Sunteți pe pagina 1din 9

Deterministic Chaos

Theory and Its


Applications to
Materials Science
Alan J. Markworth, John Stringer, and
Roger W. Rollins
Background
Over the past 30 years or so, some fascinating new developments have taken
place in a field that is now called "nonlinear science." These developments
have been of such magnitude that historian Alan Beyerchen concluded "There is
every reason to believe that the westernized world is in the early stages of an
intellectual transformation of major proportions, perhaps as significant as the
emergence of the modern worldview in
the fifteenth through seventeenth centuries."1
These new developments are rooted in
work conducted by mathematician Jules
Henri Poincare around the turn of the
century. He carried out some calculations regarding planetary orbits and
demonstrated the possibility of erratic or
"chaotic" dynamical behavior. What is
particularly interesting about such behavior is that it is exhibited by deterministic systems, that is, systems that have no
stochastic (noisy) character of any kind.
Little was done with Poincare's findings
until the 1960s when new work in nonlinear science, based on numerical models, led to the discovery of amazingly
complex behavior that can be exhibited
even by very simple deterministic systems. Since that time, the discipline has
grown enormously and has been applied
in practically every area of science and
engineering, as well as in other diverse
areas such as the financial market, political science, economics, and even Adlerian psychology.
The existence of chaotic dynamics in
nature is now known to be widespread,
which leads to the question of why it has
only recently become the subject of in-

tense study. The answer to this question


is not simple. One clue lies in the fact that
theoretical studies of chaos require extensive numerical computations which,
in turn, require the use of powerful
computers, the likes of which have become commonly available only within
the past few decades. Moreover, as far as
experiments are concerned, it is most
likely that chaos has indeed been observed on numerous occasions but has
gone unrecognized as such and may
simply have been discarded as unwanted
"noise." For example, it has been noted,
with specific reference to nonlinear electrochemical systems (although the stated
conclusion probably applies quite generally), that "Probably chaotic oscillations
were many times observed in electrochemical systems but they were treated
rather as a noise and therefore all authors
passed over them in silence."2
However, one does wonder why it took
so long for chaos to be recognized for
what it is, particularly since it can occur
in very simple systems. This question
was forcefully addressed with particular
regard to chaos in the double pendulum:
"This simple system exhibits outrageously complicated behavior. How
could anyone watch the double pendulum and continue to assume that all solutions of Newton's equations could be
developed in quasiperiodic perturbation
expansions? Did hundreds of years really go by without anyone looking at a
dynamical system in action?"3
Interest in this subject has been
spurred among the general public by the
very popular book on chaos by Gleick,4
as well as by a presentation on chaos on

the public television program NOVA and


by numerous articles in the popular
press. In many cases, the question has
been "How can this concept help me in
my area of research?" Although this has
proved difficult to answer, practical applications are beginning to emerge,
which we illustrate later in this article.
In the discussion that follows, some
comments are made regarding what
chaos actually is and what some of the
characteristics of a chaotically oscillating
system are. Two simple examples are
considered in order to illustrate these
concepts. We then describe different
ways of studying chaotic systems, some
of the pitfalls that can be encountered in
carrying out these studies, and general
areas that are currently of great interest.
Finally, we turn our attention to chaotic
dynamics in materials behavior, citing
some examples of existing work and suggesting what the future might hold.
What Is Deterministic Chaos?
At this point, we need to give a very
brief and simplified outline of what deterministic chaos is. Treatments that are
far more comprehensive can be found
elsewhere, such as in a variety of engineering-oriented books on the subject,
for example, Thompson and Stewart,5
Moon,67 and Kim and Stringer.8
Consider the state of a dynamical system, which at any given time is described by the instantaneous values of all
its state variables, such as position and
momentum. If N of these variables are
required to fully define the state of the
system, we can envisage an N-dimensional Cartesian state space wherein each
axis represents one of the state variables
and the state of the system at any given
time is described by a point within that
space. As time progresses, that point
traces a path within the state space that is
called the trajectory of the system. That
particular subspace of the state space, toward which the system converges following the dying out of any transient
behavior, is called its attractor. For example, if the system is a damped, undriven pendulum which eventually
comes to rest, its attractor is a point at the
origin of a two-dimensional state space,
the two independent state variables being position and momentum. If the
damping is removed, there is no attractor; the pendulum simply moves in an
elliptical trajectory in state space, the
properties of which are defined by its
initial position and momentum.
Consider first the kind of deterministic system to which we are most accustomed, for example, one for which the

