243

Journal of Non-Newtonian Fluid Mechanics, 36 (1990) 243-254

Elsevier Science Publishers B.V., Amsterdam

GENERALIZED CONSTITUTIVE
LIQUID CRYSTALS
PART 2. NON-HOMOGENEOUS

BRIAN J. EDWARDS , ANTONY

and RONALD G. LARSON 3

EQUATION

FOR POLYMERIC

SYSTEMS

N. BERIS *, MIROSLAV

GRMELA

I Center for Composite Materials and Department of Chemical Engineering,

University of Delaware, Newark, Delaware 19716 (U.S.A.)
* Ecole Polytechnique de Montreal, Mont&al, Qu&bec H3C 3A7 (Canada)
AT&T Bell Laboratories, Murray Hill, New Jersey 07974 (U.S.A.)
(Received

November

27, 1989)

Abstract
The Hamiltonian formulation of equations in continuum mechanics
through Poisson brackets was used in Ref. 1 to develop a constitutive
equation for the stress and the order parameter tensor for a polymeric liquid
crystal. These equations were shown to reduce to the homogeneous Doi
equations as well as to the Leslie-Ericksen-Parodi
(LEP) constitutive equations under small deformations [l]. In this paper, these equations are fitted
against the non-homogeneous Doi equations through the simulation of the
spinodal decomposition of the isotropic state when it is suddenly brought
into a parameter region in which it is thermodynamically unstable. Linear
stability analysis reveals the wavelength of the most unstable fluctuation as
well as its initial growth rate. Results predicted from this theory compare
well with the predictions of Doi for the spinodal decomposition using an
extended molecular rigid-rod theory in terms of the distribution function. .
This completes the development of a generalized constitutive equation for
polymeric liquid crystals initiated in Part 1.
Keywords: Hamiltonian fonnulation; Poisson bracket; polymeric liquid crystals; constitutive equation;
tensorial order parameter

1. Introduction
Recently, the Hamiltonian (Poisson bracket) formulation was used to
produce a more general theory for polymeric liquid crystals [l]. The gener-

244
alized constitutive
equation which is generated in Ref. 1 gives predictions
very close to the Doi theory for homogeneous
flows and reduces to the
equivalent expressions obtained from Leslie-Ericksen-Parodi
(LEP) theory
for small deformations.
In Part 1, the translational
diffusion parameter,
B, was unspecified
so
that the initial model formulation
remained as simple as possible. Now,
however, the issue of non-homogeneous
systems must be addressed. To do
this, we must first specify the dissipation
parameter,
B, which will be
accomplished
by a direct comparison with the extended model of Doi [2,3]
involving the distribution function of polymer molecules. In so doing, we are
able to rewrite his equations in a manner which is consistent with the order
parameter
representation.
A linear stability analysis is then performed
to
test the model predictions in the simple case of spinodal decomposition
of
the isotropic state (in a thermodynamically
unstable concentration
regime)
to the nematic state. Results of this test are analogous to the much more
involved calculations
of Doi [3], involving the distribution
function, and
allow specification of the parameters
b,. Thus, all parameters needed in the
new formulation
are specified here and tested so that they agree with the
presently available molecular theories of liquid crystalline structure. However, previous models have suffered from limited regions of applicability
or
from
the complexity
introduced
into
the formulation
by using a
distribution-function
approach. Our new formulation
in terms of the order
parameter tensor remains simple enough, yet faithful enough to the physics
of the liquid crystalline
systems, to use as a basis for future stability
analyses.
This paper is organized as follows. In Section 2, the parameters
in the
constitutive equation for the orientation order parameter are determined by
a direct comparison
with Dois extended rigid-rod theory. In Section 3,
spinodal decomposition
is simulated through a linear stability analysis of the
constitutive
equation
for the order parameter
tensor. Section 4 briefly
presents our conclusions from this work.
2. Generalized constitutive

equation for polymeric liquid crystals

Here, use is made of a second-order

orientational

parameter,

S, definbd

as

(2-l)
where m is the second-moment

of the distribution
(2.2)

245
where $( r, it, t) is the distribution function of the molecular orientation,
and r, n and t are the position vector, the unit vector of orientation and the
time, respectively. From the definition (2.2), both S and m are symmetric
and
= 0,

(2.3a)

tr(m) = 1.

