HYBRID PARTICLE-FINITE ELEMENT ELASTODYNAMICS SIMULATIONSOF NEMATIC LIQUID CRYSTAL ELASTOMERS
A dissertation submitted toKent State University in partial
fulfillment of the requirements for thedegree of Doctor of Philosophy
by
Badel L. Mbanga
May 2012
All rights reserved
INFORMATION TO ALL USERSThe quality of this reproduction is dependent on the quality of the copy submitted.
In the unlikely event that the author did not send a complete manuscriptand there are missing pages, these will be noted. Also, if material had to be removed,
a note will indicate the deletion.
All rights reserved. This edition of the work is protected againstunauthorized copying under Title 17, United States Code.
ProQuest LLC.789 East Eisenhower Parkway
P.O. Box 1346Ann Arbor, MI 48106 - 1346
UMI 3510768
Copyright 2012 by ProQuest LLC.
UMI Number: 3510768
Dissertation written by
Badel L. Mbanga
B.S., University of Buca, 200 I
M.A., Dalarna University, 2003
Ph.D .• Kent State University, 2012
Approved by
, Chair, Doctoral Dissertation Committee
, Members, Doctoral Dissertation Committee ----------~~~-------------Dr. Eugene C. Gartland
Dr.~~
Accepted by
~- ~ /\ - (:1.7 ~ ;/ ~--' -!.-;G:....-_(... .... _~~_~-+-__ -+ ____ l-__ ~ _____ • Chair. Department of Chemical Physics Dr~ LianjkhY Ch;6n
'--/ -J
------------------------- • Denn, College of Arts and Science Dr. Timothy Moerland
ii
ii
TABLE OF CONTENTS
LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
Acknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xi
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xii
1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Liquid Crystals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Liquid Crystal Elastomers . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.2.1 Isotropic-Nematic transition in Nematic LCE . . . . . . . . . . . . . 9
1.2.2 Photoexcitation in NLCE . . . . . . . . . . . . . . . . . . . . . . . . 11
1.2.3 Freedericksz transition in NLCE . . . . . . . . . . . . . . . . . . . . 12
1.2.4 Flexoelectric effect in NLCE . . . . . . . . . . . . . . . . . . . . . 14
1.3 Strains, strain energy, rubber elasticity . . . . . . . . . . . . . . . . . . . 15
1.3.1 Strain tensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.3.2 Strain energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
1.3.3 Rubber elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Theory of NLCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
1.4.1 De Gennes phenomenological theory . . . . . . . . . . . . . . . . . 21
1.4.2 Neo-classical theory . . . . . . . . . . . . . . . . . . . . . . . . . . 23
iii
2 FINITE ELEMENT ELASTODYNAMICS SIMULATIONS OF LCE . . . . . 28
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.1.1 Forces calculations . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.2 Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
2.3 Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
2.4 Simulations of Rubbery materials . . . . . . . . . . . . . . . . . . . . . . 37
2.5 Simulations of LCE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2.5.1 Isotropic-Nematic phase transition . . . . . . . . . . . . . . . . . . . 42
2.5.2 Semisoft Elasticity . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3 POLYDOMAIN-MONODOMAIN TRANSITION IN NEMATIC ELASTOMERS 45
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.2 Initial configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.3 Simulation and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
3.4 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4 MODELING THE STRIPE INSTABILITY IN NEMATIC ELASTOMERS . . 62
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.2 Simulations and results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.2.1 Stripes width . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2.2 Threshold for stripe instability . . . . . . . . . . . . . . . . . . . . . 70
4.2.3 Rate of strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
iv
5 MODELING DEVICES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.1 Polarization tuner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.2 Actuators . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.3 Peristaltic pumps . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.4 Self-propelled earthworm . . . . . . . . . . . . . . . . . . . . . . . . . . 79
6 CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82
6.1 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
6.1.1 Role of the Frank-Oseen elastic energy . . . . . . . . . . . . . . . 83
6.1.2 Volumetric change (swelling) of LCE . . . . . . . . . . . . . . . . 85
BIBLIOGRAPHY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87
v
LIST OF FIGURES
1 Some liquid crystal phases . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 5-Cyanobiphenyl, a commercially available liquid crystal molecule . . . . . . 2
3 Hexaazatriphenylene liquid crystal (hat), a discotic liquid crystal forming
molecule. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
4 4-cyano-resorcinol, an example of a bent-core liquid crystal molecule . . . . 3
5 Nematic and columnar phases of discotic liquid crystals. . . . . . . . . . . . 4
6 Distortions of nematic liquid crystals . . . . . . . . . . . . . . . . . . . . . . 6
7 Liquid crystal elastomers interaction with external stimuli. Image courtesy
Dr. B. Ratna . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
8 Main chain and side chain polymers . . . . . . . . . . . . . . . . . . . . . . 8
9 Isotropic nematic transition in nematic elastomers . . . . . . . . . . . . . . . 10
10 Photoexcitation in nematic elastomers. The conformation change of the Azo
dyes drives a shape change of the polymer matrix. . . . . . . . . . . . . . . . 11
11 Freederickzs transition in liquid crystals. . . . . . . . . . . . . . . . . . . . . 13
12 Relative positions of two material points in the reference (dr) and target (dR)
spaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
13 Soft elastic deformation. The director rotates to accommodate the strain. . . . 25
14 Soft vs. semisoft response in a nematic elastomer. . . . . . . . . . . . . . . 27
15 (a) and (b) show a tetrahedron in the unstrained and deformed states respectively 30
vi
16 Beam of rubber twisting. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
17 Potential (black dots) and kinetic (red dots) energy of the elastic beam expe-
riencing torsional deformation. In the absence of dissipation, the energy is
conserved (green dots). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
18 Simulation of the isotropic-nematic transition in a liquid crystal elastomer.
The sample experiences a macroscopic shape change as it is cooled down to
the nematic phase and heated back into the isotropic phase. The sample is
clamped at the top end. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
19 Cartoon of the semisoft response in a NLCE. . . . . . . . . . . . . . . . . . 44
20 A cartoon of a polydomain nematic elastomer shows no correlation between
the orientation of the director in neighboring domains. The blue arrows show
the orientation of the nematic director in the domains. . . . . . . . . . . . . 46
21 Initial configurations with different thermomechanical histories. . . . . . . . 53
22 Simulation studies of a I-PNE. (a) and (b) show the entire strip of nematic
elastomer in the polydomain and monodomain configuration respectively. In
the unstrained state, the sample strongly scatters light, whereas when strained
the sample is sandwiched between crossed polarizers and aligned with either
the polarizer or analyzer, total extinction of the incident light will occur. . . . 54
23 Texture and director configuration in a region near the center of the strip in
the polydomain state of I-PNE. . . . . . . . . . . . . . . . . . . . . . . . . 55
24 Texture and director configuration in a region near the center of the strip in
the monodomain state of I-PNE. . . . . . . . . . . . . . . . . . . . . . . . . 56
vii
25 Texture and director configuration in a region near the center of the strip in
the polydomain state of N-PNE. . . . . . . . . . . . . . . . . . . . . . . . . 57
26 Texture and director configuration in a region near the center of the strip in
the monodomain state of N-PNE. . . . . . . . . . . . . . . . . . . . . . . . 58
27 Engineering stress σ ,(black curve) and Global order parmeter, S (red curve)
vs strain for a nematic elastomer crosslinked in the isotropic phase ( I-PNE) . 59
28 Engineering stress σ ,(black curve) and Global order parmeter, S (red curve)
vs strain for a nematic elastomer crosslinked in the nematic phase ( N-PNE) . 60
29 Experiment: stretching a nematic elastomer film at an angle of 90o to the
director results in a microstructure consisting stripes of alternating director
orientation. Image courtesy H. Finkelmann . . . . . . . . . . . . . . . . . . 63
30 Simulation: stretching a nematic elastomer film at an angle of 90o to the
director. Initially a monodomain, the director field evolves to form a striped
microstructure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
31 Engineering stress (circles) and director rotation (squares) vs applied strain,
for the system shown in Figure 30. Onset of director rotation and the stress-
strain plateau both occur at the same strain. . . . . . . . . . . . . . . . . . . 70
32 Engineering stress (circles) and director rotation (squares) vs applied strain,
applied at an angle of 60o from the nematic director. . . . . . . . . . . . . . 71
33 Simulation: stretching a nematic elastomer film at an angle of 60o to the
director. Initially a monodomain, the director field rotates smoothly without
sharp gradients in orientation. . . . . . . . . . . . . . . . . . . . . . . . . . 72
viii
34 Dependence of the stress-strain response on strain rate. . . . . . . . . . . . . 73
35 Simulation: A nematic elastomer disk is stretched radially. The director field
smoothly transforms from a homogeneous monodomain to a radial configu-
ration. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
36 Cartoon of the polarization modulation by the elastomer strip. . . . . . . . . 76
37 Simulation of the photo-deformation of a beam of nematic liquid crystal elas-
tomer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
38 Simulation of the photobending of a strip of polydomain nematic liquid crys-
tal elastomer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
39 Simulation of a soft peristalsis tube made of nematic liquid crystal elastomer. 79
40 Simulation of a nematic elastomer robotic earthworm moving on a rugged
surface shaped like a size wave, with height z = A sin(kx). . . . . . . . . . . 81
ix
Acknowledgements
I thank Professors Jonathan Selinger, Mark Warner, Eugene Terentjev, T.C. Lubensky,
E. C. Gartland, and Antonio Desimone for very fruitful discussions. Dr. Fangfu Ye and
Ms. Vianney Gimenez-Pinto have greatly contributed to the work that is reported in this
dissertation. Finally, I am profoundly grateful to my advisor Professor Robin L. B. Selinger
for her guidance, support and mentoring.
x
To my family
xi
CHAPTER 1
INTRODUCTION
1.1 Liquid Crystals
In this chapter we review fundamental background information from the literature which
serves as a basis for the work that follows. Since their discovery at the end of the nineteen
century [1, 2], liquid crystals have generated a widespread interest from physical scientists
involved in areas such as soft condensed matter, cosmology [3, 4] and biological physics due
to their remarkable properties and potential applications. The liquid crystalline state is a par-
tially ordered state of matter that can be observed in compound materials with anisotropic
molecules or aggregates. This phase can itself be subdivided into several mesophases de-
pending on the type of broken symmetry. A simplistic view of the accepted nomenclature of
the liquid crystalline mesophases is shown in Figure 1.
Besides the anisotropy usually observed in their physical properties (electrical, magnetic,
optical), liquid-crystal-forming molecules share a common characteristic, which is the shape
anisotropy of the molecules. In particular, elongated molecules (rod-like) are known to form
liquid crystalline phases. A common example is 5CB, shown in Figure 2. One can also obtain
a liquid crystal phase from discotic (disk-like) molecules [5] or bent-core (also referred to as
banana) molecules. Figures 3 and 4 show examples of discotic and bent-core liquid crystals
forming molecules, respectively.
The transition between two mesophases can be induced either by a temperature change
1
2
Figure 1: Some liquid crystal phases
Figure 2: 5-Cyanobiphenyl, a commercially available liquid crystal molecule
or by a concentration change if the material is in the presence of a solvent. Materials that
fall in the former category are known as thermotropic liquid crystals, and those in the latter
are labelled lyotropic liquid crystals. There exist materials that respond to both temperature
and concentration variations. Those are known as amphotropic liquid crystals. In the works
reported here, any reference to a liquid crystal will implicitly assume a thermotropic material
unless otherwise stated.
The smectic phase is characterized by a spontaneously broken translational symmetry
along one direction as shown in Figure 1. It is best thought of as being a fluid in two dimen-
sions and a solid in the third. X-ray scattering experiments suggest a layered-like structure in
smectic materials. There exist several mesophases of smectic materials. The most studied are
3
Figure 3: Hexaazatriphenylene liquid crystal (hat), a discotic liquid crystal forming molecule.
Figure 4: 4-cyano-resorcinol, an example of a bent-core liquid crystal molecule
the smectic A and the smectic C phase. They differ in the molecular orientation relative to
the layer normal. In the smectic A phase, the molecules are on average parallel to the normal
to the layers, whereas in the smectic C phase, they are tilted with respect to the layer normal,
as illustrated by the cartoon in Figure 1.
The nematic phase is the most well studied, as it is used in most applications. It is char-
acterized by its long range orientational order and absence of translational order. For rod-like
nematogens, this means that the long axes of the molecules tend to align parallel to one an-
other. A snapshot of the centers of masses of the molecules will show no correlation, just
as in the isotropic phase. On the other hand, discotic liquid crystals will have the axes of
the disk-like molecules more or less aligned with one another in the nematic phase as in Fig-
ure 5(a). The direction of average orientation of the molecules is called the director n(r). n
4
is invariant under inversion symmetry, meaning that it is a headless vector, or defined such
that n ≡ −n . Under polarizing optical microscopy, a sample of liquid crystal in the nematic
phase exhibits a birefringence conferred to it by the anisotropy of its molecules.
(a) Nematic phase of a dis-cotic liquid crystal
(b) Columnar phaseof a discotic liquidcrystal
Figure 5: Nematic and columnar phases of discotic liquid crystals.
The transition between the isotropic and the nematic phases can be described with the
aid of an order parameter. This quantity is carefully chosen to be nonzero in the nematic
phase and zero in the isotropic phase. Moreover, it obeys the symmetry of both phases. In
the particular case of a uniaxial nematic phase, the nematic order parameter tensor Qij =
S(ninj −1
3δij) is a suitable quantity for characterizing the transition. n is the local nematic
director. S =1
2< 3vini − 1 > is the scalar order parameter order parameter. Here v is
along the long molecular axis, and one notes that S is just the average of the second Legendre
polynomial, S =
⟨1
2(3 cos2 θ − 1)
⟩, where θ = n · v. It is immediate from what precedes
that S = 0 when the nematogens are randomly oriented in space as is the case in the isotropic
5
phase, whereas S = 1 when they are perfectly aligned with one another. The Landau-de
Gennes theory for the Isotropic to nematic phase transition is built around a power series
expansion in Qij
E =1
2a(1− Tc
T)QijQij +
1
3bQijQjkQki +
1
4c[QijQij]
2 (1)
While global transformations, that is, translations and uniform rotations of the material
should cost no elastic energy, gradients in the orientation of the nematic director are penalized.
Figure 6 shows an illustration of the three types of deformations of the director field that are
penalized, namely the splay, twist, and bend deformations. Those spatial variations in the
orientation of the nematic director are penalized by the Frank-Oseen free energy:
F =1
2K11(∇ · n)2 + 1
2K22(n · ∇ × n)2 +
1
2K33(n×∇× n)2 (2)
Here K11, K22, and K33 are the elastic constants corresponding to the splay, twist and
bend deformations respectively. Note that additional terms associated with boundaries are
ignored in the equation above.
What precedes was just a brief survey of certain liquid crystals properties that are pertinent
to this dissertation. References [5–9] provide excellent reviews about the physics of liquid
crystals.
6
Figure 6: Distortions of nematic liquid crystals
1.2 Liquid Crystal Elastomers
Here we summarize findings from several research groups that have helped contribute to
a better understanding of the behavior and properties of liquid crystal elastomers. Emphasis
is put mainly on the information that is most relevant to the substance of the work reported in
this dissertation.
Liquid Crystal Elastomers (LCE) are materials that exhibit some of the elastic properties
of rubber along with the orientational order properties of liquid crystals. They are composed
of liquid crystal mesogens covalently bonded to a weakly cross-linked polymer backbone
[10,11]. Similar to low molecular weight liquid crystals, LCE respond to external stimuli such
as a temperature change, applied electric or magnetic fields, or mechanical stress [10, 12].
7
These materials display strong coupling between orientational order of the mesogens and
mechanical deformation of the polymer network. For instance in a nematic LCE, any change
in the magnitude of the nematic order parameter can induce shape change, e.g. the isotropic
nematic phase transition induces strains of up to several hundred percent in a strip of nematic
elastomer [12]. Thus LCE have been proposed for use as artificial muscles or soft actuators.
The cartoon in Figure 7 illustrates the stimulus-response of liquid crystal elastomers. Con-
versely, applied strain can also drive changes in orientational order, producing the fascinating
phenomenon of semisoft elasticity [13].
Figure 7: Liquid crystal elastomers interaction with external stimuli. Image courtesy Dr. B.Ratna
8
LCE can exist in the main chain or side chain configurations. The former has the meso-
gens linked together within the polymer backbone as in Figure 8(a), whereas in the latter,
the liquid crystal molecules are pendant and attached to the polymer backbone by flexible
spacers as shown in Figure 8(b). Note that discotic liquid crystal elastomers have also been
reported [14], with properties similar to LCE made with rod-like nematogens.
(a) main chain
(b) side chain
Figure 8: Main chain and side chain polymers
The first few attempts to blend polymers and liquid crystals date as far back as the
1960’s [10]. The effort then was mostly directed towards obtaining polymer networks with
a certain amount of frozen-in long range order. The approach was thus to crosslink poly-
mer networks in the presence of a liquid crystalline solvent in the nematic phase. Several
groups successfully obtained such materials, but it was only in 1981 that Finkelmann re-
ported the first liquid crystal elastomer formed with nematogens attached to the backbone
9
with crosslinkers [15]. Such materials had been predicted theoretically by De Gennes less
than a decade earlier in a seminal paper [16]. The Finkelmann experiment reported liquid
crystal elastomers in the smectic, cholesteric, and nematic phases. These were side-chain
LCE with a polysiloxane polymer backbone. Other approaches have been used for making
the polymer backbones; in particular, acrylate polymer backbones have been reported [?].
Polysiloxane however remains the backbone of choice due to its high anisometry, which has
been observed to display the most dramatic shape change in experiments [17] .
1.2.1 Isotropic-Nematic transition in Nematic LCE
Nematic elastomer materials constitute the main focus of this dissertation. A spontaneous
shape change as depicted in the cartoon of Figure 9 is usually observed in nematic elastomers
undergoing the isotropic to nematic phase transition. The mechanism is a simple one whereby
the polymer network is distorted and tends to depart from the average spherical shape, elon-
gating along the direction of the spontaneous director that arises upon cooling. In other words,
the radius of gyration tensor of the polymer chains acquire the anisotropy of the liquid crystal.
Conventional liquid crystals undergo a nematic to isotropic phase transition characterized by
a sharp discontinuity in the nematic order parameter and in the materials properties. For ex-
ample, a plot of the birefringence as a function of temperature will show a sharp change at the
transition. In the Ehrenfest classification of phase transitions, this is a first order transition.
