¼ 1, the roots of the poly-
nomials D1 ðfÞ, D2 ðfÞ, D3 ðfÞ, i ¼ 1; 2; 3; . . ., are all less than M.

7. Computational considerations

Calculations with macroscopic dynamical models of the form (37)–(39) (or more generally, (40), (41)) actu-
ally preceded and motivated the work in this paper. In [2,3], we found that data from dynamic experiments
with highly filled elastomers in uniaxial tension required multiple relaxation times and cubic nonlinearities
in the forcing terms for the internal strains in (39). A typical experimental response along with the model

Highly Filled Rubber in Uniaxial Tension

30

25
Force (pounds)

20

15

10

0
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
Time (sec)

Fig. 2. Fit to data for highly filled elastomer in dynamic tension/relaxation for model (dashed line) with two internal variables and
experimental data (solid line).
 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767 2765

fit (after inverse problem calculations to estimate parameters in the model) are depicted in Fig. 2 involving
data in which at least 2 or more relaxations times are required to fit the model to the data with any reasonable
degree of fidelity. These earlier findings prompted the efforts reported on here and in [21,22] to understand the
molecular foundations of such models and in particular the apparent distribution of relaxation times across
the molecules required to describe experimental data for polymers and filled rubber.
To understand and partially validate the underlying Rouse bead-spring formulation of Section 2, we carried
out preliminary calculations (reported in [22]) with the underlying bead model (5) containing the Lennard–

Fig. 3. The Rouse bead model: configuration of a polymer molecular at successive times in presence of external electromagnetic field
E ¼ ð0; sin t; 0Þ.
2766 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767

Stress versus Time Stress versus Time

0.8 0.08
test : t =0.126 s test : t =400 s
1 s 2 s
0.7 Experiment : t =0.126 s
1 s
0.07 Experiment : t =400 s
2 s

0.6 0.06
Stress, Σ z (MPa)

Stress, Σz (MPa)
0.5 0.05

0.4 0.04

0.3 0.03

0.2 0.02

0.1 0.01

0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
Time, t (s) Time, t (s)
Articular Cartilage Experiment 1 Articular Cartilage Experiment 2

Fig. 4. Stress model response (dashed line) with estimated parameters vs. experimental data (solid lines) for articular cartilage with two
different ramp strain inputs.

Jones potential and an external electromagnetic field E. A sample calculation of the motion at successive times
for an N ¼ 15 bead single strand molecule is depicted in Fig. 3 in the presence of an alternating electromag-
netic field EðtÞ ¼ ð0; sin t; 0Þ. For further examples and discussion, see [22].
The stress strain law (28) was subsequently extended ([21,22]) to include a nonlinear version of re in R of
(36) and used with the stick-slip ideas for internal strains of Section 3. Calculations with these models are given
in [21] where inverse problem techniques using quasi-static articular cartilage (a material widely accepted as
viscoelastic in nature [23]) data from [24] are used to estimate specific parameters with several different exper-
imental data sets. These validation studies yielded physically based (with molecular relevance) parameters in
the macroscopic constitutive laws. Typical findings are depicted in Fig. 4 where a uniform N j ¼ N ¼
1000; j ¼ 1; . . . ; M, nodes per molecule structure and M ¼ 10 relaxation times distributed across molecules
were used under the assumption that relaxation times were constant within a given molecular strand. That
is, in these calculations we took sjk ¼ sj ; k ¼ 1; . . . ; N j ¼ N , and ajk ¼ N1 . As detailed in [21], the theoretical mod-
elling formulations developed in this paper can be well-supported with experimental data.

8. Conclusion

We have presented a dynamic model of rubber viscoelasticity based on reptation models. Deformed molec-
ular chains or segments entrapped between cross-linked molecules or molecular chains tend to return to their
original undeformed positions. This is due to the fact that their original positions are positions of lower energy
and higher entropy. However, due to the physical and electrostatic barriers created as a result of new config-
urations and conformations, the entrapped and strained molecules only creep to their original confirmation.
The models presented here adhere to these observations and also account for multiple relaxation times that
have been experimentally observed. The approach proposed here indicates how to treat relaxation parameters
as statistical or uncertain quantities. These models can be related in a direct manner to previously derived
(pseudo-phenomenological) models based on data from quasi-static and dynamic experiments with rubber
rods in uniaxial tension and in shear as detailed in [2,3,25,5]. While the models developed here are based
on linear material relationships between CC-system and PC-system stretches, extension to the nonlinear mod-
els in [2,3] are possible; a beginning in this direction is given in [21]. Finally, the models lend themselves
directly to computational methods for deformations of viscoelastic materials.

