Вы находитесь на странице: 1из 11

Computer Simulation of Polymeric Materials.

I. Stress-Strain Behavior
ROBERT COOK, Lawrence Livermore National Laboratory,
Livermore, California 94550

Synopsis
A model of polymeric materials has been developed that includes many of the features of
condensed-phase polymer chain dynamics, central among them chain relaxation via conformational motion. The model consists of a number of chains of particles that are connected by bonds
with multiwelled potentials to approximate the energetics of conformational motion. Interactions
between particles on adjacent chains are modeled by short range repulsive potentials. We have
examined the stress-strain behavior of the model using molecular dynamics simulations and find
qualitative agreement with the observed experimental behavior of polymeric materials.

INTRODUCTION
Polymeric materials below Tg generally exhibit three distinct regions in
their stress-strain behavior as qualitatively displayed in Figure 1. The
origins of these regions can be traced to the molecular response of the
individual polymer chains to the applied stress or strain. A t low applied stress
the material behaves as a linear elastic medium with an instantaneous
modulus equal to the slope of the stress-strain curve. The modulus gives us a
measure of the instantaneous material stiffness. In this region the removal of
the applied stress results in an elastic response of the material without
hysteresis. On a molecular level we interpret this region as one where the
individual backbone bond and rotational angles, and to a lesser degree bond
lengths, open, rotate, and stretch, respectively, without any significant translational chain motion in the form of local conformational rearrangements. In a
microscopic sense the deformation is affine, since the deformation is distributed uniformly throughout the sample. A t some critical level of stress the
instantaneous modulus is abruptly reduced to near zero, and the material
extends or yields without any significant increase in the applied stress. One
often sees a dip in the yield region that is a result of the nonlinearity of the
sample under the experimental conditions and the localization of the deformation in necking. On a molecular level we interpret this yielding in terms of a
chain motion in the form of local conformational changes and associated chain
slippage. In this respect we might term the deformation microscopically
nonafine, since on the scale of the individual chains some sections are moving
relative to others. Chain failure is unlikely to significantly contribute at these
stress levels since the activation energy for bond breaking is much greater
than that for conformational or translational motion. However, because these
motions are relatively slow, this region of the stress-strain curve is particularly sensitive to temperature and to the rate of application of the stress.
Journal of Polymer Science: Part B: Polymer Physics, Vol. 26, 1337-1347 (1988)
0 1988 John Wiley & Sons, Inc.
CCC 0098-1273/88/071337-11$04.00

1338

Material failure

v)

v)

Strain
Fig. 1. A typical stress-strain c w e for a polymer below its glass-transition temperature.

After yielding, the instantaneous modulus of the material frequently increases


again, often t o a level in significant excess of the original modulus. The curve
ends abruptly with material failure. Microscopically this increase in modulus
occurs when most of the conformational softness has been stretched out,
and we are once again in a mode of opening backbone bond angles and
stretching bond lengths. Since the softer backbone rotational angles no longer
contribute, the instantaneous modulus is higher due to the stiffer backbone
bond angles. We must note before continuing that stress-strain response is
often a complicated behavior depending not only on the material but also on
the temperature regime and strain rate. This is particularly true in the
plateau region, where, for many materials, the yielding process often involves
complicated processes including bulk material flow and microvoid formation.
Also it is often the case that material failure closely follows yielding, terminating the curve before any sharp upturn. Clearly we are taking a somewhat
simplified or idealized approach a t this point.
Although we have made this analysis in terms if individual chain motions, it
is important to note that the observed experimental behavior is strongly
dependent upon the interaction between chains. It is the interactions with
neighboring chains that largely define the glassy state and prevent the
material from responding to applied stresses in a purely entropic-elastic mode.
Sufficiently below Tg,conformational freedom is inhibited by neighboring
chains. During yielding, the mechanical force provides the energy to elongate
the chains, pulling segments over conformational barriers that would otherwise be insurmountable (on a laboratory time scale) with the available
thermal energy. The entropically unfavorable configurations produced dur-

COMPUTER SIMULATION

1339

the yield region are thus frozen in. The demonstration of this with real
materials is often quite striking. If one raises the temperature of a sample that
has yielded, as one passes through TBand the frozen in conformational states
are freed, the sample will abruptly snap back to near its original length,
driven by conformational entropic forces.
Our objective in this work is to develop a model of glassy polymeric
materials and, using the techniques of computer simulation, investigate the
important features of its stress-strain behavior.* The computational constraints of computer simulation require a relatively simple model that includes the experimental features of chains with multiwelled extensional
potentials and significant interchain potentials. In addition we wish to start,
at least, with a model that may lend itself to analytic calculation. Although
we will not pursue this avenue here, work along these lines is in progress. In
the next section we will detail the model we have developed. The model
incorporates the idea of stress-induced conformational motion of individual
chains as well as the collective effects of chain interaction. Following the
discussion of the model we outline the computer simulation methods and
studies performed and then give a discussion of the results obtained from our
work.