Surprisingly, nematic elastomers show a totally different behavior near the transition. Exper-
iments have reported a transition that is neither first order nor second order, albeit smooth. It
is more like a smooth crossover between the two phases. Selinger et al. [18] have shown that
there are several mechanisms that could account for that peculiar behavior. First, they showed
10
that the underlying heterogeinity of elastomers, that is, the quenched disorder of which the
crosslink points are the sources, results in regions of different isotropic-nematic transition
temperatures. This means that the smooth transition observed is due to the fact that domains
are continuously undergoing the transition, much like an avalanche. The strain also varies like
other material properties during the transition. The other possible cause for the broadening
of the transition proposed by Selinger et al. is the non-uniform distribution of local stresses
as a result of crosslinking. We make a similar observation in our studies of the polydomain to
monodomain transition in nematic elastomers in Chapter 3. De Gennes proposed that mon-
odomain samples produced by means of the Finkelmann’s two-step crosslinking method are
paranematic when heated up above the isotropic nematic transition. That is, they have a resid-
ual anisotropy acquired during the second stage of crosslinking, and hence no real isotropic
nematic transition could be expected.
Figure 9: Isotropic nematic transition in nematic elastomers
11
1.2.2 Photoexcitation in NLCE
Experiments on LCE have demonstrated that these materials show a coupling between
optical and mechanical energy [19, 20] . A beam of light incident on a sample of liquid crys-
tal elastomer was reported to induce mechanical deformations comparable to those attained
at the N-I transition. This is a direct modulation of the degree of nematic order in the material
by light, followed by a response of the polymer matrix due to the coupling between order
and strain in LCE. Warner and Terentjev [10] have coined this phenomenon “stress-optical
coupling.” In order to observe stress-optical coupling, there must be another component be-
sides the liquid crystal molecules and the polymer network that responds to illumination,
as depicted in Figure 10. This is usually achieved by embedding a low concentration of
light-sensitive molecules (Azo dyes) in the blend [21, 22].
Figure 10: Photoexcitation in nematic elastomers. The conformation change of the Azo dyesdrives a shape change of the polymer matrix.
These molecules can exist in either the Trans (elongated) or the Cis (kinked) configuration
and undergo a conformational change from Trans to Cis when illuminated at the appropriate
12
wavelength. The Cis configuration is usually a metastable state and hence the molecules tend
to relax from Cis to Trans when the stimulus is removed.
1.2.3 Freedericksz transition in NLCE
A sample of nematic liquid crystal confined between two parallel substrates can be aligned
by an external electric or magnetic field. In a typical experiment with an electric field, the
director at the surface of the substrates is anchored and has a preferred orientation dictated
by an alignment layer. An electric field is applied perpendicular to the substrates, and the
director tends to align parallel (resp. perpendicular) to the field if the liquid crystal molecules
have a positive (resp. negative) dielectric anisotropy. Figure 11 shows a typical aligned
sample before and after application of a field.
In the one-elastic-constant approximation, one can rewrite the Frank-Oseen elastic energy
density as:
Ffrank =1
2K(∇n)2
≈ 1
2K(
π
d)2 (3)
Here K is the elastic constant and n = n(r) is the nematic director; d is the cell gap. The
total elastic energy in the cell is thus Ffrank =1
2K(
1
d)2 .
The contribution of the electrostatic energy can be expressed in the form:
Felectric = −1
2ϵo∆ϵE2d (4)
where ∆ϵ is the dielectric anisotropy, E is the applied electric field, and ϵo is the dielectric
13
permittivity of free space.
The electric energy will overcome the elastic energy when the field reaches the critical
value:
Ec =π
d
√K
ϵo∆ϵ(5)
This corresponds to a voltage Vc = π
√K
ϵo∆ϵwhich is independent of the sample thickness,
but just depends on the material parameters. The onset of re-orientation of the molecules
occurs at a critical voltage. This phenomenon is known as the Freedericksz transition and
is used in liquid crystal display applications to modulate light as it passes through the liquid
crystal cell. Similar experiments carried out on nematic elastomers have failed to show such
(a) With the eld OFF (b) With the eld ON
E
Figure 11: Freederickzs transition in liquid crystals.
a dramatic characteristic behavior unless the applied field was very large.
Using a similar argument to the aforementioned for low molecular weight liquid crystals,
and taking into account that the elastic energy scale is of the order of the shear modulus
µ, and that the contribution of the Frank-Oseen elastic energy is negligibly small, one can
approximate the electric field required to switch the orientation of the nematogens as:
Ec =
õ
ϵo∆ϵ(6)
14
From this it follows that nematic elastomers respond to a critical field and not to a critical
voltage [23–26] . This is intuitive when one considers that contrary to low molecular weight
liquid crystal where the anchoring occurs at the surfaces, in nematic elastomers, the director
is anchored throughout the bulk of the sample, hence offering more resistance to macroscopic
deformation. Note however that even small director reorientations have been shown to drive
macroscopic shape changes in unconfined samples [27].
1.2.4 Flexoelectric effect in NLCE
The flexoelectric effect is the induction of a spontaneous polarization as a response to a
mechanical deformation. Although very much pronounced in polar dielectric materials, it
does not require the material’s molecules to have a nonzero permanent dipole moment. In-
deed, the flexoelectric effect can be observed in liquid crystals of apolar molecules subjected
to splay or bend deformations [28]. The induced polarization is of the form [28]:
Pf = e1n(∇ · n) + e3n(∇× n) (7)
A direct method of measuring the flexoelectric coefficients was recently introduced by
Harden et al. [29], and allowed them to measure the flexoelectric response of bent-core ne-
matic liquid crystals. Bent-core nematic liquid crystals are special because they have a strong
permanent dipole moment conferred to them by their shape anisotropy. They found bend
flexoelectric coefficients (e3 ≈ 50nC/m) to be three orders of magnitude larger than those
obtained from estimation [30, 31] and measurements with calamitic liquid crystals such as
15
5CB [32, 33]. Simulations by Dhakal and Selinger [34] have found similar results. Subse-
quent experiments have also approached the converse flexoelectric effect, which is charac-
terized by a shape change under the application of a current. Note that the changes in shape
observed here are orders of magnitude larger than in piezoelectric devices. This motivated
experiments on bent-core nematic elastomers by several groups. Harden et al. applied a peri-
odic mechanical deformation to a thin film of bent-core nematic liquid crystal using a small
speaker-driven motor. The induced electric current they obtained with small to moderate de-
formations was in the nanoampere range. Chambers et al. [35] also obtained similar results
using a calamitic nematic liquid crystal elastomer film swollen in a solution of bent-core ne-
matic liquid crystal. Observe that both experiments yielded bend flexoelectric coefficients
comparable to that of the low molecular weight bent core nematic. These experiments rein-
force the view that nematic liquid crystal elastomers are excellent candidates for engineering
device applications that strive to convert electrical to mechanical energy, and vice-versa.
1.3 Strains, strain energy, rubber elasticity
Before delving into the details of the method we will introduce for modelling nematic
elastomers, it is useful to quickly review some concepts that are used throughout this disser-
tation.
1.3.1 Strain tensor
Elasticity is a material property observed when the forces causing deformation remain
below a certain threshold, thus allowing the material to return to its (relaxed) state before
deformation upon removal of the forces. We will refer to the space in which the material exists
16
prior to deformation as the reference space, and that in which it is found when deformed as the
target space. Consider a material point initially at a position r in the reference space as shown
in Figure 12. After a deformation of the material, the position of the material point in the
target space can be found by a mapping R(r). In order to express how neighbouring material
points are displaced with respect to one another, we define a quantity λij =∂Ri
∂rjreferred to
as the deformation gradient tensor. It can be shown that the difference in the square of the
Euclidean distances between two such neighbouring points due to the deformation is
dR2 − dr2 = (λikdrk)(λildrl)− (δikdrk)(δildrl)
= (λikλil − δikδil)drkdrl
= (λikλil − δkl)drkdrl
One can express the position of a material point in the target space as the sum of its position
in the reference space and a displacement field: R(r) = r + u(r). Note that a uniform
displacement field simply corresponds to moving the whole body, and should cost no elastic
energy. In terms of the displacement field, one can express the deformation gradient tensor
as:
λij =∂Ri
∂rj
=∂ri∂rj
+∂ui
∂rj
= δij +∂ui
∂rj
17
The aforementioned square of the change in separation between neighbouring material points
can thus be expressed as
dR2 − dr2 = (λikλil − δkl)drkdrl
=
[(δik +
∂ui
∂rk)(δil +
∂ui
∂rl)− δkl
]drkdrl
= (∂ul
∂rk+
∂uk
∂rl+
∂ui
∂rk
∂ui
∂rl)drkdrl
= 2εkldrkdrl
The quantity εkl =1
2(λikλil − δkl) is the Green-Lagrange strain tensor which is invariant
under rotations in the target frame.
Figure 12: Relative positions of two material points in the reference (dr) and target (dR)spaces.
1.3.2 Strain energy
The energy cost of deforming an elastic material can be described in several forms, all
using strain tensors and elastic moduli as main ingredients. Perhaps the most familiar form
of strain energy is that proposed by Hooke, which states that the strain energy is simply a
18
quadratic function of the strain tensor.
U =1
2Cijklεijεkl
Here εij is the Green-Lagrange strain tensor, Cijkl represents the elastic constants asso-
ciated with the material. The stiffness tensor Cijkl is a function of the shear modulus and
Poisson ratio of the material.
We note that since the Green-Lagrange strain εij is defined in the material or body frame,
we must also define Cijkl in the body frame. Thus if the sample undergoes a rotation, these
quantities conveniently rotate with the body rather than being defined in the lab frame.
Other forms of strain energies have been proposed to describe deformations of rubbery
materials, in particular the Neo-Hookean strain and the Mooney-Rivlin strain energies. These
are mainly constructed by creating a series expansion in terms of the invariants of the strain
tensor. Polar decomposition allows one to write the deformation gradient tensor as the product
of a positive-semidefinite stretching (F ) and a unitary rotation (R) tensors such that λ =
RθF . The stretching tensor is F =√λTλ, and the rotation is Rθ = λF−1. A deformation
such that F = δij is a mere rotation of the body and costs no elastic energy. In its eigenframe,
the stretching tensor is expressed as
F =√λTλ =
λ1 0 0
0 λ2 0
0 0 λ3
(8)
19
such that
C = F 2 = λTλ =
λ21 0 0
0 λ22 0
0 0 λ23
; ε =
λ21 − 1 0 0
0 λ22 − 1 0
0 0 λ23 − 1
(9)
are defined in the same eigenframe. Here C is known as the first Cauchy-Green tensor; in
terms of the components of the diagonalized stretching tensor (λ1, λ2, λ3), its invariants are :
I1(C) = λ21 + λ2
2 + λ23
I2(C) = λ21λ
22 + λ2
1λ23 + λ2
2λ23
J(C) = λ1λ2λ3
The neo-Hookean strain energy density for an incompressible material is expressed as :
Uneo−H =µ
2(I1 − 3) + Co(J − 1)2
Note that for an incompressible material, the third invariant (J) is unity, thus making the
second term in the expression above vanish. However, rubbery materials do not really deform
at constant volume, and this is taken into account by the Mooney-Rivlin strain energy density,
which allows for deviations from the volume before deformation.
UMR =1
2Cm1(I1 − 3) +
1
2Cm2(I2 − 3)
where I1 = I1J−2/3 and I2 = I2J
−4/3. The coefficients Cm1 and Cm2 are related to the shear
and bulk moduli of the materials.
20
1.3.3 Rubber elasticity
The statistical theory of rubber elasticity, in the simplest approximation, uses a random
walk model to describe a polymer chain. Given an articulated polymer chain with N freely
jointed segments of length b, the contour length of the chain is Lc = N b. We consider
a configuration in which the end-to-end distance of the chain is R. The number of such
configurations is Z(R) = P (R)Z, where Z =∑
configurations exp(−H/∥BT ) is the partition
function. Here H is just a constant, as energy plays no role in this model. P (R) is the
probability for the chain to have the end-to-end distance R and is expressed as
P (R) = (3
2πb2N2)3/2 exp[−3R2/2bLc]
The entropic free energy of the polymer chain in this configuration is thus
F = −kBT ln(Z(R))
= Fo + kBT (3R2/2bLc)
= Fo + kBT (3R2/2R2
o)
For Nx chains in a volume element, one obtains
F = Fo + µ(3R2/2R2o) (10)
where µ = NxkBT is the shear modulus.
21
1.4 Theory of NLCE
1.4.1 De Gennes phenomenological theory
The first theoretical description of nematic elastomers was proposed by De Gennes, and is
antecedent to the first successful synthesis of a NLCE. Using a completely phenomenological
approach, he postulated that coupling rubber elasticity with nematic order would produce a
new class of soft materials with unprecedented properties and potential for engineering ap-
plications. De Gennes theory suggests that the energy cost of deforming a nematic elastomer
must be due to the relative rotations between the director field and the polymer network.
Considering a rotation of the body about an axis ω, it can be represented as: ωij =
1
2(∇iuj −∇jui). Here ui are the components of the displacement vector u. This rotation can
further be decoupled into a rotation around the director (ω∥ =1
2niϵijkωjk), and two rotations
around axes perpendicular to it (ω⊥ = njωij), where n is the nematic director. Naturally,
only those rotations about axes perpendicular to the director are expected to cost energy. Let
us define an infinitesimal change in the director due to the rotation as δn = Ω × n , such
that n · δn = 0. The relative rotation btween the body and the nematic director can then be
expressed as: δn− ω⊥ = (Ω− ω)× n.
The following three contributions to the energy are allowed by symmetry.
E1 =1
2D1 [(Ω− ω)× n]2
=1
2D1ΩiΩi
22
E2 =1
2D2 n · ε · (Ω− ω)× n
=1
2D2 Ωiεjknjδ
⊥ik
where δ⊥ik = δik − nink
E3 =1
2Cijklεijεkl
As can be seen, the first expression (E1) is the cost for relative rotations between the
director and the polymer matrix. The second expression (E2) couples the strain tensor to the
relative rotations of the polymer network and the director field. Finally, E3 is simply a strain
energy.
A free energy density that completely describes the elasticity of a nematic elastomer can
thus be written as:
F =1
2K11(∇ · n)2 + 1
2K22(n · ∇ × n)2 +
1
2K33(n ×∇× n)2
+1
2D1ΩiΩi +
1
2D2Ωiεjknjδ
⊥ik +
1
2Cijklεijεkl
De Gennes theory, although purely phenomenological, is able to adequately describe the
behavior of nematic elastomers at a macroscopic level. Experiments [36] have successfully
established a relationship between the coefficients D1 and D2 and microscopic and macro-
scopic material constants for side chain nematic LCE. One minor pitfall of this theory is its
inapplicability to main chain materials, as it is prohibitively difficult to decouple the rotations
of the nematogens from that of the network in main chain LCE.
23
1.4.2 Neo-classical theory
Although a macroscopic theory of nematic liquid crystal elastomers seems satisfactory
enough, a better, more comprehensive picture of the intrinsic mechanical properties and in-
teractions of NLCE with external stimuli requires a more strict, preferably microscopic ap-
proach. The most successful model for describing NLCE is the neo-classical theory intro-
duced by Warner and Terentjev [10] . It is an extension of rubber elasticity to anisotropic
polymer chains. In this description, one of the starting considerations is that some anisotropy
is conferred to the polymer chains due to the presence of the nematogens. That is, the shape
of gyration of the polymer chain deviates from a sphere to an ellipsoid. The relevant quantity
used to describe this anisotropy is r =l∥l⊥
, where l∥ is the length in the direction parallel
to the nematic director and l⊥ is the length in the transverse direction. An important quan-
tity is the step length tensor lij , which is usually isotropic in ordinary polymers. The chain
anisotropy r is usually more pronounced in main chain NLCE than in side chain materials.
lij =
l⊥ 0 0
0 l⊥ 0
0 0 l∥
(11)
Similar to low molecular weight liquid crystals, where order on a microscopic scale can
be related to macroscopic measurable quantities such as the dielectric anisotropy, the values
of l∥, l⊥ and hence r can be obtained by making measurements on a larger scale. In particular,
cooling a sample from the isotropic to the nematic phase leads to a spontaneous elongation
along the direction of the nematic director, and contraction perpendicular to it. The ratio of
this resulting extension and contraction provides a good measure for r. It can be shown that
24
the step length tensor depends on the nematic director as:
lij = l⊥δij + (l∥ − l⊥)ninj (12)
The sample in the reference state exists in a monodomain conformation with a director no
and step length tensor loij . An arbitrarily chosen polymer chain has a fully extended contour
length Lc and end-to-end distance Ro. The mean square displacement is⟨Ro
iRoj =
1
3Lcl
oij
⟩.
In light of the the discussion of rubber elasticity from the preceding section, one can write
the end-to-end distance as a Gaussian distribution.
P (Ro) = Det[loij]exp[
−3
2Lc
Roi (l
oij)
−1Roj ] (13)
If the sample undergoes a deformation from its reference state to the current state, the
theory assumes that the deformation is affine, i.e. the separation between two crosslinks
is proportional to the shape change of the whole sample. In other words, one can write
the new end-to-end distance as Ri = λij(oRj) where λij is the macroscopic deformation
gradient tensor. The current step length tensor also differs from the reference lij = loij and the
probability distribution of the end-to-end length of the chain is
P (R) = Det [lij] exp(−3
2Lc
Ril−1ij Rj) (14)
Using this, one easily obtains the elastic energy as :
F =1
2NxkBT Tr[loij λ
Tij l
−1ij λij] (15)
Recalling that the step length tensor is a function of the nematic director, one easily sees
that strains are coupled to the director configuration. In the small strain regime, the equation
25
above, with the substitution λij = δij + εij , yields the coefficients of de Gennes’ relative
rotations between nematogens and the polymer network.
D1 = NxkBT(l∥ − l⊥)
2
l∥l⊥
D2 = NxkBTl2∥ − l2⊥
l∥l⊥(16)
1.4.2.1 Soft and semisoft elasticity
Figure 13: Soft elastic deformation. The director rotates to accommodate the strain.
The trace formula above has been successfully used in various studies of nematic elas-
tomers, yielding results in remarkable agreement with experiments. Of particular interest to
the studies reported in this dissertation is the prediction from this theory that certain modes
of deformation can be achieved at almost no energy cost. These energy-free deformations are
those in which a rotation of the system’s internal degree of freedom, i.e. the nematic direc-
tor, occurs in order to balance the cost of the elastic distortion. As its roots, soft elasticity
assumes an isotropic reference state which, when cooled into the nematic phase, results in
the elongation of the sample in the direction of the spontaneously formed director. As there
are infinitely many equiprobable possibilities for the orientation of the nematic director at the
26
transition, one can expect that finding a path that maps two or more such states is equivalent
to achieving a deformation at no energy cost.
It is easily proved that such modes of deformation exist, at least mathematically [37, 38].
As an illustration, choosing a deformation of the form
λij = l1/2ij Rθ (l
oij)
1/2,
where Rθ is an arbitrary rotation by an angle θ, one obtains a zero energy cost from the
configuration with step length tensor loij to that with step length tensor lij .