Acknowledgement

This research was supported in part by the US Air Force Office of Scientific Research under Grant
AFOSR-FA9550-04-1-0220.
 H.T. Banks et al. / Applied Mathematical Modelling 32 (2008) 2753–2767 2767

References

[1] H.T. Banks, N.G. Medhin, A molecular based dynamic model for viscoelastic responses of rubber in tensile deformation, Commun.
Appl. Nonlinear Anal. 8 (2001) 1–18.
[2] H.T. Banks, N.G. Medhin, G.A. Pinter, Nonlinear reptation in molecular based hysteresis models for polymers, Quart. Appl. Math.
62 (2004) 767–779.
[3] H.T. Banks, N.G. Medhin, G.A. Pinter, Multiscale considerations in modeling of nonlinear elastomers, J. Comput. Methods Eng. Sci.
Mech. 8 (2007) 53–62.
[4] H.T. Banks, G.A. Pinter, L.K. Potter, Existence of unique weak solutions to a dynamic system for nonlinear elastomers with
hysteresis, Differ. Integral Equat. 13 (2000) 1001–1024.
[5] H.T. Banks, G.A. Pinter, L.K. Potter, M.J. Gaitens, L.C. Yanyo, Modeling of nonlinear hysteresis in elastomers under uniaxial
tension, J. Intell. Mater. Syst. Struct. 10 (1999) 116–134.
[6] A.R. Johnson, Modeling and viscoelasticity materials using internal variables, Shock Vib. Digest 31 (1999) 91–100.
[7] E. Riande, R. Diaz-Calleja, M.G. Prolongo, R.M. Masegosa, C. Salom, Polymer Viscoelasticity – Stress and Strain in Practice,
Marcel Dekker Inc., New York, 2000.
[8] R.D. Andrews, Correlation of dynamic and static measurements on rubberlike materials, Ind. Eng. Chem. 44 (1952) 707–715.
[9] J.D. Ferry, E.R. Fitzgerald, L.D. Grandine, M.L. Williams, Temperature dependence of dynamic properties of elastomers: relaxation
distributions, Ind. Eng. Chem. 44 (1952) 703–706.
[10] F. Schwarzl, A.J. Staverman, Higher approximation methods for the relaxation spectrum from static and dynamic measurements of
viscoelastic materials, Appl. Sci. Res. A4 (1953) 127–141.
[11] D. Ter Haar, A phenomenological theory of viscoelastic behavior, Physica 16 (1950) 839–850.
[12] M.L. Williams, J.D. Ferry, Second approximation calculations of mechanical and electrical relaxation and retardation distributions,
J. Poly. Sci. 11 (1953) 169–175.
[13] M. Doi, S.F. Edwards, The Theory of Polymer Dynamics, Oxford, New York, 1986.
[14] A.R. Johnson, R.G. Stacer, Rubber viscoelasticity using the physically constrained system’s stretches as internal variables, Rubber
Chem. Technol. 66 (1993) 567–577.
[15] L.H. Sperling, Introduction to Physical Polymer Science, John Wiley & Sons, 1992.
[16] L.R.G. Treloar, The Physics of Rubber Elasticity, Clarendon, Oxford, 1975.
[17] K.C. Valanis, R.F. Landel, The strain-energy function of a hyperelastic material in terms of the extension ratios, J. Appl. Phys. 38
(1967) 2997–3002.
[18] A.S. Ackleh, H.T. Banks, G.A. Pinter, Well-posedness results for models of elastomers, J. Math. Anal. Appl. 268 (2002) 440–456.
[19] E. Coddington, N. Levinson, Theory of Ordinary Differential Equations, McGraw-Hill, New York, 1955.
[20] F.R. Gantmacher, Matrix Theory, vol. 2, Chelsea Publishing Company, New York, 1959.
[21] H.T. Banks, J.B. Hood, N.G. Medhin, J.R. Samuels, A stick-slip/Rouse hybrid model for viscoelasticity in polymers, Technical
Report CRSC-TR06-26, NCSU, November, 2006; Nonlinear Analysis: Real World Applications (in press).
[22] J.B. Hood, Molecular-based Models for Viscoelasticity of Polymers, Ph.D. Thesis, N.C. State University, Raleigh, NC, August, 2005.
[23] D.P. Fyhrie, J.R. Barone, Polymer dynamics as a mechanistic model for the flow-independent viscoelasticity of cartilage, J. Biomech.
Eng. 125 (2003) 578–584.
[24] C.Y. Huang, V.C. Mow, G.A. Ateshian, The role of flow-independent viscoelasticity in the biphasic tensile and compressive responses
of articular cartilage, J. Biomech. Eng. 123 (2001) 410–417.
[25] H.T. Banks, N.G. Medhin, G.A. Pinter, Modeling of viscoelastic shear: a nonlinear stick-slip formulation, CRSC-TR06-07,
February, 2006, Dyn. Syst. Appl. (in press).