20
Downloaded
from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1557/S0883769400037131

MRS BULLETIN/JULY 1995

Deterministic Chaos Theory and Its Applications to Materials Science

450

460

470

480

490

500

Figure 1. Numerically generated solution of Equation 1, illustrating its chaotic


(aperiodic) behavior. The solution was actually begun at time t = 0 but is shown only
for the later times, allowing time for transients to die out.

dynamics can be described in terms of


linear combinations of the state variables.
If a system of this sort is started from
two slightly different positions in state
space, the two trajectories will either remain close to each other as time progresses or will approach one another.
Some damping must be present in order
for the two trajectories to converge. The
rate of approach is exponential with
time, the coefficient of time in the exponent being called the Lyapunov exponent,
which is negative for cases such as this.
Actually, there is one Lyapunov exponent associated with each of the state
variables and, for this kind of system,
each of these is either zero or negative.
A second kind of system is one that
behaves stochastically, that is, in a truly
random manner. Here, trajectories
started from two adjacent positions in
state space have no correlation whatsoever at future times.
There is yet a third possibility: a system that is deterministic, but for which
one of the Lyapunov exponents is positive. In such cases, there is still an attractor as long as the system is damped, but
the trajectories of the system from adjacent starting points now will locally
diverge. This system is said to exhibit
deterministic chaos. The attractor for a
chaotic system is fractal, that is, it has a
dimension that is not an integer. In certain cases, it is possible for more than one
Lyapunov exponent to be positive; this is
known as hyperchaos, a subject that will
not be dealt with here.
An important characteristic of a cha-

otic system, which results from the existence of one or more positive Lyapunov
exponents, is the extreme sensitivity of
its dynamical evolution to the initial conditions. Even a very slight difference in
initial conditions, all else being equal,
results in very large differences in the instantaneous state of a chaotic system in
the long term. This feature is illustrated
in the next section with a numerical example. A second important property following from the first is that the behavior
of the system is intrinsically unpredictable.

"Intrinsically" here means that it does


not matter how accurately the state of the
system can be measured at any time; in
the long range, the evolution of the system becomes unpredictable. Predictions
can be made over limited intervals of
time, the duration of which depends
upon the accuracy of the initial state's
definition and the magnitude of the
largest positive Lyapunov exponent.
It should be emphasized that chaotic
behavior may be unpredictable, but it is
not random in a stochastic sense. Indeed,
it has an underlying order that can be
measured and quantified. This order is
demonstrated in conjunction with one
of the examples discussed in the next
section.
For chaos to occur, certain conditions
must be satisfied. First, the system must
be nonlinear in terms of its state variables.
This means that its dynamical behavior
is described in terms of the state variables in forms other than to the first
power, or in terms of products or quotients of different state variables. Second,

it must have at least three independent


state variables. These conditions are necessary, but not sufficient, for the occurrence of chaos. For example, a system
that satisfies both conditions may never
become chaotic, or it may be chaotic only
within certain ranges of values for its
state variables or its characteristic parameters (e.g., the reaction-rate parameters for a model of a chemical system).
Chaotic behavior may be undesirable
or desirable. It is undesirable in systems
where predictable behavior is important,
as in an electric-power generating system. It is desirable where good mixing is
important, as in chemical reactors.
Finally, we note that chaos is a robust
phenomenon. It is ubiquitous in its
occurrence, and it can exist over large
ranges of parameter values.
Two Examples of Deterministic
Chaos
One simple model of chaos which illustrates its usefulness is that of kneading dough containing raisins. Suppose,
for simplicity, that two raisins are initially close together. The dough is rolled
out to twice its original length, folded,
and rolled again. This process is then repeated. When the stretching begins, the
distance between the raisins immediately begins to increase. Folding the
dough has no influence on the distance
between the raisins as long as they are
both situated on the same side of the fold.
However, when they are separated by
about half the length of the original
dough, the next fold will likely occur between them, which will actually bring
them closer together. Subsequent stretching moves them further apart again.
Thus, continual mixing occurs. Extending this concept to many raisins shows
that this chaotic process (note that it is
completely deterministic,.) is required to
develop the desired uniform distribution
of raisins throughout the dough.
Another example is represented by the
following differential equation, which is
a specific case of what is known as the
"Duffing oscillator":
d2x/dt2 + 0.05dx/dt + x3 = 7.5cos(f).