(2.3b)

tr(S)

The portion of the free energy (Hamiltonian) density dependent upon the
order parameter tensor for the generalized model is
H(S)

= -&J&(S)

dV-

&J&(S)

dV,

(2.4)

where a,, is the homogeneous contribution to the entropy,

WS)

= - L%YSYol + $GG&3SP~

- &J,,S,pSP,SC~

- $a;( S&J*,
(2.5)

and Qe is the inhomogeneous (Frank elastic) contribution,

@e(S)

= - &%/G&u

- %%,s,nSvp,u - &%,S,+%,,.

(2.6)

Following Ref. 1, the evolution equation for the order parameter tensor is

(2.7)
where 6H/6Sa,, the functional derivative of H with respect to Sny, is given
by Dl,
6H
= %J(
mx,

a2Syu- a3( Sy&3, - t~yaS2k3)

+ a4(Sy&3S& - ~s,&?&S~/3) + d%,S~aSa~

- WauJ3.P - b2(+Spu,p,a+ :Spu,p,y- %Spc.s.c)
+b&+&,~

- S&%y,c,~- %dav.r

- &xuS~&3~.s)).

(2.8)

246
Here and in the following, we use Einsteins summation
convention
(i.e.
repeated indices imply summation)
and a comma to denote spatial differentiation. Subsequently,
the term proportional
to b, will be neglected since it
is higher-order
than the other non-homogeneous
terms and does not add
qualitatively
to the physics of the situation. The stress constitutive equation
is

(2.9)
With the exclusion of the B and bi parameters, which control the effects
of spatial inhomogeneities
in the orientation,
all the parameters involved in
the above three expressions have already been determined in Ref. 1 in terms
of the Doi theory for homogeneous
systems of rigid rods. This was done by
comparing
the molecular equations
for the time evolution
of the order
parameter tensor and the free energy with the equivalent expressions provided by eqns. (2.4), (2.5) and (2.7). The two expressions are identical for
spatially homogeneous
distributions
if
(2.10a)
and
al=-

9
l-2 i

u
,
3 1

a3=$U,

ai=

%U,

(2.10b)

provided that the Doi closure relationship

for the fourth-order
average is
used [l]. The term proportional
to a4 has been dropped, since tr(S . S - S. S)
= (1/2)tr(S.
S)2 [4], and since it gives identical contributions
with the aA
term to the free energy. (For a much simpler proof than [4], see Appendix
A.) As shown in [l], the viscous stress of the Doi model can also be
recovered as well, simply by letting Mayfir = ~{~~~in,,,rn~~, and the solvent
stress is obtained when r = 7,.
Thus, the parameters
of the present theory are based on a consistent
averaging of the molecular Doi theory for rigid rods. As such, some of the
molecular details (but hopefully not the important ones) are lost in the final
form of the equations,
(eqns. 2.7-2.10);
however, a major advantage
is
realized. Namely, the molecular distribution
function I,L is eliminated from
the equations
while the thermodynamic
consistency
of the equations
is

247
preserved because of the adherence to the Poisson bracket formalism.
Elimination of 4 = I/.J(
r, n, t)-which
is, in general, a six-dimensional function-makes
the numerical solution of the resulting equations practical for
interesting inhomogeneous problems, such as spinodal decomposition in its
early, intermediate and late stages and flow-induced texture evolution in
liquid crystals. Furthermore, the thermodynamic consistency guaranteed
from the Poisson bracket formalism lowers significantly the risk of observing
aphysical behavior introduced from the averaging of the more detailed
molecular description.
The remaining parameters, B and bj, are determined from the more
recent, inhomogeneous Doi theory for concentrated solutions of rigid rods
[2,3] as follows. The parameter B is determined from Dois translational
diffusivity term in the following paragraphs. The parameters bi are obtained
from the inhomogeneous excluded volume (molecular interaction) effects by
comparing the linear stability analyses of spinodal decomposition from the
isotropic to the nematic phase, as described in the next section.
The translational term in the extended Doi diffusion equation [2,3]
(neglecting the effects of the flow field and the rotational terms, which were
addressed in Ref. 1) can be rewritten as
(2.11)
which also can be identified as
(2.12)
where pA denotes the free energy density, defined from the free energy
expression in terms of the distribution function [2,3,5]
A=

p,G(r,

= ck,T

n, t) dn dV

(In #( r, n, t) + W(r,

n, t))$(r,

n, t) dn dV,

(2.13)

where w is the potential of the molecular field expressed in terms of the

interaction potential W( r - r, n, n) [2,3] as
@=

W(r-

r, n, n)#(r,

n, t) dn dV.