Although this is easily thought of for a single polymer strand, it is clearly an idealiza-
tion, as the polymer chains differ slightly in their composition; that is, a deformation may be
soft for one strand but not for the other. In other words, the onset of director rotation might
occur at different thresholds for different polymer strands, hence the term semisoft elastic-
ity [39–41]. A modified model exists [42] that accounts for the compositional fluctuations in
the sample. This phenomenon is experimentally observed when a thin sheet of monodomain
nematic elastomer is subjected to a uniaxial stretch. The stress-strain curve of such an elastic
deformation shows a very peculiar behavior that depends on the relative orientation of the
nematic director with respect to that of the imposed strain. For strains applied parallel to the
director, the stress-strain curve is similar to that of a pure elastic material, whereas strains
applied perpendicular to the director display two elastic regimes separated by a plateau. The
plateau corresponds to the soft region, i.e. that in which the nematic director rotates to ac-
commodate the strains.
27
Figure 14: Soft vs. semisoft response in a nematic elastomer.
CHAPTER 2
FINITE ELEMENT ELASTODYNAMICS SIMULATIONS OF LCE
2.1 Introduction
The dynamics of shape evolution in nematic liquid crystal elastomers is modeled using a
three dimensional finite element elastodynamics approach. This model predicts the macro-
scopic mechanical response to an external stimulus such as a change in nematic order, e.g.
by heating or cooling through the isotropic-nematic transition or, in azo-doped materials, by
exposure to light. The mechanics of nematic LCEs are thus controlled by intrinsic coupling
between nematic order and mechanical strain. Theoretical models of this coupling can be
solved analytically, e.g. if the goal is to predict the mechanical response of a representative
volume element, as described at length in [10] and references therein. The goal of the present
work, however, is more ambitious: we wish to model entire devices containing LCE actuators
and simulate their behavior on laboratory length and time scales, including dynamics as well
as static response, in three dimensions. Geometries and boundary conditions of interest are
not simple enough to allow for analytical solutions, so we turn to finite element simulation
methods to simulate the elastodynamics. Finite element methods have been used previously
to model 2-d statics of LCEs but have not, to our knowledge, previously been used to model
dynamics in 3-d [43].
Instead of using a preconfigured software package, we have developed our own finite
element simulation code based on a Hamiltonian approach. Our algorithm is based on a
28
29
marvelously simple approach to finite element elastodynamics proposed by Broughton et
al. [44, 45]. However Broughtons algorithm relies on the approximation that both strain and
rotation are small, and their finite element Hamiltonian is not invariant under rotation. As a
result, dynamics calculated from such a Hamiltonian conserve energy poorly, particularly in
the case of finite rotations. As a solution to this difficulty, we replace the linear strain tensor in
Broughtons approach with the Green-Lagrange strain tensor, a measure of deformation that is
invariant under sample rotation, containing both linear and nonlinear terms. This substitution
renders our Hamiltonian rotationally invariant, and thus our algorithm is not limited to the
small-rotation limit. The resulting dynamics shows remarkable numerical stability, and total
energy and momentum are both conserved to high precision.
In the spirit of the finite element method, we model a sample of nematic elastomer by
discretizing its volume into a mesh of tetrahedral elements. Consider a 3-d tetrahedral element
composed of an elastic material. Any arbitrary deformation u, v, or w of the element can be
described via a standard set of linear mapping functions that find the displacement of a given
interior point of an element by interpolating the displacements of its vertices. Note that using
such an affine mapping results in a uniform strain inside the element is uniform. The choice
of tetrahedral elements in 3-dimensions is not arbitrary, as simplexes or elements with n + 1
vertices in n dimension make this linear mapping very convenient.
30
u(x, y, z) = a1 + a2x+ a3y + a4z
v(x, y, z) = b1 + b2x+ b3y + b4z
w(x, y, z) = c1 + c2x+ c3y + c4z
In the unstrained states (Figure 15(a)), the vertices of the tetrahedron are at coordinates
(x1, y1, z1), (x2, y2, z2), (x3, y3, z3), and (x4, y4, z4). Any displacement of the four vertices
can be written as a function of mapping coefficients as shown below:
(a) Unstrained (b) Deformed
Figure 15: (a) and (b) show a tetrahedron in the unstrained and deformed states respectively
31
u1
u2
u3
u4
=
1 x1 y1 z1
1 x2 y2 z2
1 x3 y3 z3
1 x4 y4 z4
a1
a2
a3
a4
v1
v2
v3
v4
=
1 x1 y1 z1
1 x2 y2 z2
1 x3 y3 z3
1 x4 y4 z4
b1
b2
b3
b4
w1
w2
w3
w4
=
1 x1 y1 z1
1 x2 y2 z2
1 x3 y3 z3
1 x4 y4 z4
c1
c2
c3
c4
Here u, v, and w are displacements in the x, y, and z directions respectively. The coeffi-
cients of the mapping functions ai, bi, and ci are known as the shape factors and are calculated
every time step. They are calculated by simply inverting the matrix above. The inverse matrix
however is obtained once at the start of the simulation, that is, it needs not be updated as the
reference state remains the same throughout the simulation. When the tetrahedral element
experiences a deformation, with displacement fields u, v, and w, we can calculate the defor-
mation gradient tensor, strain tensor, the total elastic energy, and the forces on its vertices as
follows:
32
λij =
1 + a2 a3 a4
b2 1 + b3 b4
c2 c3 1 + c4
ε11 =∂u
∂x+
1
2
[(∂u
∂x
)2
+
(∂v
∂x
)2
+
(∂w
∂x
)2]
= a2 +1
2(a22 + b22 + c22)
ε22 =∂v
∂y+
1
2
[(∂u
∂y
)2
+
(∂v
∂y
)2
+
(∂w
∂y
)2]
= b3 +1
2(a23 + b23 + c23)
ε33 =∂w
∂z+
1
2
[(∂u
∂z
)2
+
(∂v
∂z
)2
+
(∂w
∂z
)2]
= c4 +1
2(a24 + b24 + c24)
The off-diagonal components of the strain tensor are found as well.
ε12 =1
2
(∂u
∂y+
∂v
∂x
)+
1
2
(∂u
∂x
∂u
∂y+
∂v
∂x
∂v
∂y+
∂w
∂x
∂w
∂y
)=
1
2(a3 + b2) +
1
2(a2a3 + b2b3 + c2c3)
33
ε13 =1
2
(∂u
∂z+
∂w
∂x
)+
1
2
(∂u
∂x
∂u
∂z+
∂v
∂x
∂v
∂z+
∂w
∂x
∂w
∂z
)=
1
2(a4 + c2) +
1
2(a2a4 + b2b4 + c2c4)
ε23 =1
2
(∂v
∂z+
∂w
∂y
)+
1
2
(∂u
∂y
∂u
∂z+
∂v
∂y
∂v
∂z+
∂w
∂y
∂w
∂z
)=
1
2(b4 + c3) +
1
2(a3a4 + b3b4 + c3c4)
Rewriting the strain energy U =1
2Cijklεijεkl in terms of the strain tensor, one obtains:
U =1
2Cxxxx(ε
211+ε222+ε233)+2Cxxyy(ε11ε22+ε22ε33+ε33ε11)+4Cxyxy(ε
212+ε223+ε231) (17)
The elastic constants Cijkl are obtained as
Cxxxx = λ+ 2µ
Cxxyy = λ
Cxyxy = µ
where λ and µ are the are the bulk and shear moduli respectively. These depend on the
Young’s modulus(E) and Poisson’s ratio(ν) of the material as shown below.
λ =E ν
(1− 2ν)(1 + ν)
µ =E ν
2(1 + ν)
34
2.1.1 Forces calculations
The force on the ith node is then obtained as derivatives of U with respect to the node
displacement.
Fx,i = −Cxxxxε11M2i(1 + a2)− Cxxxxε22M3ia3 − Cxxxxε33M4ia4
−Cxxyyε11M3ia3 − Cxxyyε22M2i(1 + a2)− Cxxyyε22M4ia4
−Cxxyyε33M3ia3 − Cxxyyε33M2i(1 + a2)− Cxxyyε11M4ia4
−2Cxyxyε12M3i(1 + a2)− 2Cxyxyε12M2ia3 − 2Cxyxyε23M4ia3
−2Cxyxyε23M3ia4 − 2Cxyxyε31M4i(1 + a2)− 2Cxyxyε31M2ia4
Fy,i = −Cxxxxε11M2i(1 + b2)− Cxxxxε22M3ia3 − Cxxxxε33M4ia4
−Cxxyyε11M3ia3 − Cxxyyε22M2i(1 + a2)− Cxxyyε22M4ia4
−Cxxyyε33M3ia3 − Cxxyyε33M2i(1 + a2)− Cxxyyε11M4ia4
−2Cxyxyε12M3i(1 + a2)− 2Cxyxyε12M2ia3 − 2Cxyxyε23M4ia3
−2Cxyxyε23M3ia4 − 2Cxyxyε31M4i(1 + a2)− 2Cxyxyε31M2ia4
Fz,i = −Cxxxxε11M2i(1 + a2)− Cxxxxε22M3ia3 − Cxxxxε33M4ia4
−Cxxyyε11M3ia3 − Cxxyyε22M2i(1 + a2)− Cxxyyε22M4ia4
−Cxxyyε33M3ia3 − Cxxyyε33M2i(1 + a2)− Cxxyyε11M4ia4
−2Cxyxyε12M3i(1 + a2)− 2Cxyxyε12M2ia3 − 2Cxyxyε23M4ia3
−2Cxyxyε23M3ia4 − 2Cxyxyε31M4i(1 + a2)− 2Cxyxyε31M2ia4
(18)
35
Here the quantities M2i, M3i, and M4i are the components of the inverse of the shape matrix
defined as
M =
1 x1 y1 z1
1 x2 y2 z2
1 x3 y3 z3
1 x4 y4 z4
−1
.
It is easy to see from the above that for a single tetrahedral element with no external forces
or constraints, all these node forces sum to zero component by component, and the total
torque also must sum to zero. This is a consequence of the conservation of linear and angular
momentum. For a mesh of connected elements, again with no external forces or constraints,
each node receives force contributions from each element of which it is a member. The sum
of forces on any node is in general not zero. Observe that the forces on the vertices should be
derivatives of the total elastic energy. This is important to point out because in the following
chapters, the elastic energy will have other terms added, which will need to be taken into
account when calculating the node forces.
2.2 Algorithm
To construct a Hamiltonian we also need to specify the kinetic energy of the system. Here
we follow the method used by Broughton et al. [44] and apply the “lumped mass approxima-
tion”, which assumes that all the mass is concentrated in the nodes of the finite element mesh.
In our simulation, the initial position and velocity of each node in the material are specified
in the initial state. The subsequent dynamics of the system are calculated explicitly using a
finite time step. After each time step, the potential energy in Equation 17 is calculated for
36
each element as a function of the corner nodes displacements from their initial positions in
the reference state. Forces on each node are calculated as a derivative of the total potential
energy in all adjacent volume elements with respect to the nodes position. The node positions
and velocities are then updated using the velocity Verlet method [46]:
xi(t+∆t) = xi(t) + vi(t)∆t+1
2ai(t)(∆t)2
vi(t+∆t) = vi(t) +ai(t) + ai(t+∆t)
2∆t
This finite element explicit dynamics algorithm closely resembles the familiar molecu-
lar dynamics method and is almost as easy to code. However, here we are moving nodes
rather than atoms, and instead of an interatomic potential we are using the continuum elastic
potential energy, expressed as a function of node displacements.
2.3 Dissipation
To add internal damping associated with velocity gradients in the sample, we use a mod-
ified form of Kelvin dissipation. In its standard form, the Kelvin dissipation force (e.g. be-
tween two particles, or between two nodes in a finite element mesh) is proportional to the
velocity difference between them (see e.g. [47]). This form conserves linear momentum but
violates conservation of angular momentum; internal dissipation forces could create torque,
which is of course unphysical. We modified the Kelvin dissipation form to provide for con-
servation of angular momentum, that is, dissipation forces between any pair of nodes must
act along the line of sight between them, so they create no torque [48]. We also scale the
dissipation force so it depends on the effective strain rate between two nodes rather than
37
their absolute velocity difference. With these modifications, the dissipation force between a
pair of neighboring nodes separated by distance d is F12 = −η(v1 − v2) · (r1 − r2)
d12r12, with
η = 10−7 kg.m/sec. The resulting dissipation is isotropic in character and does not depend
on the orientation of the director field.
2.4 Simulations of Rubbery materials
Figure 16: Beam of rubber twisting.
Equipped with the tools described in the previous sections, we venture into a simple test
case which consists in simulating an isotropic piece of rubber of size 10mm× 5mm× 1mm
that is subjected to a torsional deformation; see Figure 16 . The material parameters used
for this simulation are a Young’s modulus E = 1.5MPa, and Poisson ratio very close to the
38
e
Figure 17: Potential (black dots) and kinetic (red dots) energy of the elastic beam experi-encing torsional deformation. In the absence of dissipation, the energy is conserved (greendots).
incompressibility limit ν = 0.499. Note that a Poisson ratio of ν = 0.5 is not numerically
achievable as that would result in an infinite bulk modulus. As shown in Figure 17, in the
absence of dissipative forces, energy is conserved up to a part in 108.
It is worth mentioning here that prior to performing all the simulations reported from this
point on, several numerical experiments were performed for benchmarking. In particular, the
discretization of the volume into a tetrahedral mesh was handled in such a way that the distri-
bution of element sizes was close to uniform. The appropriate time step for each simulation
was chosen to be smaller than any characteristic time scale of the system. All benchmarks
showed that for a Poisson ratio ν = 0.499, the sample deforms with volume fluctuations are
less than a part in a thousand.
39
2.5 Simulations of LCE
To apply the method described in the previous section to nematic elastomers, there must
be extra terms in the potential energy to account for the nematic interaction and the strain-
order coupling. In the simplest approximation we consider a linear coupling between the
strain tensor and the nematic order as proposed by de Gennes [16]. One can thus rewrite the
potential energy as:
U = Ustrain − αεijQij (19)
The nematic order parameter tensor Qij is defined in the material frame and transforms like
the strain tensor; that is, rotations of the whole sample in the laboratory frame leave these
quantities unchanged. While one normally couples the left strain tensor to the order parameter
tensor in order to have a frame indifferent energy [49] , the coupling above is also valid if
the nematic director nref used to construct Qij lives in the reference frame; that is, nref is
obtained by an inverse transformation from the target to the reference space before coupling
the Green-Lagrange strain tensor εij to Qij . The said inverse transformation is precisely
described by the unitary matrix (rotation) obtained by polar decomposition of the deformation
gradient tensor.
To illustrate this, let us define by nt the local nematic director in the target space. Recall that
the polar decomposition of the deformation gradient tensor introduced earlier reads:
λ = RU = V R
Where U and V are positive definite and R is unitary. From this, it follows that λT = RT V .
First, coupling the left Green tensor to the order parameter reads:
40
(λλT )(nt ⊗ nt) = nt · (λλT )nt
= | λT nt |2
= | RTV nt |2
= | V nt |2
If we write nref = RT nt , then coupling the right Green tensor to the order parameter
tensor in the reference space reads:
(λTλ)(nref ⊗ nref ) = | λnref |2
= | λRT nt |2
= | V nt |2
Clearly, coupling the left Green tensor to the target order tensor is equivalent to coupling the
right Green tensor to the appropriately transformed order tensor described above.
We have developed two distinct approaches to simulating nematic elastomers. In the first
and the simplest approximation, we can control the state of the nematic order and observe the
mechanical response of the sample. In particular, by increasing (resp. reducing) the amount of
order in the sample, one can mimic the cooling (resp. heating) of the material, hence allowing
to study the shape change of nematic LCE undergoing a Nematic-Isotropic phase transition.
In chapters 3 and 4, we present a more advanced approach in which we simultaneously model
shape change and microstructural evolution. This approach allows us to track the dynamics
of the nematic director field and predict mechanical response. The main assumption used is
that the nematic director is always in quasi-static equilibrium with the slowly evolving strain
41
field. This is justified by the observed discrepancy between the time scales of the network and
director relaxations [50]. This permits one to study the nucleation and evolution of stripes
and the polydomain to monodomain transition. It is appropriate to highlight here that this
approach that considers isotropic rubber elasticity and a coupling between strains and order
tensors is not at odds with others such as the neoclassical which is widely used. One can
show that one form can be recovered from the other [51]. To such end, we start by rewriting
the de Gennes form of the energy in the following form.
U =1
2µ[Tr(λλT )− αQ · (λλT )]
=1
2µ[Tr(λλT )− αλλT · S(n⊗ n− 1
3I)]
=1
2µ[Tr(λλT )− α S n · (λλT )n− 1
3Tr(λλT )]
=1
2µ[Tr(λλT )− α S | λT n |2 −1
3Tr(λλT )]
=1
2µ[(1 +
αS
3)Tr(λλT )− α S | λT n |2]
Similarly, assuming an isotropic reference state, i.e. lo = I , we write neoclassical energy
as:
U =1
2µTr(l−1λλT )
=1
2µTr
[((1
l∥− 1
l⊥
)n⊗ n+
1
l⊥
)λλT
]=1
2µTr
[(1
l∥− 1
l⊥
)(λλT )(n⊗ n) +
1
l⊥λλT
]=1
2µ
[(1
l∥− 1
l⊥
)| λT n |2 + 1
l⊥Tr(λλT )
]Equating the two expressions above, the phenomenological constant α that couples the
strain tensor to the nematic order parameter tensor is be expressed in terms of the components
42
of the step length tensor as: αS = le
(1
l⊥− 1
l∥
). Here one defines le =
3
l∥ + 2l⊥.
2.5.1 Isotropic-Nematic phase transition
We present here a simulation of a thin strip of nematic elastomer undergoing the transition
from the isotropic to the nematic phase. The film has dimensions 1.5 mm × 0.5 mm with a
thickness of 100 µm. The shear and bulk moduli are µ = 5.7×105Pa and Br = 2.8×107Pa
, respectively, comparable to that of an isotropic rubber. In this simulation, the strain-order
coupling parameter is α = µ. The sample is initially in the high temperature isotropic phase,
and we use the first approach described above to evolve the simulation. The strength of the
nematic scalar order parameter is increased, and as a response, the sample elongates in one
direction and contracts in the other two. In a transition from the isotropic to nematic phase
in NLCE, the spontaneous nematic director can point in any direction, and this will normally
lead to an uncontrolled shape change of the sample. In order to break the symmetry, one end
of the sample of the sample is constrained not to move. This study provides a qualitative
agreement with the large shape change observed in similar experiments [12].