MODEL
Our model,? schematically displayed in Figure 2, consists of a number of
spring-particle chains that represent individual polymer chains. An important
feature of the model is that it is mathematically a purely one-dimensional
system. Each particle is characterized by only a single coordinate on an axis
running parallel to the chain direction. The figure suggests a parallel format
for chains packed in a two- (or three-) dimensional array. However, the
adjacency of chains is kept track of independently, and there is no explicit
distance between chains, since there is only one coordinate dimension. Thus
by varying, for a given chain, which particles are adjacent to which neighboring chains, we can create an entangled system. The choice of this special
one-dimensionalsystem is motivated by the immense saving in computational
time and is justified by the expectation that the most significant motions and
forces are in the chain direction, which is also the same direction that stresses
and strains are applied. This feature of the model does, of course, eliminate
the study of the effect of forces and motions perpendicular to the chain
direction, although shearing forces can still be dealt with as described shortly.
In order to model the conformational motions of polymer chains, intrachain
bonds are characterized by a double-welled potential displayed in Figure 3a.
The two low energy states can, for example, be thought of as representing
gauche and trans states in a four-atom sequence. In this context note that
the particles in the model do not necessarily represent the individual atoms of
a real polymer chain. More generally, the short and long states model the
existence of a conformational potential reduced to one dimension. This ap*We also note some related work that has appeared; see Ref. 2. The work of Weiner and
collaborators reports on a related approach to some of these problems; see especially Ref. 3.
Some calculations using the same basic model have been published previously; see Ref. 4.

COOK

1340

Fig. 2. A schematic of our model system. Note that the model is mathematically one-dimensional; adjacency of chains is kept track of independently.

proach to conformational modeling is not new with this work, having been
used by Pear and Weiner5 and Helfand6 previously. In the interest of
computational simplicity, we have created the double-welled potential from
spliced parabolas as
1
v,= + -yl(
2

H-

+ ,y3(

- Ul)2

-y2(1 -

1-

a2)2

for 1 < b,

forb, < 1 < b,

(3)

for 1 > b,

where 1 is the distance between two particles. The various parameters are
picked such that the potential and its derivative are continuous at b, and b,.
For the work described here we have used the symmetric double-welled
potential illustrated in Figure 3a with parameters as listed in Table I. This
symmetric system with dimensionless parameters was picked for this preliminary work to explore the model, and it should be noted that using the
same general form we can generate double-welled potentials with different
curvatures in different regions and/or different well depths and barrier
heights. Additionally, by extending the potential to many wells, we can
simulate a more complex conformational transition sequence.
The potential we have picked to use between particles on different chains is
the repulsive potential shown in Figure 3b. Mathematically, this is also a

COMPUTER SIMULATION

2.1
1.8

1341

.9

1.2

1.5

1.8

1.5

.6

.3
0
.6

2.1

Spring length
(a)

.40

.30

.10

-. 6

-.4

-.2
0
.2
Relative displacement

.4

.6

(b)
Fig. 3. The (a) intrachain double-welledpotential and (b) interchain repulsive potential.

COOK

1342

TABLE I
Parameters for the Attractive and Repulsive Potentials, eqs. (3) and (4)

Note: The energetic parameters are dimensionless, and distances are measured units of the
equilibrium length of the short well.

spliced-parabola potential of the form.