(2.14)

In eqns. (2.4), (2.5) and (2.6), the Landau-de Gennes expression for the
free energy is used in lieu of the alternative free energy expression provided
by eqn. (2.13). Furthermore, the parameters a, in eqn. (2.59, listed in eqn.
(2.10), were obtained by fitting the Doi free energy expression [5]. Therefore,

248

in order to compare the translational

term of eqn. (2.7) with eqn. (2.11), we
need to obtain an alternative expression to eqn. (2.11) for the portion of the
free energy depending
on the order parameter
tensor, H(S), which will
involve eqn. (2.4) in terms of an integral over the entire space r 8 n. One
way to obtain this is to assume an approximation
for the free energy of the
form
1 dH

H=Ho+-2-dS:S.

(2.15)

This approximation
becomes exact in the limit of small S when the free
energy reduces to a quadratic functional of S. Thus, the free energy A is
now written as

A=&+

:~[n,n,-',S,,]~#(r,

n,

t) dn dV/,

Ya

(2.16)

where A, involves the portion of the free energy which does not depend
S (kinetic energy). Then, pA can be approximated
as

on

(2.17)
As a consequence,
our formalism as

eqn. (2.11) can be rewritten

in a form compatible

with

(2.18)
If now eqn. (2.18) is multiplied by n,np and integrated
over n, then an
equivalent
expression for the evolution of the order parameter
tensor is
obtained:

where
the
decoupling
approximation
(n,npnin,n,n,)
= (n,n,)
( npng)(n,n,)
is used in order to preserve
the symmetry
of B, and
6,,( ~H/&l$,,) = 0 since the functional derivative is traceless. Equation (2.19)
does not, in general, guarantee the symmetry and unit trace characteristics
of m. However, the corresponding
equations for mpa and tr(m) can be
formulated and used together, as in eqn. (2.7), for the definition of am,,/&.
(We have tried to make the comparison
as simple as possible.) Then,
comparison
of eqn. (2.7) with a modified form of eqn. (2.19) leads to the
following expression for B
B ally& = mq [(h,,-A,)m,,+h,6,,]mgj,

(2.20)

249
where
A,,=---

Dll
2ck,T

D,

(2.21)

___
2ck,T

AL=

Equation (2.20) is positive-definite,

provided that A,, 2 A _L, as is always the
case for polymeric liquid crystals. Thus, this definition is acceptable from a
thermodynamic
point of view (see Refs. 1 and 6).
Although the qualitative
character of the sixth-order
tensor B is well
represented by the above theoretical description, the exact numerical values
of it are not, due to the assumptions
involved in the derivation.
An
alternative approach can be constructed
as follows. In the small concentration limit,
0, and at equilibrium, the distribution
function is constant,
#=$,=1/47r.
F or cases which are perturbed slightly from equilibrium, the
distribution
function can be approximated
by a truncated series in terms of
the order parameter tensor S, which as a first-order approximation
is

w=

#=A exp(PS:nn)=&(l+jB:nn),
with the parameter
S = $(nn

- +a)#

p determined
dn = &S:

or, /3 = H/2. Therefore,

eqn. (2.13) reduces to
A = ck,T
= ck,T
= ck,T

s
/
/

(2.22)
from the consistency

/nnnn

dn = %S,

in this limit, the free energy expression

(2.23)
provided

by

In+ 4 dn dV
nn)-&l+

(-ln(4r)

+ FS:

( - ln(4r)

+ YS : S) dV;

where use was made of the identities

&

requirement

n,np dn = +Saa,

F-S:

nn) dn dV

(2.24)
[7]
(2.25a)

and
(2.25b)

SH
6s = lSck,TS.