2.5.2 Semisoft Elasticity
Following the description of the semisoft elastic response presented in the preceding chap-
ter, we can model this phenomenon. If we consider an element with the director initially
oriented at an angle ϕ with respect to, say, the z-direction, then a uniaxial strain applied in
the in y-direction has the effect of rotating the director to a new angle θ. The strain energy
for the deformation as described in the neoclassical theory then becomes a function of θ.
43
Figure 18: Simulation of the isotropic-nematic transition in a liquid crystal elastomer. Thesample experiences a macroscopic shape change as it is cooled down to the nematic phaseand heated back into the isotropic phase. The sample is clamped at the top end.
U = µ1
4(1
r− 1)(2a2(a3 cos(2θ)((r − 1) cos(2ϕ) + r + 1) + (r − 1) sin(2ϕ)((b3 + 1) cos(2θ)
−b2 sin(2θ))− (r − 1) sin(2θ) cos(2ϕ)− (r + 1) sin(2θ)) + 2(r − 1) sin(2ϕ)
((a3(b3 + 1)− b2) sin(2θ) + (a3b2 + b3 + 1) cos(2θ))
+(r − 1) cos(2ϕ)((a23 + b22 − b3(b3 + 2)− 2) sin(2θ) + 2(a3 − b2(b3 + 1)) cos(2θ))
+a22 sin(2θ)(−((r − 1) cos(2ϕ) + r + 1)) + a23r sin(2θ) + 2a3r cos(2θ)
+a23 sin(2θ) + 2a3 cos(2θ)− b22r sin(2θ) + b23r sin(2θ)
+2b3r sin(2θ) + 2b2r cos(2θ) + 2b2b3r cos(2θ)− b22 sin(2θ)
+b23 sin(2θ) + 2b3 sin(2θ) + 2b2 cos(2θ) + 2b2b3 cos(2θ)
−2c22 sin(2θ) + 2c23 sin(2θ) + 4c2c3 cos(2θ))
44
Figure 19: Cartoon of the semisoft response in a NLCE.
Solving for θ is straightforward and gives the orientation of the director for a given strain.
In particular, for the most commonly studied experiment in which the strain is imposed per-
pendicular to the director orientation, one uses ϕ =π
2. This yields a rotation of the director
by an amount
θ =1
2tan−1
(2((a2 + 1)a3 + b2(b3 + 1)r + c2c3)
(a2 + 1)2 − a23 + b22r − (b3 + 1)2r + c22 − c23
)
This suggests that one can always know the state of the nematic director given the strain.
However, as appealing as this may appear, it will not be our method of choice in the subse-
quent chapters. This is simply because here the strain is treated as an external global variable
that couples to the local nematic director. In order to study microstructure formation and evo-
lution, it is important to allow the strains to vary locally in response to changes in the director
orientation and vice-versa.
CHAPTER 3
POLYDOMAIN-MONODOMAIN TRANSITION IN NEMATIC ELASTOMERS
3.1 Introduction
Liquid crystal elastomers, when crosslinked, usually exist in the polydomain configura-
tion, consisting of a large array of randomly oriented, micron sized domains, see Figure 20.
The domains may exhibit a local average degree of order, but globally there is no long range
order. A sample of such material will strongly scatter light that is incident upon it due to the
size of the domains.
Monodomain liquid crystal elastomers can be obtained by means of the Finkelmann two-
step crosslinking method, which consists in a weak crosslink followed by another crosslink
under the influence of an aligning field, e.g. an electric or magnetic field, or external strain.
Deformation of an initially polydomain nematic elastomer film induces a transition to
the monodomain configuration. We model the resulting microstructural evolution and stress-
strain response using a novel finite element elastodynamics simulation approach. We explore
how the thermomechanical history of the sample, e.g. its crosslink density and phase at time
of network formation, affects the width of the poly-monodomain transition and the associated
stress-strain behavior. We find that when the sample is cross-linked in the isotropic phase,
the material shows a semi-soft response with a well-defined plateau in the stress-strain curve.
By contrast, when the sample is cross-linked in the nematic phase, the resulting strong local
disorder broadens the transition, and the plateau is much less pronounced. These simulation
45
46
Figure 20: A cartoon of a polydomain nematic elastomer shows no correlation between theorientation of the director in neighboring domains. The blue arrows show the orientation ofthe nematic director in the domains.
results yield qualitative agreement with recent experimental observations. We also study the
rate-dependent material response under uniaxial extension. This simulation approach allows
us to explore the fundamental physics governing dynamic mechanical response of nematic
elastomers and also provides a potentially useful computational tool for engineering device
applications.
When a polydomain LCE thin film is stretched uniaxially, its orientational domains align,
producing long-range orientational order. This poly-to-monodomain (P-M) transition has
been well-characterized [52–54]. Some polydomain materials exhibit semi-soft mechanical
response, while others do not, with no universal behavior. It was not clear what aspect of
composition or processing determines the nature of a sample’s mechanical response.
47
Recent work by Uruyama and coworkers [55] shed some light on this mystery by study-
ing the P-M transition in samples with the same chemical composition but different thermo-
mechanical history. They prepared polydomain samples in two different ways. Nematic-
crosslinked polydomain nematic elastomers, or N-PNE, were prepared by cross-linking in
the nematic phase with no aligning field. N-PNE samples display short-range orientational
order with approximately micron-sized domains. Isotropic-crosslinked polydomain nematic
elastomers, or I-PNE, were prepared by cross-linking in the isotropic phase with no aligning
field and then cooling through the I-N transition. I-PNE samples display stronger disorder,
with orientational domains too small to observe via polarization microscopy.
Uruyama et al found that under uniaxial extension, I-PNE samples show a sharp P-M tran-
sition with a clear semi-soft mechanical response, while N-PNE materials show a broadened
P-M transition and no pronounced plateau in the stress-strain curve. When the applied strain
was relaxed, both types of samples recovered their initial shape, but N-PNE also recovered
the same initial polydomain texture, indicating a strong local memory effect.
Discrete lattice models in both 2-d [56,57] and 3-d [58] have provided key insights into the
role of local heterogeneity in both monodomain and polydomain LCE. Yu et al. explored how
heterogeneity affects long-range correlations in director orientation in polydomain LCE in the
absence of applied strain. Uchida [56] modeled the P-M transition in LCE films cross-linked
in the isotropic state, demonstrating that local heterogeneity broadens the transition region. J.
Selinger and Ratna [58] showed that the Isotropic-Nematic (I-N) transition in monodomain
LCE may be broadened by heterogeneity in either random fields or bonds. In these lattice-
based approaches, strain is treated as a global variable playing the role of an external applied
48
field; thus the details of sample shape evolution are not predicted.
Desimone et al. [43] used 2-d finite element elastostatics methods to model the mechanical
response of a thin monodomain LCE film under uniaxial strain. They modeled the sample
as a 2-d homogenized composite of domains with different orientations. This innovative
approach successfully reproduced both the soft mechanical response and the sample’s overall
shape evolution, but without explicitly modeling the resulting microstructure or the associated
rate-dependence.
In this chapter, we describe 3-d finite element elastodynamics simulation studies of the
P-M transition in both N-PNE and I-PNE polydomain samples, with spatial resolution of the
nematic director field down to the micron scale. This model allows us to simulate mechanical
response, shape change, microstructural evolution, and strain-rate effects. We use the model
to investigate in detail how a sample’s mechanical response depends on its thermo-mechanical
history. We also quantify the degree of global orientational order induced via applied strain.
To study the mechanical response of LCE thin films, we carry out computer simulations
using the 3-d finite element elastodynamics approach described in the previous chapter. A
sample of arbitrary shape is discretized into a nonuniform mesh of volume elements using
the Netgen algorithm [59]. Each element represents an approximately micron-sized nematic
domain. The deformations of an element are expressed in terms of shape functions that
interpolate the position of every point within the element from the position of its vertices [60].
The shape function used in this case is an affine one, and for simplicity, we use elements of
simplectic form, here tetrahedra in 3-d; this implies that the strain is uniform within each
element, though it may vary from one element to the next. We define a free energy density
49
for each element that comprises three parts:
U = Ustrain + Ustrain−order + Umemory (20)
The first term is the well-known neo-Hookean strain energy:
Ustrain =1
2µTr(λkiλkj) +K(Det[λij]− 1)2 (21)
where λij is the deformation gradient tensor . Volume conservation is maintained by a large
bulk modulus K. The second term Ustrain−order describes the coupling between mechanical
strain and orientational order:
Ustrain−order = −αεij(Qij −Qoij) (22)
Here we use the rotationally invariant Green-Lagrange strain tensor εij = 12(λikλkj −
δij), defined locally for each element. Local orientational order is characterized by Qij, the
symmetric and traceless uniaxial order parameter tensor. Qoij is the local order parameter
tensor present at the time of crosslinking. The parameter α is proportional to the density
of crosslinks in the sample. Both Qoij and α depend on the details of sample preparation as
discussed below.
The third term Umemory describes the crosslink memory effect:
Umemory =1
2β(Qij −Qo
ij)2 (23)
This term biases the nematic director to remain parallel to its orientation at the time of
crosslinking, thus playing the role of a local field whose orientation is defined by Qoij . The
50
strength of the field is assumed to be uniform throughout the sample and is defined by the
parameter β. We expect that samples crosslinked in the nematic phase have ’stronger’ mem-
ory of their initial state, and thus larger β, than those crosslinked in the isotropic phase. The
uniaxial order parameter tensor Qij has three independent degrees of freedom. We assume
no biaxiality in the system and hold the nematic scalar order parameter S constant within the
element; we further assume that S does not vary spatially across the sample. Thus, the only
degrees of freedom of Qij that are allowed to vary are the polar and azimuthal angles that
define the orientation of the local nematic director.
When the sample is subjected to an external uniaxial strain, the forces on the vertices
of each element are calculated as derivative of the free energy density F = − U . The
contribution to the force from the elastic potential energy is :
Fx,i = K (2 (b4c3 − (b3 + 1) (c4 + 1))M2iZ + 2 (a3 (c4 + 1)− a4c3)M3iZ
+2 (a4 (b3 + 1)− a3b4)M4iZ)
+µ ((a2 + 1)M2i + b2M3i + c2M4i)
Fy,i = K (2 (b2 (c4 + 1)− b4c2)M2iZ + 2 (a4c2 − (a2 + 1) (c4 + 1))M3iZ
+2 ((a2 + 1) b4 − a4b2)M4iZ)
+µ (a3M2i + (b3 + 1)M3i + c3M4i)
Fz,i = K (2 ((b3 + 1) c2 − b2c3)M2iZ + 2 ((a2 + 1) c3 − a3c2)M3iZ
+2 (a3b2 − a2 (b3 + 1)− b3 − 1)M4iZ) +
µ (a4M2i + b4M3i + (c4 + 1)M4i)
51
where
Z = (c2 (a4 (b3 + 1)− a3b4) + c3 ((a2 + 1) b4 − a4b2)
− (c4 + 1) (−a3b2 + a2 (b3 + 1) + b3 + 1) + 1)
The momenta of the elements are computed using the lumped mass approximation whereby
the mass of each element is equally distributed to its vertices, also refered to here as nodes
[45, 61]. The simulation evolves in two steps. First, holding Qij fixed, the nodes forces are
calculated and integrated forward in time via the velocity Verlet algorithm. The new strain
tensor can then be defined from the current and old positions of the nodes. The next step
consists in holding the nodes positions fixed and relaxing the nematic order parameter tensor
in order to minimize the free energy density. Since the local nematic director is defined as
n = (sin θ cosϕ, sin θ sinϕ, cos θ) with respect to the laboratory frame, relaxing the nematic
order tensor amounts to minimizing the free energy density with respect to its degrees of free-
dom, namely the azimuthal and polar angles of the local director ϕ and θ respectively. Note
that the only terms that depend on ϕ and θ are Ustrain−order and Umemory. The minimization
is performed using Powell’s method, which is a form of conjugate gradient method that does
not require the derivative of the function to be computed.
minimizeθ,ϕ
U(θ, ϕ)
This algorithm allows us to simultaneously model microstructural evolution and macro-
scopic shape change of the sample.
52
3.2 Initial configuration
The thermomechanical history of the sample as stated above plays a very important role in
the sample’s mechanical response. Here we will attempt to study the response of two samples,
one crosslinked deep in the nematic phase, and the other in the isotropic phase. It is worth
describing how we obtain the initial configuration prior to the imposition of the strain. With
the sample discretized in a tetrahedral mesh, we assign a nematic director to each element.
To mimic an isotropic crosslinked nematic elastomer, we start with a random distribution
of the orientations of the nematic directors (Figure 21(a)). Urayama et al [55] have shown
that the nematic elastomers with nematic genesis show the schlieren texture observable in
low molecular weight liquid crystals under polarizing microscopy. In our studies, a nematic
crosslinked initial state (Figure 21(b)) is achieved by running a simulation that takes the
sample into the nematic phase. This is done by using the Lebwohl-Lasher model of nematic
liquid crystals. It is simply equivalent to a lattice approximation of the Maier-Saupe theory.
It uses a Hamiltonian of the form
H = J∑<i,j>
[1− (ni · nj)2]
where the summation is carried over neighbouring pairs (i, j). J is the field coupling strength.
Note that the nematic directors are defined at the centroids of the elements, and, although the
mesh is not regular, the distance between the centroids of neighbouring elements is approx-
imately constant throughout the mesh. The Monte Carlo Metropolis algorithm is used to
anneal the sample, and one obtains different sizes of correlation length by quenching. Note
that this simulation involves only the nematic directors, and not the elastic degrees of freedom
53
of the sample.
(a) Isotropic genesis (IPNE) (b) Nematic genesis (NPNE)
Figure 21: Initial configurations with different thermomechanical histories.
3.3 Simulation and results
We present here simulation results obtained for a uniaxial stretching of a polydomain
film of dimensions 1.5 mm × 0.5 mm with a thickness of 100 µm, with shear modulus
µ = 5.7 × 105Pa, bulk modulus Br = 2.8 × 107Pa. The strain-order and memory cou-
pling parameters for are α = µ and β = 0.3µ, respectively. In these simulations, we used
γ = 10−7J . The strip of nematic elastomer was discretized in a mesh of approximately
80,000 tetrahedral elements. The initial director orientation of the directors is either random
(isotropic crosslinked), or composed of uncorrelated domains of size about an order of mag-
nitude larger than the element size (nematic crosslinked); that is, there is no global order in
the system. The sample is clamped at its end, i.e. the components of the displacement and
velocity perpendicular to the direction of imposed strain are always zero for the nodes in these
regions. All other nodes of the finite element mesh are allowed to move in any direction. The
54
(a) ∆λ/λ = 0
(b) ∆λ/λ = 0.5
Figure 22: Simulation studies of a I-PNE. (a) and (b) show the entire strip of nematic elas-tomer in the polydomain and monodomain configuration respectively. In the unstrained state,the sample strongly scatters light, whereas when strained the sample is sandwiched betweencrossed polarizers and aligned with either the polarizer or analyzer, total extinction of theincident light will occur.
clamped regions are moved at a constant speed of 1 mm/sec.
As the experiments on this materials are usually carried out in a quasi static manner due
to their slow stress relaxation [62], it was of paramount importance in these simulations to
investigate the strain rate dependance of the transition. High rates of strain resulted in out of
equilibrium situations, often missing entirely the dynamics of the polydomain-monodomain
transition when the latter occured within a narrow range of strains. A discussion of the strain
rate dependance of the elastic response in this materials can be found in the next chapter. This
was easily alleviated by using smaller time steps, with the drawback that the computation
time was greatly increased. Our programs thus had to be optimized for parallel computations
using MPI, and the simulations were executed on a cluster architecture hosted by the Ohio
Supercomputer Center. The typical time step used in the simulations shown here is 0.1 µs.
55
Figure 23: Texture and director configuration in a region near the center of the strip in thepolydomain state of I-PNE.
While the simulation evolves as described in the previous section, we track the time evolution
of the local director in each element, together with each element and the overall mesh shape
change. As demonstrated in [52], we observe that the polydomain to monodomain transition
in nematic elastomers proceeds by rotation of the domains rather than domain growth. During
the deformation, the sample goes from a scattering state as a result of the nonuniformity in
the director orientations to a transparent one.
56
Figure 24: Texture and director configuration in a region near the center of the strip in themonodomain state of I-PNE.
The textures as would be observed under optical polarizing microscopy are shown in
Figures 25 and 26. The local directors in the regions near the clamps are not allowed to
rotate, resulting in the formation of defects in the director orientation. These regions strongly
scatter light, even after the director rotation is complete far from the clamps. The resulting
stress-strain behavior is shown in Fig. 27 for samples with crosslinked in the isotropic phase,
and in Fig. 28 for a sample crosslinked in the nematic phase. The stress here is the engineering
57
Figure 25: Texture and director configuration in a region near the center of the strip in thepolydomain state of N-PNE.
stress which is obtained by dividing the average of the normal forces applied to the sample
by the area of the cross-sectional where these forces are applied. The same plots also show
the dependence of the global order parameter S of the sample on the imposed strain. S is
simply the average of the second Legendre polynomial (S = ⟨P2(cos θ)⟩) over the whole
mesh, where θ is the angle between the local director and the direction of the applied strain.
For samples crosslinked in the nematic phase, the stress-strain curve displays a linear
58
Figure 26: Texture and director configuration in a region near the center of the strip in themonodomain state of N-PNE.
regime followed by a plateau in the range of strains that correspond to the rotation of the
domains towards the direction of the applied stress. Upon completion of the directors rotation,
the linear regime is recovered. The height of the plateau, and hence the amount of work
required to achieve the transition is higher for samples crosslinked deep in the nematic phase.
Samples crosslinked in the isotropic phase on the other hand display a much more pronounced
plateau, with an ideally soft behavior. We thus anticipate the polydomain to monodomain
transition to be a reversible process for samples crosslinked in the nematic phase and not
59
for those crosslinked in the isotropic phase. The slope of the stress-strain curve in the linear
regime (after the rotation of the domains is completed) however does not reveal any detail on
the history of the sample.
DL/L
Figure 27: Engineering stress σ ,(black curve) and Global order parmeter, S (red curve) vsstrain for a nematic elastomer crosslinked in the isotropic phase ( I-PNE) .
It can be seen from the range of strains over which the transition occurs that samples
crosslinked in the isotropic phase offer less resistance to the rotation of the domains than their
counterpart crosslinked in the nematic phase. One can thus think of LCE crosslinked in the
isotropic phase as having a weak anchoring of the director to the polymer matrix, and those
crosslinked in the nematic phase as having a strong anchoring. It is worth noting however
that the amount of order in the obtained monodomain state is roughly the same regardless of
the whether the sample was crosslinked in the isotropic or in the nematic phase.
60
DL/L
Figure 28: Engineering stress σ ,(black curve) and Global order parmeter, S (red curve) vsstrain for a nematic elastomer crosslinked in the nematic phase ( N-PNE) .