+--(Ax-rc)
1 kr,
2 rc - r,

=o

forr,<Ax<r,
for lAxl > r,

where Ax is the relative displacement of two particles on adjacent chains. The


values of the parameters are also listed in Table I. The choice of the
interchain particle potential as a purely repulsive potential was motivated by
a desire for simplicity and by the realization that the dominant interaction we
hope to model by this potential is the hindrance involved with one chain or
particle sliding past another. The use of two different parabolic curvatures in
different regions (0 to f r, and fr, to f rc) was motivated by an effort to
enhance the strength of the very short-range repulsion forces. The cut-off at
Ax = frc = f0.5 allows both the potential and force to be continuous with
zero. The net effect of this potential is that, with respect to a pair of neighbors
on an adjacent chain, the position of lowest energy for a particle is directly
between them. Let us emphasize again that the mathematical nature of the
model is purely one dimensional, and as such this interchain potential, which
depends only upon the relative coordinates of the two particles in question, is
much like a shearing force. Also, we note again that we can model entangled
chains through the control of adjacency. For example we could set chain
adjacency such that the first three particles of chain i were adjacent to chain
j , but the next three particles were adjacent to chain k .
The collection of chains is generated in the following manner. As an input
parameter, in addition to the dimensionless potential parameters discussed
above, we include the desired overall length (as measured in units of the
equilibrium length of the short bond) of the collection of chains and the

COMPUTER SIMULATION

1343

desired fraction of initial short and long bonds. Then, placing a particle a t
the origin for each chain, we place sequential particles a t either the short or
long equilibrium lengths, arbitrarily chosen to be 1 or fi in this work, on the
basis of a random number generator and the preselected probability of short
and long linkages. In this work we have made the short and long states
equally probable as one might expect from the potential in Figure 3a where
the wells are of equal depth. We note, however, that the processing history
might leave a polymer sample with a conformational distribution very different from that expected from Boltzmann statistics. This procedure continues
until the coordinate of a placed particle is a t a value greater than the desired
initial chain length, Lo. A determination is then made to see if the last
particle is closer to Lo than the one preceding it. If so, the process ends for
that chain; if not, the last particle is removed. When all the chains have been
created in this fashion, we have a collection of chains, the left end particles of
which are all a t the origin while the right end particles are distributed about
the value Lo. The number of particles in any one chain will vary about an
average of 1 + 2Lo/(a1 aa).The chains are arranged in a hexagonal array,
so that each chain has six nearest neighbor chains. At this point more
complicated adjacency rules could be imposed as discussed earlier to create
entangled systems. In the work discussed here we will not explore this
avenue. Periodic boundary conditions are used in directions perpendicular to
the chain direction. The cross-section of the array of chains is most easily
visualized as showing hexagonal packing in a rhombus-shaped figure. The
coordinates of each particle on a given chain are then adjusted by adding 6/2
to each coordinate, where 6 is the distance Lo - xl, x1 being the coordinate of
the last particle in the chain. This small shift has the effect of making the
orientation of the particles on either end symmetric with respect to one
another. Note a t this point that particles are at potential minima with respect
to neighboring particles on the same chain but not with respect to particles on
adjacent chains. To remedy this, the system of chains is now allowed to fully
relax.
Next it is necessary to develop a method for applying an extensional force
to the collection of chains. For convenience we have chosen to conduct our
stress-strain computer experiments by controlling the applied stress rather
than the sample elongation as is generally the case experimentally. The
application of the extensional force might have been done by applying the
force to the end particles of each chain. We rejected this approach for two
reasons. First, we plan in the future to study chains with broken bonds, and
this type of extensional force would lead in some cases to the equivalent of
pull-out, an interesting phenomena but not one we wish to pursue at this
time. More importantly, forces acting directly on individual chains is inconsistent with the experimental analog where the ends of the macroscopic
sample are clamped and then the extensional force is applied. To mimic this
feature we have chosen to apply extensional forces as follows. We first
determine the average position of the end particles on each end of the
collection of chains. Another pair of particles is placed a t these points,
denoted as x l c and xrC, and connected to all the particles a t that end by
simple Hookean springs at their equilibrium lengths, which may be slightly
different for each chain. The Hookean force constants are all identical and in
this work are set equal to y. We can express the potential associated with

1344

COOK

these particles as

where n is the number of chains, m iis the number of particles in chain i, and
is the equilibrium length of the clamping springs attached to chain i.
The extensional forces are then applied to these clamping particles. The
molecular dynamics methods are described in the next section.