(2.26)

250

On the other
reduces to

hand,

in the same small concentration

limit,

eqn. (2.11)

(2.27)
and gives, through multiplication
by n,np - (1/3)~?,~
n, the evolution equation for Sup

and integration

over

which leads, through pre-averaging

of the translational
diffusivity,
to an
expression similar to eqn. (2.7) with the sixth-rank tensor B identified as

BWY&r= &4&,,-

b)m,,+ Ls,c]s,p

(2.30)

Note that the expression for the sixth-rank tensor B obtained through eqn.
(2.30) is very similar to that of eqn. (2.20) in the limit of an isotropic
(S -+ 0) dilute solution of rigid rods, the only difference
being that the
factor 2/15 is replaced by l/9. This small difference is attributed
to the
closure approximation
necessary for the development
of eqn. (2.20) even in
the limit of no intermolecular
interactions.
Either eqn. (2.20) or eqn. (2.30)
can, in principle, be used to describe translational
effects for concentrated
solutions, each one involving specific simplifying assumptions. Which one is
more appropriate
can be seen only after the completion
of more detailed
(numerical) calculations in the non-linear regime. As seen in the following
section, for the linear stability analysis of the spinodal decomposition.
both
formulae for B give practically identical results.
3. Kinetics of spinodal decomposition
As an example of the applicability
of the present model to the study of
non-homogeneous
systems, the initial stages of the spinodal decomposition
are calculated as a liquid crystal in the isotropic state (S = 0) is brought into
a parameter region where the isotropic state is unstable. Recently, Doi [3]
has investigated the same problem by examining the initial behavior of the
fluctuation
of the concentration
and orientation
in the isotropic phase. A
linear stability analysis of a Fourier component
of the fluctuation
led to a
system of coupled differential
equations for the components
of the order
parameter
tensor which revealed three different
fluctuation
modes; the
growth rates and the most unstable wavelength
were subsequently
determined. In this section we show that the same results can be produced
using the present model with much less effort.

251
In the initial stages of the disturbance, it
(u = O), and that the parallel translational
mechanism for the molecules (A = 0, A I =
and retaining only terms linear with respect
parameter tensor, eqn. (2.7) becomes

VDll

= - 27 ( a2Sap
-

is reasonable to assume no flow

diffusion is the only relaxation
0) [3]. Under these assumptions,
to the components
of the order

blKQ3,p.p
- b&&,p,s

+ :Spp,r,,

- f%~%+,~));~,~;
(3 -2)

where we have used

(3.3)
which is the limiting (S = 0) expression for B arising from either of eqns.
(2.20) or (2.30) with the numerical (order one) constant
v assuming the
values of l/2 or 3/5 respectively.
To investigate the time evolution of various modes, let Sk be the kth
Fourier component of the order parameter tensor
Sk = Real[ Ak( t) eikx3],

(3.4

in general complex, where Ak(t) is a traceless, symmetric tensor. Substitution of eqn. (3.4) into eqn. (3.2) leads to a system of five independent,
ordinary differential equations coupling the five independent
components
of
Ak( t). As already observed by Doi [3], these equations can be separated into
five independent
sets of equations, each one governing the (initial) evolution
in time of five orientational
modes. These are equations involving Ak12,
A kll - Ak22, Ak13, Ak23 and Ak33In particular, the equations have exactly the same form discovered by Doi
[3] and separate the fluctuation modes into three types.
(1) The twist
A k22, is
&A

2vDll

k12=

3L2

mode, with similar equations

8bl

-K4
9L2

followed

Ak12,

by A,,,

and A,,, -

(3.5)

length of the molecule.

where K = kL/2 with L being a characteristic
equation (eqn. (3.2) in Ref.
Equation (3.5) is the same as the corresponding

252
3) with the only (minor) difference being that the numerical factor 2v/3
appears instead of 4/7, provided that the coefficient
b, is defined as
h&.

(3.6)

to a
For U < 3 the coefficient of Akr2 in eqn. (3.5) is negative, corresponding
negative eigenvalue, which implies a decaying fluctuation
for every wavelength k. For UP 3 however, the fluctuation
will grow for small enough
wavelengths with the maximum growth rate A,
h

= 6vL+(1 - u/3)*
n2
UL2

attained

for the most

(3.7)

)
unstable

wavenumber

k,,,

k,, = ;

(3.8)

Thus U = 3 corresponds
to the critical concentration
beyond which the
isotropic state becomes unstable to infinitesimal perturbations,
in agreement
with the free energy analysis of the static system.
(2) The bend
$Ah13

mode, with similar equations

= - -2vDll

K*+

3L2

followed

by A,,,

and A,,,,

is

(3.9)

4(269'L:b2)K4]Ak,3.