3.4 Discussion
This chapter presented simulations of the polydomain to monodomain transition in liq-
uid crystal nematic elastomers samples with different crosslinking histories. Studying these
materials at the continuum level, we used a model that makes no assumption on the detailed
chemical structure, and hence strive to extract universal characteristics of the response of liq-
uid crystal elastomers subjected to external stimuli. The mechanical response of polydomain
nematic elastomers has been investigated in the presence of an external mechanical stimulus,
taking into account the details of the sample preparation. We found that the thermomechan-
ical history of the sample plays a crucial role in determining the dynamics of the transition,
and the simulation results are in a good qualitative agreement with recent experiments on this
61
class of material [55].
CHAPTER 4
MODELING THE STRIPE INSTABILITY IN NEMATIC ELASTOMERS
4.1 Introduction
In the nematic phase, due to strong coupling between mechanical strain and orientational
order, nematic liquid crystal elastomers display strain-induced instabilities [10] associated
with formation and evolution of orientational domains. In a classic experiment, Kundler and
Finkelmann [63] measured the mechanical response of a monodomain nematic LCE thin film
stretched along an axis perpendicular to the nematic director. They observed a semisoft elastic
response with a pronounced plateau in the stress-strain curve arising at a threshold stress.
Accompanying this instability they observed the formation of striped orientational domains
with alternating sense of director rotation, and a stripe width of 15 µm. They repeated the
experiment with samples cut at different orientations to the director axis, and found that the
instability was absent when the angle between the initial director and the stretch axis was less
than 70o [63] ; in this geometry, instead of forming stripes, the director rotates smoothly as a
single domain.
This peculiar behavior reminiscent of a martensitic transformation, promises a bright fu-
ture to nematic liquid crystal elastomers for engineering applications based on soft actua-
tors [64]. An interesting application that was proposed was building an acoustic wave polar-
izer by modulating the internal degree of freedom, namely the nematic director.
Using the aforementioned 3-d finite element elastodynamics simulation, we investigate
62
63
Figure 29: Experiment: stretching a nematic elastomer film at an angle of 90o to the direc-tor results in a microstructure consisting stripes of alternating director orientation. Imagecourtesy H. Finkelmann
the onset of stripe formation in a monodomain film stretched along an axis not parallel to the
nematic director.
In an earlier work, DeSimone et al. [43] carried out numerical simulation studies of the
stripe instability using a two-dimensional finite element elastostatic method. Each area el-
ement in the system was considered as a composite of domains with different orientations.
This simulation model was the first to reproduce successfully the soft elastic response of
nematic elastomers, but did not attempt to resolve the resulting microstructural evolution.
Uchida carried out more detailed studies of director evolution in nematic elastomers using a
two-dimensional lattice model where macroscopic strain is treated as a global variable analo-
gous to an external field, but did not attempt to describe the non-uniform strain and resulting
shape evolution of the sample.
Here we explore this elastic instability in more detail by simultaneously modeling the
sample’s mechanical response, shape evolution, and the associated microstructural evolution
as a function of strain. We use a Hamiltonian-based 3-d finite element elastodynamics model
with terms that explicitly couple strain and nematic order. By resolving the finite element
64
mesh down to the micron scale, we resolve the formation of orientational domains, and be-
cause the model is dynamic rather than static in character, we can examine the effects of strain
rate. We use the simulation to explore the dependence of mechanical response on deformation
geometry.
4.2 Simulations and results
We model this instability in a thin film of nematic elastomer which has been cross-linked
in the nematic phase [65]. Using public domain meshing software [59] we discretize the
volume of the sample into approximately 78, 000 tetrahedral elements. For each volume
element we assign a local variable n that defines the nematic director, and Qij =12S(3ninj −
δij), which is the associated symmetric and traceless nematic order tensor. The initial state is
taken to be a monodomain with n = no in every element; this configuration is defined as the
system’s stress-free reference state.
There are many approaches to finite element simulation of the dynamics of elastic media
[66]; we make use of an elegant Hamiltonian approach developed by Broughton et al. [44,45],
generalizing it to three dimensions and the case of large rotations. We write the Hamiltonian
of an isotropic elastic solid as:
Helastic =∑p
Vp1
2Cijklε
pijε
pkl +
∑i
1
2miv
2i . (24)
Here the first term represents elastic strain energy, with p summing over volume elements. Vp
is the volume of element p in the reference state. For an isotropic material the components of
the elastic stiffness tensor Cijkl are determined from only two material parameters, namely the
65
shear and bulk moduli [67]. As an approximation, Broughton et al developed this formulation
using the linear strain tensor, but we instead use the rotationally invariant Green-Lagrange
strain tensor εij = 12(ui,j + uj,i + uk,iuk,j), where u is the displacement field. We note that
using the linearized strain tensor would make the Hamiltonian unphysical, as rotation of the
sample would appear to cost energy. The second term represents kinetic energy in the lumped
mass approximation [45] whereby the mass of each element is equally distributed among its
vertices, which are the nodes of the mesh. Here i sums over all nodes, mi is the effective
mass and vi the velocity of node i.
To account for the additional energy cost associated with the presence of a director field,
we add to the potential energy,
Hnematic =∑p
Vp
[−αεpij(Q
pij −Qrp
ij ) + β(Qpij −Qrp
ij )2]. (25)
The first term describes coupling between the strain and order parameter tensors using a form
proposed by DeGennes [16]. Here Qprij defines the nematic order in the element’s reference
state. The prefactor α controls the strength of this coupling, and DeGennes [16] argued that
it is of the same order of magnitude as the shear modulus µ. Variables Qij , Qprij , and εij
are all defined in the body frame, i.e. they are invariant under rotations in the target frame.
See [68] for the relation between Qij in the body and lab frames. The second term describes
“cross-link memory,” that is, the tendency of the nematic director to prefer its orientation at
crosslinking. Thus there is an energy cost to rotate the director away from its reference state,
with coupling strength β.
66
The strain tensor εij within each tetrahedral element is calculated in two steps. We calcu-
late the displacement u of each node from the reference state, then perform a linear interpola-
tion of the displacement field within the volume element in the reference state. The resulting
interpolation coefficients represent the derivatives ui,j needed to calculate the components of
the strain tensor. Details were described in chapter 2 and can be found in any introductory text
on finite element methods, e.g. [60]. At this level of approximation, the strain is piecewise
constant within each volume element. The effective force on each node is calculated as the
negative derivative of the potential energy with respect to node displacement.
To evolve the system forward in time, we assume the director is in quasistatic equilibrium
with the strain; that is, the time scale for director relaxation is much faster than that for strain
evolution as observed by Urayama [69]. The first part of each step is elastodynamics: holding
Qij in each element constant, the equations of motion f = ma for all node positions and
velocities are integrated forward in time using the Velocity Verlet algorithm [46], with a time
step of 10−8 sec. In the second part of each step, we relax the nematic director in each element
to instantaneously minimize the element’s potential energy. Because the director relaxes from
a higher energy state to a lower energy state without picking up conjugate momentum, this
is a source of anisotropic dissipation. Thus in our model, as in real nematic elastomers,
strains that rotate the director cause more energy dissipation than those applied parallel to the
director [70].
To add internal damping associated with velocity gradients in the sample, we use a mod-
ified form of Kelvin dissipation. In its standard form, the Kelvin dissipation force (e.g. be-
tween two particles, or between two nodes in a finite element mesh) is proportional to the
67
velocity difference between them (see e.g. [47].) This form conserves linear momentum but
violates conservation of angular momentum; internal dissipation forces could create torque,
which is of course unphysical. We modified the Kelvin dissipation form to provide for con-
servation of angular momentum, that is, dissipation forces between any pair of nodes must
act along the line of sight between them, so they create no torque [48]. We also scale the
dissipation force so it depends on the effective strain rate between two nodes rather than their
absolute velocity difference. With these modifications, the dissipation force between a pair of
neighboring nodes separated by distance d is F12 = −η(v1 − v2) · (r1 − r2)
d12r12. The result-
ing dissipation is isotropic in character and does not depend on the orientation of the director
field.
We simulate uniaxial stretching in an initially monodomain nematic elastomer film of
size 1.5 mm × 0.5 mm with a thickness of 50 µm, with shear modulus µ = 5.7 × 105Pa,
bulk modulus Br = 2.8 × 107Pa, and parameters α = µ, β = 0.3µ, and ζ = 10−7kg.m/s.
We first consider the case where the director is initially oriented along the y axis, transverse
to the direction of applied strain. The sample is clamped on two sides and the clamped
regions are constrained to move apart laterally at a constant speed of 1 mm/sec. The resulting
microstructural evolution is shown in Fig.30. Here color represents Jones matrix imaging
of the director field as viewed through crossed polarizers parallel to the x and y directions;
blue corresponds to a director parallel to the polarizer or analyzer, and red corresponds to a
director at a 45o angle to either. While the simulated sample is three-dimensional, the film’s
microstructure does not vary significantly through the thickness and can thus be visualized in
2-d.
68
Figure 30: Simulation: stretching a nematic elastomer film at an angle of 90o to the director.Initially a monodomain, the director field evolves to form a striped microstructure.
4.2.1 Stripes width
At a strain of 8.5%, the director field in the sample becomes unstable and orientational
domains form, nucleating first from the free edges of the film. Heterogeneity in the finite ele-
ment mesh serves to break the symmetry and nucleate the instability. By 9% strain, the whole
film is occupied by striped orientational domains with alternating sense of director rotation.
69
The stripes are not uniform in width, being slightly larger near the free edges. Near the center
of the sample, each individual stripe has a width of about 25 µm, which is of the same or-
der of magnitude as that observed in experiment [63]. This value is in reasonable agreement
with the theoretical estimate by Warner and Terentjev [10] who predicted a stripe width of
h ∼√ξL/
√1− 1/λ3
1; where ξ is the nematic penetration length , L is the sample width,
and λ1 is the strain threshold of the instability. Indeed , using ξ = 50nm, L = 1.5× 10−3m,
and λ1 = 1.09, one finds h ≈ 18.14µm. The stripes coarsen as the elongation increases.
Eventually this microstructure evolves into a more disordered state with stripes at multiple
orientations. By reaching 35% strain, the stripes have vanished and the film is again in a mon-
odomain state with the director oriented with the direction of strain. Only the regions near
the clamped edges do not fully realign, in agreement with experimental observations [63]
and with the simulation studies of DeSimone [43]. We will explore the dependence of stripe
width on aspect ratio and other parameters in future work.
The resulting stress-strain response is semi-soft [68] in character, as shown in Figure 31.
The initial elastic response is linear, followed by an extended plateau running from about
8.5% to over 30% strain, after which there is a second linear regime. We also measure the
average director rotation⟨sin2(ϕ)
⟩and observe that the thresholds for both the stress-strain
plateau and the rotation of the nematic director occur at the same strain. This finding demon-
strates, in agreement with theory [10,68], that the reorientation of the system’s internal degree
of freedom–namely the nematic director–reduces the energy cost of the deformation.
70
σ(
N/
mm
2)
0
0. 02
0. 04
0. 06
0. 08
0. 1
0. 12
0
0.2
0.4
0.6
0.8
1
DL/L
0 0 .1 0. 2 0 .3 0. 4 0 .5
S
S
S
Figure 31: Engineering stress (circles) and director rotation (squares) vs applied strain, forthe system shown in Figure 30. Onset of director rotation and the stress-strain plateau bothoccur at the same strain.
4.2.2 Threshold for stripe instability
We also performed simulations for monodomain nematic elastomer films with the initial
director orientation at different angles to the pulling direction. In Figure 32 we plot the film’s
stress-strain response when strain is applied at an angle of 60o from the nematic director,
which shows no plateau, and likewise director rotation shows no threshold behavior. As
shown in Figure 33, the director rotates to align with the strain direction without forming
stripes. We performed additional simulations with the director at angles of 70o and 80o to
the pulling direction and again found no stripe formation and no plateau in the stress-strain
response.
71
σ(
N/
mm
2)
0
0. 02
0. 04
0. 06
0. 08
0. 1
0. 12
0
0. 2
0. 4
0. 6
0. 8
1
DL/L0 0 .1 0. 2 0 .3 0. 4 0 .5
S
S
Figure 32: Engineering stress (circles) and director rotation (squares) vs applied strain, ap-plied at an angle of 60o from the nematic director.
4.2.3 Rate of strain
Because liquid crystal elastomers have relatively slow stress relaxation [71], mechanical
experiments are often performed using static strains, i.e. the sample is allowed to relax be-
tween successive elongations. However, these static strain experiment may not accurately
reflect the behaviour of these materials when used in applications. It is thus paramount to
study the rate dependence of the strain of the material’s response. We tried varying the
applied strain rate. Figure34 compares the stress-strain response for samples strained at 1
mm/sec and 5 mm/sec. The higher strain rate produces a significant stress overshoot, and
stripe formation occurs at a strain of 15%. This finding suggests that the threshold strain for
the instability depends in a significant way on strain rate.
72
Figure 33: Simulation: stretching a nematic elastomer film at an angle of 60o to the direc-tor. Initially a monodomain, the director field rotates smoothly without sharp gradients inorientation.
4.3 Discussion
The simulations presented here were performed at far higher strain rates, e.g. 50% per
second, than those used in typical experiments [62,63] where the material is allowed to relax
for minutes or hours between strain increments. In future work we plan to apply our model to
examine deformation of nematic elastomers at slower strain rates and as a function of sample
73
(
Figure 34: Dependence of the stress-strain response on strain rate.
geometry. We will also examine the role of initial microstructure and thermomechanical
history in determining mechanical response. Using the same finite element approach, we
can also test the predictions of other proposed constitutive models, and model geometries
of interest for potential applications. Through this approach we hope to bridge the divide
between fundamental theory of these fascinating materials and engineering design of devices.
CHAPTER 5
MODELING DEVICES
The novel electrical, optical and mechanical properties of liquid crystal elastomers allow
one to envision a vast field of possible engineering applications. The method we have intro-
duced for studying the response of these materials to external stimuli also allows us to model
devices, thus making a step towards bridging our understanding of the fundamental micro-
scopic properties of LCE and their applications. In this chapter, a number of examples of
applications of nematic liquid crystal elastomers are presented, ranging from soft photoactu-
ators to robotic earthworms.
5.1 Polarization tuner
The basic principle of operation of a wave plate consists in modifying the state of po-
larization of the light incident upon it by changing the relative phase of it’s ordinary and
extraordinary waves. The accepted taxonomy refers to a wave plate as λn
wave plate if it re-
tards one state of polarization by πn
with respect to the other. Here n is an integer. The most
commonly used are the half wave plate and the quarter wave plate. Observe that the quarter
wave plate will change the polarization state of light from linear to circular, and vice-versa.
To explore mechanical response of nematic elastomer films in a more complex geometry, we
simulated the radial stretching of a circular monodomain film of diameter 1 cm and thick-
ness 100 µm, with the nematic director oriented initially along the y axis, as indicated by
the arrow in Figure 35. Boundary conditions were imposed that clamp the sample around74
75
Figure 35: Simulation: A nematic elastomer disk is stretched radially. The director fieldsmoothly transforms from a homogeneous monodomain to a radial configuration.
its circumference and stretch radially in all directions, pulling the edge outward at constant
speed. Figure 35 shows the film at different stages of its extension, demonstrating that the
director field smoothly changes from a monodomain to a radial configuration, with no stripe
instability.
With a careful choice of the sample’s thickness, this deformed circular sheet of nematic
elastomer could be used as a tunable spatial polarization converter as described in [72].
5.2 Actuators
The photoresponse of nematic elastomers was extensively discussed by Warner and Ter-
entjev [10]. Camacho-Lopez et al. demonstrated that an azo-doped liquid crystal elastomer
beam anchored on one edge bends spontaneously when illuminated on one side by a fast
laser pulse. Our simulations permit us to model the mechanical response of such a system. In
76
Figure 36: Cartoon of the polarization modulation by the elastomer strip.
particular, we consider a nematic elastomer beam in its initial nematic state with its director
aligned horizontally as in Figure 37. Then we switch the top layer of the sample, shown in
red, from nematic to isotropic, by setting the nematic order tensor Qij to zero only within
volume elements on the top of the sample. This mimics the fact that the light is rapidly at-
tenuated as it goes through the material. The resulting surface contraction induces a rapid
bend, pulling the beam into a curved position. The resulting radius of curvature depends on
many variables including the thickness and elastic properties of the beam; the strength of the
nematic-strain coupling; the magnitude of the reduction in the nematic order parameter in the
surface layer; and the thickness of the surface layer. The time response of the beam depends
also on the kinetics of the trans-cis photoisomerization of azobenzene chromophores in the
material.
Ikeda has studied the photoresponse of sheets of nematic elastomers illuminated by UV
light. In his experiments, he was able to control the folding of the sheet by simply rotating the
77
Figure 37: Simulation of the photo-deformation of a beam of nematic liquid crystal elastomer.
polarization of the incident (linearly) polarized light. We have obtained similar results from
our simulations with sheets of polydomain nematic elastomers that are illuminated by a beam
of linearly polarized light. See Figure 38. In these simulations, as in the previous case, only
a thin portion of the film responds to the incident light. The difference between these and the
previous simulations is that the domains on the top layer of the film do not all contract by
the same amount, but by an amount proportional to the angle between the local director and
polarization state of the incident light.
5.3 Peristaltic pumps
Because of their exceptional ability to change shape and mimic the behavior of muscles,
nematic elastomers are a good candidate for the engineering of peristaltic pumps. Peristalsis
is the process by which propagating undulation in a tube or channel induces motion of its
contents. This is for example the mechanism by which food is transported through the human
digestive system. We have modeled two highly idealized conceptual designs for peristaltic
pumps composed of nematic elastomers [61]. The first of these is a tube-shaped structure
78
Figure 38: Simulation of the photobending of a strip of polydomain nematic liquid crystalelastomer.
79
with the nematic director oriented along the tubes long axis. To induce a propagating wave,
we apply a modulation to the strength of the scalar nematic order parameter along the length
of the tube with a selected wavelength and frequency. Such modulation could be created
e.g. by non-uniform heating or by a pattern of laser illumination switched on/off periodically
along the tube to create a moving wave. Such a device might be useful for transport of highly
viscous fluids or slurries. A second configuration is also shown; here we induce a similar
propagating oscillation in a thin film designed to cover a rigid channel and move the fluid
inside. Alternatively a pair of such films might be used on opposite sides of a channel.
Figure 39: Simulation of a soft peristalsis tube made of nematic liquid crystal elastomer.