SIMULATION METHODS
The actual simulation of the model described above requires addressing
several key questions. Central to these is the problem of temperature control
and energy dissipation. We have chosen in this initial study to work a t 0 K
and t o damp out kinetic energy as it is produced by adding a viscous damping
term to the equations of motion to be solved. Our motivation in this matter is
as follows. First, our goal a t this point is to understand the simple mechanics
of the model, not its temperature dependence. Thermal energy sufficiently
below Tg is manifest primarily in atomic vibrations rather than conformational freedom, which has been largely frozen out. Thus entropic effects owing
to conformational motion should be relatively unimportant, a t least on
laboratory time scales. A future paper on the temperature dependence of our
model will examine these questions more closely. Maintaining 0 K is another
problem. As we apply extensional forces to the model, some of the work done
is dissipated as heat, which, in the case of the dynamics of the model, means
the generation of particle velocities. In a model with only a few hundred
particles there is no easy way to transfer this heat to the surroundings, and
thus it continues to grow, ultimately dominating the potential energy and the
force derived therefrom. Although the process of heat dissipation holds some
important questions that we plan to address in future work, it is not our
intention a t this time to pursue them until we develop a firmer understanding
of the purely mechanical aspects of the model. Thus, in order to maintain 0 K
we have studied the model as if it were immersed in a viscous medium.
Mathematically then, we are solving the N-coupled equations of motion of the
form

where N is the total number of particles, U, is the sum of U, and U, for every
interparticle interaction and Uc, x, is the coordinate value of particle i, q is
the damping coefficient on the particle velocity u,, and FeXtis the external
extensional force applied only to the two clamping particles described above.
Computationally we apply a small increment of extensional force and then,
while holding it constant, slowly damp the motions so that the system relaxes
to its lowest potential energy configuration consistent with the applied force
and the initial condition. The configuration a t this level of elongational force
is then recorded before another increment of force is applied. Thus what we
follow is the minimum energy configuration pathway from a state of no
external elongational force to a point when the external forces have completely stretched the system out. The length of the system is monitored by the

COMPUTER SIMULATION

1345

distance between the two clamping particles. Additionally, since we have the
configuration at any point during the extension, we can keep track of
the fraction of long and short bonds. We note here that there are certainly
more efficient ways for energy minimization than damped molecular dynamics. Our motivation in using this approach stems from our planned finite
temperature studies as indicated earlier.
All simulations were carried out using standard simple difference techniques. The treatment of the damping was formally inexact in that u ( t ) was
approximated by u ( t - St/2); however, as a minimization tool the method is
acceptable since the final state reached is the same. Systems of from 25 chains
averaging 12 particles long to 81 chains averaging 45 particles long were run to
extensions of about 170% of the initial length. Computational time on a
VAX-750 varied from less than an hour for the smaller systems to several
days for the largest systems.

RESULTS AND DISCUSSION


In Figure 4 we show with a solid line and the left-hand ordinate a plot of
FexJn, the applied stress per chain (equivalent to a force per cross-sectional
area) as a function of the elongation ratio L/L, for an n = 81 chain system
with initial length Lo = 60. Systems of different length and number of
particles per chain give qualitatively the same results. For our model the
strain is simply the elongation ratio minus one. The dashed line and the
right-hand ordinate show a plot of the fraction of bonds that are long as a
function of system elongation. Clearly the form of the stress-elongation curve
is similar to that discussed in the Introduction, suggesting that our model is
consistent with the experimental observation of stress-strain behavior in

14 r

1.o

.9
I

'u

n L

-L

0-

P
u , G

1.0

.4

1.1

1.2

1.3

1.4

1.5

1.6

Elongation
Fig. 4. Simulation results for a system of 81 chains arranged hexagonally (9 X 9) with average
length of 60. The solid line and left-hand ordinate show stress as a function of the system
elongation, L/L,. The dashed line and right-hand ordinate show the fraction of long bonds as a
function of system elongation.

COOK

1346

polymeric materials. Let us now examine the various regions of the computergenerated curve more closely.
At low degrees of stress the curve is linear, reflecting an elastic region.
The initial modulus, measured as stress per chain (FeXt/n)over the strain
( L / L o - 1) is 40.4. This quantity can be calculated directly from the model as
follows. Let n , and n, be the average initial number of short and long bonds
per chain. Then

Lo = nsal + n p 3

(7)

If we assume that the initial application of an elongational force acts to


extend all the bonds slightly, the average extension will be Fex,/y,where F,,
is the force per chain and y ( = y1 = y3 in this work) is the restoring force
coefficient for each bond. We have neglected the interchain restoring forces,
since they are weaker. Thus