Equation (3.9) is the same as the corresponding

equation (3.3) in Ref. 3 with
the only (minor) difference that the numerical factor 2v/3 appears instead
of 12/7 provided that the coefficient h, is defined as
(3.10)
(3) The splay
and Ak22, is
$A,;;=

--

mode, with the following

2vDll

3L2 [i

coupled

equations

for AA33, A,,,

1h33?
K4A
f%1 w 8bI

8(3h, + 2&)
27L2
-K4
9L2

A,,,

K4

- -

A,

3L2

(3.11)

27L2

x33>

(3.12)

2vDll 1
$A,,,= - __
3L2 [i

K4 Am

2vD,,
- 3~2

81,,
~KA,33-

(3.13)

253
This mode has two independent equations (if the traceless condition is
satisfied at all times). There are two eigenvalues, both of them remaining
negative as long as U is less than the critical value of 3. If U > 3, however,
these eigenvalues can become positive for small wavenumbers. These characteristics are also exhibited by the corresponding equations for the splay
mode of Doi [3].
Finally, all the growth rates of the above modes have the same dependence on the wavenumber as that discovered by Doi [3]
(3.14)
In conclusion, this similarity of the predictions of the present model for the
spinodal decomposition with the results from the molecular theory of Doi
provides additional evidence that the essential physics is correctly incorporated in the present formulation.
4. Conclusion
In this part of the paper, we have introduced the dissipation tensor B by a
first-order comparison with the more complex, non-homogeneous theory of
Doi [3]. Using the spinodal decomposition results of Doi [3] as a reference, it
was possible to calculate the parameters b, and b, in terms of molecular
variables. Then the Frank elasticity parameters can be estimated from Dois
molecular theory through their interrelationship with the bi shown in Refs. 1
and 8. Although the non-homogeneous Doi theory, based upon the distribution function, 4, is too complicated for general stability analyses, the new
formulation in terms of the order parameter tensor is simple enough for
application to complex flow studies. This represents a unique opportunity to
study the phenomenon of spinodal decomposition, as well as the physics of
flow-induced texture evolution. Future work will address this subject.
Acknowledgement
The authors (BJE, ANB) would like to acknowledge financial support
provided by NSF through an ERC grant to the Center for Composite
Materials.
References
1 B.J. Edwards, A.N. Beris and M. Grmela, J. Non-Newtonian
51-72.
2 M. Doi, J. Chem. Phys., 88 (1988) 4070-4075.

Fluid Mech., 35 (1990)

254
3 M. Doi, J. Chem. Phys., 88 (1988) 7181-7186.
4 D.C. Wright and N.D. Mermin, Crystalline Liquids: The Blue Phases, Rev. Mod. Phys., 61
(1989) 385-432.
5 M. Doi, J. Polym. Sci., Polym. Phys. Ed., 19 (1981) 229-243.
6 A.N. Beris and B.J. Edwards, J. Rheol., 34 (1990) 55-78.
7 R.B. Bird, C.F. Curtiss, R.C. Armstrong and 0. Hassager, Dynamics of Polymeric Liquids,
Vol. 2, Kinetic Theory, 2nd Ed., John Wiley, New York, NY, 1987.
8 B.J. Edwards and A.N. Beris, J. Rheol., 33 (1989) 1189-1193.

Appendix A
For any matrix

c, the Cayley-Hamilton

c *c - l,c + I$ - I$-l

theorem

states that

(~4.1)

= 0,

where

64.2)

I, = trc,
I,=

(1/2)[(trc)*-

tr(c.c)],

(A.3)

and
I3 = det(c).
Multiplying
tr(c.c.c.c)
Substitution
tr(S.S.S.S)

(A.4)
eqn. (A.l) by c*, and taking the trace yields
-I$r(c.c.c)

+I,tr(c.c)

of S for c, knowing
= (1/2)[tr(S.S)]*.

-I,trc=O.

that trS = 0, immediately

(A-5)
yields
(A@