5.4 Self-propelled earthworm
Earthworms move by a propagating wave of muscle contraction, alternately shortening
and lengthening the body along its length. This motion can be replicated in a nematic elas-
tomer by applying a modulation in the magnitude of the nematic order parameter, much as
80
in the pumps modeled above. To enable motion of the center of mass, an earthworm also
needs friction with a nearby surface. The underside of a real earthworm is decorated with
tiny bristles (setae) that allow it to anchor one section of the body while another section is in
motion. To mimic this behavior we could also imagine placing bristles on the underside of a
nematic elastomer earthworm robot. Alternatively, asymmetric static friction can be created
by suitable surface morphology on the substrate. For instance, an elongated monodomain
nematic elastomer film can be made to crawl by placing it on an asymmetric substrate, e.g.
on brushed velvet. Local contraction and elongation can be induced by moving a heat source
just above the film from one end to the other. In the simulation, shown in Video 1, we ap-
ply perfect static friction so that the simulated earthworm cannot slide backwards. Kinetic
friction is also included with a finite value of the friction coefficient.
81
Figure 40: Simulation of a nematic elastomer robotic earthworm moving on a rugged surfaceshaped like a size wave, with height z = A sin(kx).
CHAPTER 6
CONCLUSION
Coupling between strain and orientational order in nematic liquid crystal elastomers can
be observed through a variety of experiments. When an LCE film is subject to an external
mechanical strain, it breaks down into orientational domains of alternating director orienta-
tion. The resulting orientational domain patterns can be observed either optically or using
x-ray techniques [10]. The director reorientation is shown to compensate the energy cost of
the elastic deformation, and is visible in the stress-strain response of the material in the form
of a plateau. The spontaneous change in shape of a strip of liquid crystal elastomer near the
nematic-isotropic phase transition is also an evidence of this coupling [12]. A liquid crys-
tal elastomer doped with azo dyes will deform upon illumination by light of an appropriate
wavelength as a result of the conformation change of the azo dyes that tend to reorient the
liquid crystalline molecules, and hence distort the network strands [20, 64, 73].
We presented simulations of shape change and microstructural evolution in nematic elas-
tomer films using 3-d finite element elastodynamics. Studying these materials at the contin-
uum level, we derived a simulation model that makes no assumption on the detailed chemical
structure, and hence attempts to probe universal characteristics of the response of liquid crys-
tal elastomers subjected to external stimuli. In this dissertation, the mechanical response of
nematic LCE has been investigated in the presence of an externally applied uniaxial strain,
and the simulations results are in a good qualitative agreement with experiments on this class
82
83
of material [11, 55]. This work aimed at bridging the divide between the fundamental micro-
scopic theory of nematic liquid crystal elastomers and their applications.
Clearly, the results showed here are in no way an exhaustive presentation of the contribu-
tion that can be expected of the methods we have developed. There are several avenues that
remain to be studied using our method. As an example, the interaction of nematic elastomers
with an electric field can be modeled by modifying the free energy density to account for the
interaction between a nematic director and an electric field. Similarly, the flexoelectric effect
and the converse flexoelectric effect in NLCE [35] are readily available for simulation. A
minor variation to the nematic interaction part of our free energy density will also allow one
to study buckling instability in cholesteric liquid crystal elastomers [74].
6.1 Future work
6.1.1 Role of the Frank-Oseen elastic energy
It can be seen from the preceding chapters that spatial variations of the orientation of
the nematic director were not penalized by in the simulation reported. At first glance, this
seem to pose a strong limitation to the strength of this method we employ for modeling LCE.
However, one must expect this contribution to the energy to be negligibly small, i.e. several
orders of magnitude smaller than all others. This expectation is motivated by the fact that
the energy scales involved differ considerably. A characteristic length scale that emerges in
nematic elastomers is the nematic penetration length ξ =√
K/µ ≈ 10−8m, using K ≈
10−11N and µ ≈ 105Pa. It is a rough estimate of how far perturbations of the orientational
order would persist in a competition between Frank elasticity and that of the rubbery network.
The natural question is whether the smallness of this term justifies its omission from most
84
reported studies of the deformation of LCE (Desimone, Warner, Kundler); is it sufficient to
break a symmetry? e.g. can its presence shift the threshold for stripe instability? While such
a contribution is expected to be of very little importance in the mechanical response of these
soft materials [52], we outline here how it will be accounted for in future work. In the one
elastic constant approximation, one can write the energy penalty for spatial variations of the
director orientation as
Unem =K
2
∫dV Qij,kQij,k (26)
Upon discretization, it approximates to
Unem ≈ 1
2γ
N∑p=1
nb∑q=1
h3
(Qp
ij −Qqij
h
)2
(27)
where h is the element length, N is the total number of elements, nb is the number of neigh-
bours of the pth element, and γ is the Frank elastic constant. In the current formulation of
the coupling between order and strains, the director is defined at the centroid of the element,
whereas the strains are obtained from the displacement of the nodes, which in turn are in-
terpolated within the element. In other words, strains are piecewise-linear and the nematic
order tensor Qij is piecewise-constant throughout the mesh. In order for the gradients of the
order parameter tensor to be defined, one would need to have a node-centered definition of
Qij , which will be piecewise-linear, just as the strain tensor. In our two-step dynamics, given
an increment in displacement and hence a strain tensor ε∗ij , the relaxation of the directors will
85
consist in minimizing the energy.
0 =N∑p=1
∂Utotal
∂Qpij
(28)
=N∑p=1
[−αε∗ij + β(Qp
ij −Q0ij) + γ
nb∑q=1
h3
(Qp
ij −Qqij
h2
)]
Which is a system of N equations to be solved simultaneously for the Qp. This effectively
means that our algorithm will be of the form :
1- Initialize the system
2- Apply a deformation, find the local strains ε∗ij .
3- Solve for the local Qij as shown above.
4- Go to 2.
6.1.2 Volumetric change (swelling) of LCE
The stimulus-induced shape change of elastomeric materials opens door to a wide range
of engineering applications. Heretofore, efforts have been mainly devoted to obtaining three-
dimensional structures from the imposition of a two-dimensional pattern on an isotropic ma-
terial [75,76]. The reverse problem has been approached as well by Dias et al. [77] who made
a demonstration of how one finds the growth pattern that generates a certain shape. Here we
intend to consider a different class of materials, and show that additional frustrations occur
which might preclude the attainment of the desired shape, or enhance some saillant features
of the otained shape.
86
A thin strip of elastomer with a nonuniform crosslink density will, upon exposure to an exter-
nal stimulus such as a solvent, undergo nonuniform deformations. Predicting the state of the
deformed material becomes even more arduous when the polymer strands have an anisotropic
shape of gyration as is the case in nematic elastomers. When a nematic elastomer is put in
the presence of a solvent, it undergoes a volumetric change that depends on both the sym-
metry of the solvent, and the state of the elastomer. Experiments [27, 78–80] have shown
that monodomain nematic elastomers, when embedded in an isotropic solvent elongate along
the nematic director, and shrink in the transverse directions. This anisotropic swelling is not
observed in polydomain samples. Other studies have focused on the mechanical and elec-
trooptical properties of nematic elastomers swelled in the presence of a constraint such as a
nematic solvent, or an aligning field [81]. The orientation of the liquid crystal mesogens was
shown to play a role in the selection of bending directions in swollen nematic elastomers [82].
We intend to study the case in which the pattern of the crosslinks defines a metric, and hence
a desired shape upon relaxation. Starting from the phenomenological formalism of the strain-
order coupling presented in this work, we will investigate how the presence of the nematic de-
gree of freedom frustrates the shape selection expected for the case of non-uniformly swelled
isotropic elastomer. Clearly, a first, more accessible question will be that of finding the equi-
librium shape of a uniformly swelled nematic elastomer with a patterned nematic director. We
will also study how morphological transitions may occur by altering the pitch of a swelled
cholesteric liquid crystal elastomer, e.g. by prescribing a swelling pattern corresponding to
an unduloid whose eccentricity is a function of the cholesteric pitch.
BIBLIOGRAPHY
[1] F. Reinitzer, Monatsch Chem., vol. 9, p. 421, 1888.
[2] Z. Lehmann, Z. Physikal Chem., vol. 4, p. 462, 1889.
[3] M. J. Bowick, L. Chandar, E. A. Schiff, and A. M. Srivastava, Science, vol. 263, no.5149, pp. 943–945, 1994.
[4] I. Chuang, R. Durrer, N. Turok, and B. Yurke, Science, vol. 251, no. 4999, pp. 1336–1342, 1991.
[5] S. Chandrasekhar and G. Ranganath, Rep. Prog. Phys., vol. 53, p. 57, 1990.
[6] S. Chandrasekhar, Liquid Crystals. Cambridge University Press, 1977.
[7] P. G. de Gennes and J. Prost, The Physics of Liquid Crystals (International Series ofMonographs on Physics), 2nd ed. Oxford University Press, USA, 1995.
[8] P. J. Collings and M. Hird, Introduction to Liquid Crystals: Chemistry and Physics.Taylor and Francis, 1997.
[9] M. J. Stephen and J. P. Straley, “Physics of liquid crystals,” Rev. Mod. Phys., vol. 46,pp. 617–704, Oct 1974.
[10] M. Warner and E. Terentjev, Liquid Crystal Elastomers. Oxford Science Publications,2003.
[11] H. P. H. R. Brand and P. Martinoty, Soft Matter, vol. 2, p. 182, 2006.
[12] J. Naciri, A. Srinivasan, H. Jeon, N. Nikolov, P. Keller, and B. R. Ratna, Macro-molecules, vol. 36, pp. 5868–5875., 2003.
[13] F. Ye, R. Mukhopadhyay, O. Stenull, and T. C. Lubensky, Phys. Rev. Lett., vol. 98,no. 14, p. 147801, 2007.
[14] H. Bengs, H. Finkelmann, J. Kpfer, H. Ringsdorf, and P. Schuhmacher, Makromol.Chem., Rapid Commun., vol. 14, p. 445, 1993.
[15] H. Finkelmann, H. Kock, and G. Rehage, Makromol. Chem. Rapid Commun., vol. 2, p.317, 1981.
[16] P. G. De Gennes, C. R. Acad. Sci. B, vol. 281, p. 101, 1975.
[17] H. Finkelmann, H. Benthack, and G. Rehage, 27th Intern. Symposium on Macro-molecules, vol. 2, p. 961, 1981.
87
88
[18] J. V. Selinger, H. G. Jeon, and B. R. Ratna, Phys. Rev. Lett., vol. 89, p. 225701, 2002.
[19] H. Finkelmann, E. Nishikawa, G. G. Pereira, and M. Warner, Phys. Rev. Lett., vol. 87,no. 1, p. 015501, 2001.
[20] M. C. Lopez, H. Finkelmann, P. Palffy-Muhoray, and M. Shelley, Nature Matter, vol. 3,p. 307, 2004.
[21] D. Corbett and M. Warner, Phys. Rev. Lett., vol. 96, no. 23, p. 237802, 2006.
[22] L. H. He, Phys. Rev. E, vol. 75, no. 4, p. 041702, 2007.
[23] D. J. Burridge, Y. Mao, and M. Warner, Phys. Rev. E, vol. 74, no. 2, p. 021708, 2006.
[24] A. M. Menzel and H. R. Brand, Phys. Rev. E, vol. 75, no. 1, p. 011707, 2007.
[25] E. M. Terentjev, M. Warner, R. B. Meyer, and J. Yamamoto, Phys. Rev. E, vol. 60, no. 2,pp. 1872–1879, 1999.
[26] C.-C. Chang, L.-C. Chien, and R. B. Meyer, Phys. Rev. E, vol. 56, no. 1, pp. 595–599,1997.
[27] K. Urayama, Y. O. Arai, and T. Takigawa, Macromolecules, vol. 38, no. 8, pp. 3469–3474, 2005.
[28] R. B. Meyer, Phys. Rev. Lett., vol. 22, no. 18, pp. 918–921, May 1969.
[29] J. Harden, B. Mbanga, N. Eber, K. Fodor-Csorba, S. Sprunt, J. T. Gleeson, and A. Jakli,Phys. Rev. Lett., vol. 97, no. 15, p. 157802, Oct 2006.
[30] A. I. Derzhanski and H. P. Hinov, Physics Letters A, vol. 62, no. 1, pp. 36 – 38, 1977.
[31] W. Helfrich, Physics Letters A, vol. 35, no. 6, pp. 393 – 394, 1971.
[32] A. Derzhanski and A. G. Petrov, Physics Letters A, vol. 36, no. 6, pp. 483 – 484, 1971.
[33] L. M. Blinov, M. I. Barnik, H. Ohoka, M. Ozaki, and K. Yoshino, Phys. Rev. E, vol. 64,no. 3, p. 031707, Aug 2001.
[34] S. Dhakal and J. V. Selinger, Phys. Rev. E, vol. 81, no. 3, p. 031704, Mar 2010.
[35] J. Harden, M. Chambers, R. Verduzco, P. Luchette, J. T. Gleeson, S. Sprunt, and A. Jakli,Applied Physics Letters, vol. 96, no. 10, p. 102907, 2010.
[36] S. M. Clarke, A. R. Tajbakhsh, E. M. Terentjev, and M. Warner, Phys. Rev. Lett., vol. 86,pp. 4044–4047, Apr 2001.
[37] S. Conti, A. DeSimone, and G. Dolzmann, Phys. Rev. E, vol. 66, no. 6, p. 061710, 2002.
[38] P. I. C. Teixeira and M. Warner, Phys. Rev. E, vol. 60, no. 1, pp. 603–609, 1999.
89
[39] J. S. Biggins, E. M. Terentjev, and M. Warner, Phys. Rev. E, vol. 78, no. 4, p. 041704,2008.
[40] J. M. Adams and M. Warner, Phy. Rev. E, vol. 72, no. 1, p. 011703, 2005.
[41] O. Stenull and T. C. Lubensky, Phys. Rev. E, vol. 69, no. 5, 2004.
[42] G. C. Vervey and M. Warner, Macromolecules, vol. 28, p. 4303, 1995.
[43] S. Conti, A. DeSimone, and G. Dolzmann, Phys. Rev. E, vol. 66, no. 6, p. 061710, 2002.
[44] J. Q. Broughton, F. F. Abraham, N. Bernstein, and E. Kaxiras, Phys. Rev. B, vol. 60,no. 4, pp. 2391–2403, Jul 1999.
[45] R. E. Rudd and J. Q. Broughton, Phys. Rev. B, vol. 58, no. 10, pp. R5893–R5896, 1998.
[46] M. Allen and D. Tildesley, Computer Simulations of Liquids. Clarendon Press, Oxford,1987.
[47] M. Marder, Journal of the Mechanics and Physics of Solids, vol. 54, pp. 491–532, 2006.
[48] E. L. Wilson, Three Dimensional Static and Dynamic Analysis of Structures. Comput-ers and Structures, Inc., 1995.
[49] X. Xing, R. Mukhopadhyay, T. C. Lubensky, and L. Radzihovsky, Phys. Rev. E, vol. 68,no. 2, p. 021108, 2003.
[50] K. Urayama, S. Honda, and T. Takigawa, Phys. Rev. E, vol. 74, no. 4, p. 041709, 2006.
[51] N. Uchida and A. Onuki, Europhys. Lett., vol. 45, no. 3, pp. 341–347, 1999.
[52] S. V. Fridrikh and E. M. Terentjev, Phys. Rev. E, vol. 60, p. 1847, 1999.
[53] N. Uchida, Phys. Rev. E, vol. 62, no. 4, pp. 5119–5136, 2000.
[54] I. K. S. M. Clarke, E. M. Terentjev and H. Finkelmann, Macromolecules, vol. 31, p.4862, 1998.
[55] M. K. K. Urayama, E. Kohmon and T. Takigawa, Macromolecules, 2009.
[56] N. Uchida, Phys. Rev. E, vol. 60, no. 1, pp. R13–R16, 1999.
[57] Y.-K. Yu, P. L. Taylor, and E. M. Terentjev, Phys. Rev. Lett., vol. 81, no. 1, pp. 128–131,1998.
[58] J. V. Selinger and B. R. Ratna, Phys. Rev. E, vol. 70, no. 4, p. 041707, Oct 2004.
[59] Documentation and software available at http://www.salome-platform.org/.
[60] W. B. Zimmerman, Process Modelling and Simulation with Finite Element Methods.World Scientific, 2004.
90
[61] R. Selinger, B. Mbanga, and J. Selinger, Proceedings-SPIE the International Society forOptical Engineering, vol. 6911, p. 6911, 2008.
[62] S. M. Clarke and E. M. Terentjev, Phys. Rev. Lett., vol. 81, p. 4436, 1998.
[63] I. Kundler and H. Finkelmann, Macromol. Rapid Commun., vol. 16, p. 679, 1995.
[64] Y. Yu, M. Nakano, and T. Ikeda, Nature, vol. 425, p. 145, 2002.
[65] B. L. Mbanga, F. Ye, J. V. Selinger, and R. L. B. Selinger, Phys. Rev. E, vol. 82, p.051701, Nov 2010.
[66] T. Belytschko, W. K. Liu, and B. Moran, Nonlinear Finite Elements for Continua andStructures. Wiley, 2000.
[67] M. P. Marder, Condensed Matter Physics. Wiley, 2000.
[68] O. Stenull and T. C. Lubensky, Phys. Rev. E, vol. 69, no. 2, p. 021807, 2004.
[69] K. Urayama, S. Honda, and T. Takigawa, Phys. Rev. E, vol. 74, no. 4, p. 041709, 2006.
[70] E. M. Terentjev, I. V. Kamotski, D. D. Zakharov, and L. J. Fradkin, Phys. Rev. E, vol. 66,no. 5, p. 052701, 2002.
[71] S. M. Clarke and E. M. Terentjev, Phys. Rev. Lett., vol. 81, no. 20, pp. 4436–4439, 1998.
[72] Y.-H. Wu, Y.-H. Lin, H. Ren, X. Nie, J.-H. Lee, and S.-T. Wu, Optics Express, vol. 13,p. 4638, 2005.
[73] H. Finkelmann, E. Nishikawa, G. G. Pereira, and M. Warner, Phys. Rev. Lett., vol. 87,p. 015501, 2001.
[74] R. Verduzco, G. Meng, J. A. Kornfield, and R. B. Meyer, Phys. Rev. Lett., vol. 96, no. 14,p. 147802, Apr 2006.
[75] E. Efrati, E. Sharon, and R. Kupferman, Journal of the Mechanics and Physics of Solids,vol. 57, no. 4, pp. 762 – 775, 2009.
[76] E. Sharon and E. Efrati, “The mechanics of non-euclidean plates,” Soft Matter, vol. 6,pp. 5693–5704, 2010.
[77] M. A. Dias, J. A. Hanna, and C. D. Santangelo, Phys. Rev. E, vol. 84, p. 036603, Sep2011.
[78] K. Urayama, Y. Okuno, and S. Kohjiya, Macromolecules, vol. 36, no. 16, pp. 6229–6234, 2003.