Lo + - ( n s
Y
Fext

+ n,)

and this combined with eq. (6) gives the initial modulus

Using the parameters from Table I and noting that for this computer run
n J ( n , + n,) = 0.4912, consistent with a code input of equal probabilities for
short and long bonds, calculation of the modulus from eq. (9) gives 40.59, in
excellent agreement with the computer-generated value of 40.4.
We can also note that during this initial period, up to an elongation of a
little less than 1.1, the plot of fraction long remained flat, indicating very few
short-to-long transitions. This is consistent with our view of the deformation
occurring microscopically affinely without significant conformational transitions.
At elongations greater than 1.1 the instantaneous modulus drops to a value
of about 10 a t the same time that the fraction of long bonds steadily increases
to about 0.9. This region mimics the plateau region, where conformational
changes are the primary contributors to sample elongation. Modifying the
model to include more than two potential minima and a greater possible total
extension for a bond would undoubtedly reduce the modulus in this region
further as well as more realistically represent the many barriers that are
crossed as a real chain is extended.
In the elastic region, unloading of the force would allow the system to relax
t o its original length, few conformational changes having taken place. However, in the plateau region if we begin to unload the system we get the elastic
response appropriate for a system with the current fraction of long bonds.
Basically what we see is relaxation to no load a t a length extended from the
original length. The same result is seen in real materials that have yielded. On
unloading a real sample one sees a small elastic response to an equilibrium
length in excess of that of the undeformed sample. This is, of course, a time
and temperature dependent phenomena, however a t a temperature sufficiently
below T lengths remain unchanged on laboratory time scales.

COMPUTER SIMULATION

1347

As the stress per chain is increased the curve swings up again, and the
modulus once again increases. The deformation is once again affine: all bonds
are being stretched. The modulus does not grow to a high value as with many
real materials only because of the symmetry of the double-welled potential
used in this work. A more realistic potential would have risen much more
steeply when extending from the long well, a feature that would be reflected
in a much greater ultimate modulus. Since bond rupture is not possible with
the potential used in this work, we cannot expect the equivalent of material
failure. Rather, the simulation is stopped when nearly all bonds are extended.
One of our primary interests was to examine the nature of correlated
short-to-long transitions during the computer experiment. In the absence of
an interchain potential, all short bonds in a given chain climb the barrier
together and pass over in unison owing to the equal distribution of forces. As
the interchain potential is introduced it breaks up this correlation, since all
bonds do not share the extensional forces equally. Instead what we see is
numerous small clusters of transitions, where one bond passing over a barrier
will carry with it bonds on neighboring chains. Thus the correlation changes
from along a single chain to between neighboring chains as the interchain
potential is increased.
This work is continuing along several lines. First we are continuing our
characterization of stress-strain behavior by a careful examination of the
importance of barrier height and the double-well extensional distance, a , - u3.
We are also looking a t extended potentials that include several wells of
different depths separated by barriers of differing heights and steepnesses.
Additionally we plan to examine temperature dependent phenomena with
Brownian dynamics simulations. In summary, we have developed a useful
model for the study of the mechanical properties of polymeric materials. In
this paper we have shown its application to the modeling and interpretation
of stress-strain behavior. It is our expectation that the model and derivatives
of it will be useful in the interpretation and study of a number of material
properties.
The author thanks Dr. Witold Brostow for several useful discussions. This work was performed
under the auspices of the U.S. Department of Energy by the Lawrence Livermore National
Laboratory under Contract No. W-7405-ENG-48.

References
1. See, for example, H. R. Allcock and F. W. Lampe, Contemporary Polymer Chemistry,
Prentice Hall, Englewood Cliffs, N.J., 1981, p. 536.
2. D. Brown and J. H. R. Clarke, J . Chem. Phys., 84, 2858 (1986); Y. Termonia, P. Meakin,
and P. Smith, Macromolecules, 18, 2246 (1985); Zbid., 19, 154 (1986).
3. D. H. Berman and J. H. Weiner, J . Chem. Phys., 83, 1311 (1985); J. H. Weiner and D. H.
Berman, Macromolecules, 17,2015 (1984); J. H. Weiner and T. W. Stevens, Macromolecules, 16,
672 (1983).
4. R. Cook and M. B. Mercer, Muter. Chem. Phys., 12, 571 (1985); W. Brostow and D. P.
Turner, J . Rheology, 30, 767 (1986).
5. J. H. Weiner and M. R. Pear, Macromolecules, 10, 317 (1977).
6. E. Helfand, J. C h m . Phys., 69, 1010 (1978).

Received October 20, 1986


Accepted February 20,1987

Вам также может понравиться