[79] K. Urayama, S. Honda, and T. Takigawa, Macromolecules, vol. 39, no. 5, pp. 1943–1949, 2006.
[80] K. Urayama, R. Mashita, Y. O. Arai, and T. Takigawa, Macromolecules, vol. 39, no. 24,pp. 8511–8516, 2006.
91
[81] A. Fukunaga, K. Urayama, T. Takigawa, A. DeSimone, and L. Teresi, Macromolecules,vol. 41, no. 23, pp. 9389–9396, 2008.
[82] Y. Sawa, K. Urayama, T. Takigawa, A. DeSimone, and L. Teresi, Macromolecules,vol. 43, no. 9, pp. 4362–4369, 2010.
92
Appendix 1 : Fortran codes
The following items will also be available on www.e-lc.org
Fortran codes.
program main
call setconfig call forcecalc call pullforce
call initial call dynamics
call closeall
end
93
parameter(ntet=10422) parameter(ntri=999) parameter(np=3193) integer tetlist(ntet,4) integer npick(4),ntetnab(ntet) integer itime, pref, prefa integer npairs,pullme(np), tetnab(ntet,4) integer ione(100000),itwo(100000) integer myrank, numprocs, ierr, tag, myid! integer status(MPI_STATUS_SIZE) real*8 Rot(3,3),centcent(ntet,4) real*8 k1, r, thet1, dir(ntet),ax,bx,cx,dx
c damping parameter(gamma=0.0) parameter(r=1.5) parameter(k1=0.0001) parameter(thet0=0.0)
parameter(d_0 = 1.0e-5)! parameter(thet1=0.0) parameter(dt=1.0E-7) parameter(dt2o2=dt*dt*0.5d0) parameter(dto2=dt*0.5d0) parameter(irealtime=int(2.0/dt))
! parameter(cxxyy=2.79530E+02)! parameter(cxyxy=5.0E+5)! parameter(cxxxx= Cxxyy + 2.0*Cxyxy)! parameter(cxxxx=100.0*Cxyxy)! parameter(cxxyy=1.0*Cxyxy)
parameter(alpha =0.1)! parameter(bet = 1.1*alpha / 4.0) parameter(mu = 1.0)
parameter(rho=1500.0)
character filename1*16 character filename2*16 character filename3*16 character filename4*16 character filename5*16
integer:: istep, nsteps, tetmax(ntet)
character(len=80) filename, frame
real*8 x(np),y(np),z(np), mass(np) real*8 vx(np),vy(np),vz(np) real*8 fx(np),fy(np),fz(np) real*8 fx1(np),fy1(np),fz1(np) real*8 x0(np),y0(np),z0(np) real*8 fxold(np),fyold(np),fzold(np) real*8 st(3,3), xtime
real*8 u(4),v(4),w(4), theta, phi real*8 a(4),b(4),c(4) real*8 eps(3,3) , lambda(3,3), lam(3,3,ntet) real*8 q(3,3),qfac, qloc(ntet),qlocal real*8 q0(3,3), q0field(3,3), qnab(3,3), q0sav(3,3,ntet)
94
real*8 fxsum, fysum,fzsum, vtemp(ntet), vol_n(np), mass_n(np) real*8 xm(ntet,4,4),x00(ntet), y00(ntet), z00(ntet) real*8 xmin,xmax,ymin,ymax,zmin,zmax,xlength,ylength,zlength real*8 etot,petot,pe1,pe2,pe3,ketot,pxtot,pytot,pztot real*8 nx(ntet),ny(ntet), nz(ntet), vec(3,3) real*8 nx0(ntet),ny0(ntet), nz0(ntet) real*8 fztot, fytot real*8 xlength0,ylength0,zlength0 character*4 job
95
common x,y,z,vx,vy,vz,x0,y0,z0,u,v,w,a,b,c,fx,fy,fz,fx1,fy1,fz1, & fxold,fyold,fzold,mass,x00,y00,z00,dir,ntetnab,nx,ny,nz,nx0, & ny0,nz0,pi1,lam,lambda,eps,q,qfac,qloc,vtemp,vol_init, & vol_now,vol_n,mass_n,q0sav,q0,q0field, & qnab,rot,xm,fxsum, fysum,fzsum,theta,phi,thet1, & energy,etot,petot,ketot,pxtot,pytot,pztot,xtime,pe, & xmin,xmax,ymin,ymax,zmin,fytot,zmax,vmax,ax,bx,cx,dx, & xlength,ylength,zlength,xlength0,ylength0,zlength0,qlocal, & tetlist,tetnab,centcent,itime,npairs,ione,itwo,pullme, & irun, pref, myrank,numprocs,ierr,tag,myid, & prefa,nb,ibond,filename1,filename2, & filename3,filename4,filename5
96
subroutine setconfig include 'test-inits1.f'! include 'test-inits2.f' integer lda,nn,iflag real*8 aa(4,4),aasave(4,4),bb(4,4) include 'test-inits2.f'
vmax = 1.0e-2
pi1=4*atan(1.0)fytot = 0.0
filename1='theta2.dat'filename2='pic2.xyz'filename3='st-st2.dat'filename4='dir2.dat'filename5='vol2.dat'
open(unit=37,file=filename1,status='unknown')open(unit=28,file=filename2,status='unknown')open(unit=18,file=filename3,status='unknown')open(unit=31,file=filename4,status='unknown')open(unit=117,file=filename5,status='unknown')
lda=4 nn=4 iflag=0
open(unit=41,file='nodes.dat',status='old')
do 2 i=1,np read(41,*)pref,x(i),y(i),z(i) 2 continue close(unit=41)
open(unit=42,file='tetras.dat',status='old') do 4 itet=1,ntet read(42,*)tetlist(itet,1),tetlist(itet,2), & tetlist(itet,3),tetlist(itet,4)
dir(itet) = thet0 4 continue close(unit=42)
call maxmin
xlength0=xmax-xminylength0=ymax-yminzlength0=zmax-zmin
ylength=ylength0
! if(myid ==0) then write(6,*)'in setconfig xmin-max:',xmin,xmax
write(6,*)'in setconfig ymin-max:',ymin,ymaxwrite(6,*)'in setconfig zmin-max:',zmin,zmax
! endif
97
do 608 itet=1,ntet x00(itet)=0.d0 y00(itet)=0.d0 z00(itet)=0.d0 do 609 i=1,4 ip=tetlist(itet,i) x00(itet)=x00(itet)+x(ip) y00(itet)=y00(itet)+y(ip) z00(itet)=z00(itet)+z(ip) 609 continue x00(itet)=x00(itet)/4.d0 y00(itet)=y00(itet)/4.d0 z00(itet)=z00(itet)/4.d0 608 continue
call nbors
do 5 i=1,np pullme(i)=0 if(y(i).eq.ymin)pullme(i)=-1 if(y(i).eq.ymax)pullme(i)=1
5 continue
c generate aa(i,j) matrix and invert it
do 10 itet=1,ntet do 11 i=1,4 ip=tetlist(itet,i) aa(i,1)=1. aa(i,2)=x(ip) aa(i,3)=y(ip) aa(i,4)=z(ip) 11 continue
do 112 i=1,4 do 13 j=1,4 aasave(i,j)=aa(i,j) 13 continue 112 continue
call matinv(aa,lda,nn,iflag)
do 20 i=1,4 do 21 j=1,4 xm(itet,i,j)=aa(i,j) 21 continue 20 continue
10 continue
do 12 i=1,np x0(i)=x(i) y0(i)=y(i) z0(i)=z(i) vx(i)=0.d0 vy(i)=0.d0
98
vz(i)=0.d0 12 continue
c generate a list of nearest neighbor pairs to usec in the damping calculation
npairs=0! if(myid ==0) write(6,*)'finding pairs'
ihaveit=0
if(ihaveit.eq.0)then
do 240 itet=1,ntet
if(mod(itet,500).eq.0)write(6,*)itet
ip1=tetlist(itet,1) ip2=tetlist(itet,2) ip3=tetlist(itet,3) ip4=tetlist(itet,4)
call checkmatch(ip1,ip2) call checkmatch(ip1,ip3) call checkmatch(ip1,ip4) call checkmatch(ip2,ip3) call checkmatch(ip2,ip4) call checkmatch(ip3,ip4) 240 continue
! write(6,*)'we have ',npairs,' unique near nbor node pairs'
open(unit=50,file='ioneitwo.dat',status='unknown')
write(50,*)npairs do 51 ip=1,npairs write(50,*)ip,ione(ip),itwo(ip) 51 continue close(unit=50)
else open(unit=50,file='ioneitwo.dat',status='old') read(50,*)npairs do 151 ip=1,npairs read(50,*)kk,ione(ip),itwo(ip) 151 continue close(unit=50)! write(6,*)'file read successfully'! write(6,*)'we have ',npairs,' unique near nbor node pairs' endif
do 601 i=1,npmass(i)=0.0
99
601 continue vol0=0.d0
vol_init =0.0do 222 itet=1,ntet
call calcvol0(itet,vol0)vol_init = vol_init + vol0
do 602 i=1,4mass(tetlist(itet,i))=mass(tetlist(itet,i)) + 0.25*vol0*rho
602 continue
vtemp(itet)=vol0222 continue
pysum=0.d0 totmass=0.d0
xcom =0.5*(xmax - xmin)ycom =0.5*(ymax - ymin)zcom =0.5*(zmax - zmin)
do 672 i = 1, npvy(i) = vmax*(y(i)-ycom)/ycom
pysum=pysum+vy(i)*mass(i) totmass=totmass+mass(i)672 continue
write(6,*)'totmass=',totmass write(6,*)'pysum=',pysum
write(6,*)'Vol_init',vol_init
vzave=pysum/totmass do 673 i=1,np vy(i)=vy(i)-vzave 673 continue
return end
100
subroutine initialinclude 'test-inits1.f'
include 'test-inits2.f'
strain=100*(ylength-ylength0)/ylength0 write(18,*) strain, -fytot write(6,*)itime,' strain=',strain,' force= ',-fytot
write(28,*)npwrite(28,*)do i=1,npwrite(28,*)'A',x(i),y(i),z(i)
enddo
vol_now=0.d0
do itet=1,ntet call volchange(itet,vol0)
vol_now = vol_now + vol0enddo
del_vol = 100*(vol_now-vol_init)/vol_initwrite(*,*) 'volume change',del_vol,'percent'write(117,*) strain, del_vol
ketot = 0.0do i=1, npketot = ketot + 0.5*mass(i)*(vx(i)**2 + vy(i)**2 + vz(i)**2)enddo
write(*,*) itime,petot, ketot, petot + ketot
write(6,*)itime,' strain=',strain,' force=',-fytot write(6,*)'initial state has potential energy',petot
xtime=0.d0
returnend
101
subroutine nbors include 'test-inits1.f' include 'test-inits2.f'
write(*,*) 'in nbors' open(unit=47, file='tetnab', status='unknown')
open(unit=48, file='ntetnab', status='unknown')
do i=1,ntet do k=1,4 tetnab(i,k)=0 ntetnab(i)=0 centcent(i,k)=0.d0 enddo enddo
do itet=1,ntet-1do jtet=itet+1,ntet
inombre=0 do kk=1,4 ip=tetlist(itet,kk) do mm=1,4 in=tetlist(jtet,mm) if(ip.eq.in)inombre=inombre+1 enddo enddo
if(inombre .eq. 3) then ntetnab(itet)=ntetnab(itet)+1 tetnab(itet,ntetnab(itet))=jtetrr=dsqrt((x00(itet)-x00(jtet))**2 + (y00(itet)-y00(jtet))**2 +
& (z00(itet)-z00(jtet))**2) centcent(itet,ntetnab(itet))=(rr / d_0)**2! write(*,*)centcent(itet,ntetnab(itet)) ntetnab(jtet)=ntetnab(jtet)+1 tetnab(jtet,ntetnab(jtet))=itet centcent(jtet,ntetnab(jtet))=(rr / d_0)**2 endif
enddo
enddowrite(*,*)'smile12'
do itet=1,ntet if(ntetnab(itet).gt.4)then write(6,*)'error in subroutine nbors' write(6,*)'too many neighbors for tetrahedron= ',itet write(6,*)'number of nbors is=',ntetnab(itet) endif enddo
do itet=1,ntet
102
do k=1,ntetnab(itet) write(47,*) itet,k,tetnab(itet,k),centcent(itet,k) enddo
write(48,*) itet,ntetnab(itet) enddo close(unit=47)
close(unit=48)write(*,*) 'out nbors'
return end
103
subroutine dynamicsinclude 'test-inits1.f'
include 'test-inits2.f'
do 999 itime=1,irealtime
call nodeupdate
if (mod(itime,100).eq.0)thencall maxmincall pullforce
call savepic endif
call oldforcescall forcecalccall velocities
999 continue
returnend
104
subroutine nodeupdateinclude 'test-inits1.f'
include 'test-inits2.f'
xtime=xtime+dt
do 410 i=1,np if(pullme(i).eq.0)then x(i)=x(i)+dt*vx(i)+dt2o2*fx(i)/mass(i) y(i)=y(i)+dt*vy(i)+dt2o2*fy(i)/mass(i) z(i)=z(i)+dt*vz(i)+dt2o2*fz(i)/mass(i)
elsey(i)=y(i)+dt*vy(i)
endif410 continue
returnend
105
subroutine velocities ! Velocities update using the Verlet method include 'test-inits1.f' include 'test-inits2.f'
do i=1,np if(pullme(i).eq.0)then vx(i)=vx(i)+dto2*(fx(i)+fxold(i))/mass(i) vy(i)=vy(i)+dto2*(fy(i)+fyold(i))/mass(i) vz(i)=vz(i)+dto2*(fz(i)+fzold(i))/mass(i) endif enddo return end
106
subroutine epstensor(itet) include 'test-inits1.f' include 'test-inits2.f'
do j=1,4 npick(j)=tetlist(itet,j) enddo
do 200 j=1,4 i=npick(j) u(j)=x(i)-x0(i) v(j)=y(i)-y0(i) w(j)=z(i)-z0(i) 200 continue
c now that I have the u,v,w vectors, all I have to do is a matrixc multiply to find the shape function coefficients.
do 210 i=1,4 a(i)=0.d0 b(i)=0.d0 c(i)=0.d0 210 continue do 211 j=1,4 do 212 i=1,4 a(i)=a(i)+xm(itet,i,j)*u(j) b(i)=b(i)+xm(itet,i,j)*v(j) c(i)=c(i)+xm(itet,i,j)*w(j) 212 continue 211 continue
lam(1,1,itet)= 1.0 + a(2) lam(2,2,itet)= 1.0 + b(3) lam(3,3,itet)= 1.0 + c(4) lam(1,2,itet)= a(3) lam(2,1,itet)= b(2) lam(1,3,itet)= a(4) lam(3,1,itet)= c(2) lam(2,3,itet)= b(4) lam(3,2,itet)= c(3)
eps(1,1)=a(2)+0.5d0*(a(2)*a(2)+b(2)*b(2)+c(2)*c(2)) eps(2,2)=b(3)+0.5d0*(a(3)*a(3)+b(3)*b(3)+c(3)*c(3)) eps(3,3)=c(4)+0.5d0*(a(4)*a(4)+b(4)*b(4)+c(4)*c(4)) eps(1,2)=0.5d0*(a(3)+b(2)+a(2)*a(3)+b(2)*b(3)+c(2)*c(3)) eps(2,1)=eps(1,2) eps(1,3)=0.5d0*(a(4)+c(2)+a(2)*a(4)+b(2)*b(4)+c(2)*c(4)) eps(3,1)=eps(1,3) eps(2,3)=0.5d0*(b(4)+c(3)+a(3)*a(4)+b(3)*b(4)+c(3)*c(4)) eps(3,2)=eps(2,3)
return end
107
subroutine forcecalc include 'test-inits1.f' include 'test-inits2.f'
petot=0.d0
do 100 i=1,np fx(i)=0.d0 fy(i)=0.d0 fz(i)=0.d0100 continue
do itet = 1, ntet! call frank(itet) call test(itet)
enddo ! call rayleigh return end
108
subroutine force(itet) include 'test-inits1.f' include 'test-inits2.f'! Force calculation
pe =0.0
call epstensor(itet)
do i = 1,3 do j = 1,3
lambda(i,j) = lam(i,j,itet) enddo enddo
do j=1,4 npick(j)=tetlist(itet,j) enddo
a1 =a(1)b1 =b(1)c1 =c(1)a2 =a(2)b2 =b(2)c2 =c(2)a3 =a(3)b3 =b(3)c3 =c(3)a4 =a(4)b4 =b(4)c4 =c(4)
thet1 =0.5*ATan2(2*cx + (0.5*(2*a4*b4*(1 + (-1 + alpha)*r) + 1(1 + a2)*b2*(1 + (alpha - r)*r) +
1 a3*(1 + b3)*(1 + (alpha - r)*r) + 1 ((-1 - a2)*b2 + a3*(1 + b3))*(1 + r*(-2 + alpha + r)) 1 *Cos(2*thet0) - (1 + a2 + a3*b2 + b3 + a2*b3) 1 *(1 + r*(-2 + alpha + r))*Sin(2*thet0)))/r, 1 ax - bx + (0.25*(-(a2*(2 + a2)) - a3**2 + b2**2 1 + b3*(2 + b3) + 2*b4**2 + 2*a4**2*(-1 + r) 1 + (alpha*(-(a2*(2 + a2)) - a3**2 - 2*a4**2 1 + b2**2 + b3*(2 + b3)) + 2*(-1 + alpha)*b4**2)*r 1 + (a2*(2 + a2) + a3**2 - b2**2 - 2*b3 - b3**2)*r**2 + 1(1 + r*(-2 + alpha + r))*((2 + a2*(2 + a2) - a3**2 1 - b2**2 + b3*(2 + b3))* Cos(2*thet0) + 1 2*((1 + a2)*a3 - b2*(1 + b3))*Sin(2*thet0))))/r)
dir(itet) = thet1
fxsum=0.d0 fysum=0.d0 fzsum=0.d0
do 14 i=1,4 j=npick(i)
109
m2i=xm(itet,2,i) m3i=xm(itet,3,i) m4i=xm(itet,4,i)
fxsum = -0.5*m4i*((2*(-c2 - b3*c2 + b2*c3)*(b3 - a4*c2 1 - a4*b3*c2 + a4*b2*c3 - b4*c3 + c4 + b3*c4 1- a3*(b2 - b4*c2 + b2*c4) + a2*(1 + b3 - b4*c3 + c4 + b3*c4) - 1 k1))/k1 + 2*a4*(1 + (-1 + 1/r)*Cos(thet1)**2) 1 + 2*b4*(-1 + 1/r)*Cos(thet1)*Sin(thet1) 1 + alpha*(2*a4*Cos(thet1)**2 + 2*b4*Cos(thet1)*Sin(thet1))) - 1 0.5*m3i*((2*(-b2 + b4*c2 - b2*c4)*(b3 - a4*c2 - a4*b3*c2 1 + a4*b2*c3 - b4*c3 + c4 + b3*c4 - a3*(b2 - b4*c2 + b2*c4) 1 + a2*(1 + b3 - b4*c3 + c4 + b3*c4) - k1))/ 1 k1 + (1 + (-1 + 1/r)*Cos(thet1)**2)*(2*a3 1 + 2*a3*(-1 + r)*Sin(thet0)**2 + (1 + a2)*(-1 + r) 1 *Sin(2*thet0)) + 2*(-1 + 1/r)*Cos(thet1)*(b2*(-1 + r) 1 *Cos(thet0)*Sin(thet0) + (1 + b3)*(1 1 + (-1 + r)*Sin(thet0)**2))*Sin(thet1) + 1 alpha*(Cos(thet1)**2*(2*a3*Cos(thet0)**2 - 1 (1 + a2)*Sin(2*thet0)) + 2*Cos(thet1) 1 *((1 + b3)*Cos(thet0)**2 - b2*Cos(thet0) 1 *Sin(thet0))*Sin(thet1))) - 1 0.5*m2i*((2*(1 + b3 - b4*c3 + c4 + b3*c4) 1 *(b3 - a4*c2 - a4*b3*c2 + a4*b2*c3 - b4*c3 1 + c4 + b3*c4 - a3*(b2 - b4*c2 + b2*c4) + 1 a2*(1 + b3 - b4*c3 + c4 + b3*c4) - k1))/k1 1 + (1 + (-1 + 1/r)*Cos(thet1)**2)* 1 (2 + 2*a2 + 2*(1 + a2)*(-1 + r)*Cos(thet0)**2 1 + a3*(-1 + r)*Sin(2*thet0)) + 1 (-1 + 1/r)*Cos(thet1)*(b2 + b2*(-1 + r)*Cos(thet0)**2 1 + (1 + b3)*(-1 + r)*Cos(thet0)*Sin(thet0))*Sin(thet1) + 1 (-1 + 1/r)*Cos(thet1)*(b2*(1 + (-1 + r)*Cos(thet0)**2) 1 + (1 + b3)*(-1 + r)*Cos(thet0)*Sin(thet0))*Sin(thet1) + 1 alpha*(Cos(thet1)**2*(2*(1 + a2)*Sin(thet0)**2 1 - a3*Sin(2*thet0)) + 2*Cos(thet1)*(-((1 + b3) 1 *Cos(thet0)*Sin(thet0)) + b2*Sin(thet0)**2)*Sin(thet1)))
fysum = -0.5*m4i*((2*(a3*c2 - c3 - a2*c3)*(b3 - a4*c2 - 1 a4*b3*c2 + a4*b2*c3 - b4*c3 + c4 + b3*c4 - 1 a3*(b2 - b4*c2 + b2*c4) + a2*(1 + b3 - b4*c3 1 + c4 + b3*c4) - k1))/ 1 k1 + 2*a4*(-1 + 1/r)*Cos(thet1)*Sin(thet1) 1 + alpha*(2*a4*Cos(thet1)*Sin(thet1) + 2*b4*Sin(thet1)**2) 1 + 2*b4*(1 + (-1 + 1/r)*Sin(thet1)**2)) - 1 0.5*m3i*((2*(1 - a4*c2 + c4 + a2*(1 + c4))*(b3 - a4*c2 1 - a4*b3*c2 + a4*b2*c3 - b4*c3 + c4 + b3*c4 - a3*(b2 1 - b4*c2 + b2*c4) + a2*(1 + b3 - b4*c3 + c4 + b3*c4) 1 - k1))/k1 + 2*(-1 + 1/r)*Cos(thet1)*((1 + a2)*(-1 + r) 1 *Cos(thet0)*Sin(thet0) + a3*(1 + (-1 + r)*Sin(thet0)**2))* 1 Sin(thet1) + (2 + 2*b3 + 2*(1 + b3)*(-1 + r)*Sin(thet0)**2 1 + b2*(-1 + r)*Sin(2*thet0))*(1 + (-1 + 1/r)*Sin(thet1)**2) + 1 alpha*(2*Cos(thet1)*(a3*Cos(thet0)**2 - (1 + a2) 1 *Cos(thet0)*Sin(thet0))*Sin(thet1) + (2*(1 + b3) 1*Cos(thet0)**2 - b2*Sin(2*thet0))*Sin(thet1)**2)) - 1 0.5*m2i*((2*(a4*c3 - a3*(1 + c4))*(b3 - a4*c2 1 - a4*b3*c2 + a4*b2*c3 - b4*c3 + c4 + b3*c4 - a3*(b2 1 - b4*c2 + b2*c4) + a2*(1 + b3 - b4*c3 + c4 + b3*c4) - k1))/ 1 k1 + (-1 + 1/r)*Cos(thet1)*(1 + a2 + (1 + a2)*(-1 + r) 1 *Cos(thet0)**2 + a3*(-1 + r)*Cos(thet0) 1 *Sin(thet0))*Sin(thet1)+
110
1 (-1 + 1/r)*Cos(thet1)*((1 + a2)*(1 + (-1 + r) 1 *Cos(thet0)**2) + a3*(-1 + r) 1 *Cos(thet0)*Sin(thet0))*Sin(thet1) + 1 (2*b2 + 2*b2*(-1 + r)*Cos(thet0)**2 + (1 + b3) 1 *(-1 + r)*Sin(2*thet0))*(1 + (-1 + 1/r)*Sin(thet1)**2) + 1 alpha*(2*Cos(thet1)*(-(a3*Cos(thet0)*Sin(thet0)) 1 + (1 + a2)*Sin(thet0)**2)*Sin(thet1) 1 + (2*b2*Sin(thet0)**2 - (1 + b3) 1 *Sin(2*thet0))*Sin(thet1)**2))
fzsum = -0.5*(2*(1 + c4) + (2*(1 - a3*b2 + b3 + 1 a2*(1 + b3))*(b3 - a4*c2 - a4*b3*c2 + a4*b2*c3 1 - b4*c3 + c4 + b3*c4 - a3*(b2 - b4*c2 + b2*c4) + 1 a2*(1 + b3 - b4*c3 + c4 + b3*c4) - k1))/k1)*m4i - 0.5*m2i* 1 (c2 + (2*(-a4 - a4*b3 + a3*b4)*(b3 - a4*c2 - a4*b3*c2 1 + a4*b2*c3 - b4*c3 + c4 + b3*c4 - a3*(b2 - b4*c2 + b2*c4) 1 + a2*(1 + b3 - b4*c3 + c4 + b3*c4) - k1))/ 1 k1 + c2*(-1 + r)*Cos(thet0)**2 + c2*(1 + (-1 + r) 1 *Cos(thet0)**2) + 2*c3*(-1 + r)*Cos(thet0)*Sin(thet0)) - 1 0.5*m3i*((2*(a4*b2 - b4 - a2*b4)*(b3 - a4*c2 - a4*b3*c2 1 + a4*b2*c3 - b4*c3 + c4 + b3*c4 - a3*(b2 - b4*c2 + b2*c4) 1 + a2*(1 + b3 - b4*c3 + c4 + b3*c4) - k1))/ 1 k1 + 2*c2*(-1 + r)*Cos(thet0)*Sin(thet0) 1 + 2*c3*(1 + (-1 + r)*Sin(thet0)**2))
! write(*,*) fxsum, fysum, fzsum! call volchange(itet,vol0) fx(j)=fx(j)+fxsum*vtemp(itet) fy(j)=fy(j)+fysum*vtemp(itet) fz(j)=fz(j)+fzsum*vtemp(itet)14 continue
return end
111
subroutine volchange(itet,vol0) ! Calculate the change in volume of an element.
include 'test-inits1.f' include 'test-inits2.f'
do 778 j=1,4 npick(j)=tetlist(itet,j) 778 continue
xa=x(npick(1)) xb=x(npick(2)) xc=x(npick(3)) xd=x(npick(4))
ya=y(npick(1)) yb=y(npick(2)) yc=y(npick(3)) yd=y(npick(4))
za=z(npick(1)) zb=z(npick(2)) zc=z(npick(3)) zd=z(npick(4))
a11=xa-xb a12=ya-yb a13=za-zb
a21=xb-xc a22=yb-yc a23=zb-zc
a31=xc-xd a32=yc-yd a33=zc-zd
vol0=a11*a22*a33+a12*a23*a31+a13*a21*a32 vol0=vol0-a13*a22*a31-a11*a23*a32-a12*a21*a33 vol0=vol0/6.
vol0=abs(vol0)
return end
112
subroutine frank(itet) include 'test-inits1.f' include 'test-inits2.f'
ax = 0.0bx = 0.0cx = 0.0
jtet =ntetnab(itet)
do k=1,jtetdx = centcent(itet,k)ax = ax + (cos(dir(tetnab(itet,k))) / dx )**2bx = bx + ( sin(dir(tetnab(itet,k)))/ dx )**2 cx = cx + 0.5*sin(2.0*dir(tetnab(itet,k))) / ( dx )**2enddo
ax = 0.1*axbx = 0.1*bxcx = 0.1*cx
return end
113
subroutine pullforce include 'test-inits1.f' include 'test-inits2.f'
fytot = 0.d0do 602 i=1, np if(pullme(i).eq.1) thenfytot=fytot + fy(i)endif
602 continue return end
114
subroutine oldforces include 'test-inits1.f'
include 'test-inits2.f'
do 420 i=1,np fxold(i)=fx(i) fyold(i)=fy(i) fzold(i)=fz(i) 420 continue
returnend
115
subroutine savepicinclude 'test-inits1.f'
include 'test-inits2.f'
strain=100*(ylength-ylength0)/ylength0 write(18,*) strain, -fytot write(6,*)itime,' strain=',strain,' force= ',-fytot
write(28,*)npwrite(28,*)do i=1,npwrite(28,*)'A',x(i),y(i),z(i)
enddo
vol_now=0.d0
thetave =0.0
do itet=1,ntet call volchange(itet,vol0)
vol_now = vol_now + vol0 thetave = thetave + sin(dir(itet))**2
write(31,*) itet, dir(itet)
enddowrite(37,*) strain, thetave/real(ntet)write(6,*) 'dir_rot',thetave/real(ntet)
del_vol = 100*(vol_now-vol_init)/vol_initwrite(*,*) 'volume change',del_vol,'percent'write(117,*) strain, del_vol
returnend
116
SUBROUTINE MATINV (A,LDA,N,IFLAG) implicit real*8(a-h,o-z)CC-----------------------------------------------------------------------C MATINV WRITTEN BY CHARLES P. REEVE, STATISTICAL ENGINEERINGC DIVISION, NATIONAL BUREAU OF STANDARDS, GAITHERSBURG,C MARYLAND 20899CC FOR: COMPUTING THE INVERSE OF A GENERAL N BY N MATRIX IN PLACE,C I.E., THE INVERSE OVERWRITES THE ORIGINAL MATRIX. THE STEPS C OF THE ALGORITHM ARE DESCRIBED BELOW AS THEY OCCUR. ROWC INTERCHANGES ARE DONE AS NEEDED IN ORDER TO INCREASE THEC ACCURACY OF THE INVERSE MATRIX. WITHOUT INTERCHANGES THISC ALGORITHM WILL FAIL WHEN ANY OF THE LEADING PRINCIPALC SUBMATRICES ARE SINGULAR OR WHEN THE MATRIX ITSELF ISC SINGULAR. WITH INTERCHANGES THIS ALGORITHM WILL FAIL ONLYC WHEN THE MATRIX ITSELF IS SINGULAR. THE LEADING PRINCIPALCC [A B C]C SUBMATRICES OF THE MATRIX [D E F] ARE [A] AND [A B] .C [G H I] [D E]CC SUBPROGRAMS CALLED: -NONE-CC CURRENT VERSION COMPLETED JANUARY 15, 1987CC REFERENCE: STEWART, G.W., 'INTRODUCTION TO MATRIX COMPUTATIONS',C ACADEMIC PRESS, INC., 1973C-----------------------------------------------------------------------C DEFINITION OF PASSED PARAMETERSCC * A = MATRIX (SIZE NXN) TO BE INVERTED (REAL)CC * LDA = LEADING DIMENSION OF MATRIX A [LDA>=N] (INTEGER)CC * N = NUMBER OF ROWS AND COLUMNS OF MATRIX A (INTEGER)CC IFLAG = ERROR INDICATOR ON OUTPUT (INTEGER) INTERPRETATION: C -2 -> TOO MANY ROW INTERCHANGES NEEDED - INCREASE MXC -1 -> N>LDAC 0 -> NO ERRORS DETECTEDC K -> MATRIX A FOUND TO BE SINGULAR AT THE KTH STEP OFC THE CROUT REDUCTION (1<=K<=N)CC * INDICATES PARAMETERS REQUIRING INPUT VALUES C-----------------------------------------------------------------------C PARAMETER (MX=100) DIMENSION A(LDA,*),IEX(MX,2) IFLAG = 0CC--- CHECK CONSISTENCY OF PASSED PARAMETERSC IF (N.GT.LDA) THEN IFLAG = -1 RETURN ENDIFCC--- COMPUTE A = LU BY THE CROUT REDUCTION WHERE L IS LOWER TRIANGULARC--- AND U IS UNIT UPPER TRIANGULAR (ALGORITHM 3.4, P. 138 OF THEC--- REFERENCE)C
117
NEX = 0 DO 70 K = 1, N DO 20 I = K, N S = A(I,K) DO 10 L = 1, K-1 S = S-A(I,L)*A(L,K) 10 CONTINUE A(I,K) = S 20 CONTINUECC--- INTERCHANGE ROWS IF NECESSARYC Q = 0.0 L = 0 DO 30 I = K, N R = ABS(A(I,K)) IF (R.GT.Q) THEN Q = R L = I ENDIF 30 CONTINUE IF (L.EQ.0) THEN IFLAG = K RETURN ENDIF IF (L.NE.K) THEN NEX = NEX+1 IF (NEX.GT.MX) THEN IFLAG = -2 RETURN ENDIF IEX(NEX,1) = K IEX(NEX,2) = L DO 40 J = 1, N Q = A(K,J) A(K,J) = A(L,J) A(L,J) = Q 40 CONTINUE ENDIFCC--- END ROW INTERCHANGE SECTIONC DO 60 J = K+1, N S = A(K,J) DO 50 L = 1, K-1 S = S-A(K,L)*A(L,J) 50 CONTINUE A(K,J) = S/A(K,K) 60 CONTINUE 70 CONTINUECC--- INVERT THE LOWER TRIANGLE L IN PLACE (SIMILAR TO ALGORITHM 1.5,C--- P. 110 OF THE REFERENCE) C DO 100 K = N, 1, -1 A(K,K) = 1.0/A(K,K) DO 90 I = K-1, 1, -1 S = 0.0 DO 80 J = I+1, K S = S+A(J,I)*A(K,J) 80 CONTINUE A(K,I) = -S/A(I,I)
118
90 CONTINUE 100 CONTINUECC--- INVERT THE UPPER TRIANGLE U IN PLACE (ALGORITHM 1.5, P. 110 OFC--- THE REFERENCE) C DO 130 K = N, 1, -1 DO 120 I = K-1, 1, -1 S = A(I,K) DO 110 J = I+1, K-1 S = S+A(I,J)*A(J,K) 110 CONTINUE A(I,K) = -S 120 CONTINUE 130 CONTINUECC--- COMPUTE INV(A) = INV(U)*INV(L)C DO 160 I = 1, N DO 150 J = 1, N IF (J.GT.I) THEN S = 0.0 L = J ELSE S = A(I,J) L = I+1 ENDIF DO 140 K = L, N S = S+A(I,K)*A(K,J) 140 CONTINUE A(I,J) = S 150 CONTINUE 160 CONTINUECC--- INTERCHANGE COLUMNS OF INV(A) TO REVERSE EFFECT OF ROW C--- INTERCHANGES OF AC DO 180 I = NEX, 1, -1 K = IEX(I,1) L = IEX(I,2) DO 170 J = 1, N Q = A(J,K) A(J,K) = A(J,L) A(J,L) = Q 170 CONTINUE 180 CONTINUE RETURN END
119
SUBROUTINE jacobi(A,V,NEQ,TL)IMPLICIT REAL*8 (A-H,O-Z)DIMENSION A(NEQ,NEQ),V(NEQ,NEQ)
ZERO = 0.0D0SUM = ZEROTOL = DABS(TL)
C---- SET INITIAL EIGENVECTORS -------------DO 200 I=1,NEQDO 190 J=1,NEQIF (TL.GT.ZERO) V(I,J) = ZERO
190 SUM = SUM + DABS(A(I,J))IF (TL.GT.ZERO) V(I,I) = 1.0
200 CONTINUEC---- CHECK FOR TRIVIAL PROBLEM -----------
IF (NEQ.EQ.1) RETURNIF (SUM.LE.ZERO) RETURNSUM = SUM/DFLOAT(NEQ*NEQ)
C-------------------------------------------C---- REDUCE MATRIX TO DIAGONAL ------------C-------------------------------------------400 SSUM = ZERO
AMAX = ZERODO 700 J=2,NEQIH = J - 1DO 700 I=1,IH
C---- CHECK IF A(I,J) IS TO BE REDUCED -----AA = DABS(A(I,J))IF (AA.GT.AMAX) AMAX = AASSUM = SSUM + AAIF (AA.LT.0.1*AMAX) GO TO 700
C---- CALCULATE ROTATION ANGLE ----------AA=ATAN2(2.0*A(I,J),A(I,I)-A(J,J))/2.0SI = DSIN(AA)CO = DCOS(AA)
C---- MODIFY "I" AND "J" COLUMNS --------DO 500 K=1,NEQTT = A(K,I)A(K,I) = CO*TT + SI*A(K,J)A(K,J) = -SI*TT + CO*A(K,J)TT = V(K,I)V(K,I) = CO*TT + SI*V(K,J)
500 V(K,J) = -SI*TT + CO*V(K,J)C---- MODIFY DIAGONAL TERMS -------------
A(I,I) = CO*A(I,I) + SI*A(J,I)A(J,J) =-SI*A(I,J) + CO*A(J,J)A(I,J) = ZERO
C---- MAKE "A" MATRIX SYMMETRICAL -------DO 600 K=1,NEQA(I,K) = A(K,I)A(J,K) = A(K,J)
600 CONTINUEC---- A(I,J) MADE ZERO BY ROTATION ------700 CONTINUEC---- CHECK FOR CONVERGENCE -------------
IF(DABS(SSUM)/SUM .GT.TOL)GO TO 400RETURNEND
120