(1)

Equation 1 describes the displacement


x of a one-dimensional, forced damped
oscillator as a function of time t (both in
dimensionless units). The only difference between this expression and that
which a sophomore engineering student
can solve in closed form is that the displacement-dependent restoring force
here is cubic in x (that is, a "stiffening
spring") rather than linear. This par-

Downloaded from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms of use, available at
MRS BULLETIN/JULY 1995 http://dx.doi.org/10.1557/S0883769400037131
http:/www.cambridge.org/core/terms.

21

Deterministic Chaos Theory and Its Applications to Materials Science

ticular equation has been discussed


elsewhere 9 in considerable detail. It
represents a nonlinear system with three
independent state variables: position, velocity (or momentum), and the phase of
the driving force. Thus, it satisfies the
criteria (necessary but not sufficient) for

the occurrence of chaos. A numerically


generated solution is presented in Figure 1. The initial conditions that were selected were x = 0 and dx/dt = 0 at t = 0,
although the solution is shown beginning at a later time to allow transient behavior to die out. Clearly, we no longer

450

500

Figure 2. Shown here is the solution presented in Figure 1 (solid curve) plus a
second solution (dashed curve) obtained by changing the value of the displacement
x at the initial time t = 0 from x = 0 to x = 7 x 10~s. Despite the very small change
in this initial condition, the two solutions are, in the long term, quite different, which
illustrates the strong dependence of a chaotic system upon initial conditions.

have the simple, sinusoidal variation of


displacement with time that, for example, would characterize the long-term
solution of Equation 1 if the x3 term were
simply replaced with x. Instead, the motion is quite complex, although some patterns can be discerned. This is indeed an
example of deterministic chaos, the
existence of which could be quantitatively demonstrated by calculating the
Lyapunov exponents.
To illustrate the strong dependence of
long-term behavior on initial conditions,
was mentioned in the previous section,
consider what happens when one of the
initial conditions from the previous
paragraph is changed ever so slightly. Let
the displacement at t = 0 be changed
from x = 0 to x = 1 X 10 b. The "longterm" solution is shown in Figure 2 and
is compared with that from Figure 1.
Clearly, the two solutions are quite different from one another.
In Figure 3, the velocity v = dx/dt is
plotted versus the displacement x for the
example shown and the interval covered
in Figure 1. This figure is thus a projection of the chaotic trajectory onto the x-v
plane in state space with the trajectory
lying on the attractor.
Chaotic behavior actually has an underlying order. To illustrate that order,
we again consider the trajectory of the
oscillator given by Equation 1 as projected onto the x-v plane. However, contrary to what is shown in Figure 3, which
includes every point on the trajectory, we
now include only those points which occur at integer multiples of the period of
the forcing function. In other words,
only points that occur every 2TT units of
time are included. The result, called a
Poincare section, is plotted in Figure 4.
This figure contains about 15,000 points
and therefore covers a much larger interval than that used to generate Figures 13. The complex, layered structure that
can be seen results from "stretching"
and "folding" actions, as in the previous
example of stretching and folding dough.
The Poincare section can be regarded
as a kind of cross section of the chaotic
attractor, which vividly illustrates its
extremely complex but nonrandom
structure.

Two Ways of Studying Dynamic


Systems
-4

Figure 3. A state-space plot ofx versus v = dx/dt for the solution of Equation 1
shown in Figure 1. This trajectory lies on the chaotic attractor.

There are two ways of looking at a dynamical system which may exhibit deterministic chaos. The first is that adopted
by Lorenz10 and by many others since
that time: A set of differential equations
is chosen that is believed to describe the
system in question. Solutions of these

Downloaded from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms of use, available at
22
MRS BULLETIN/JULY 1995
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1557/S0883769400037131

Deterministic Chaos Theory and Its Applications to Materials Science

equations are then obtained by numerical integration, in which the evolution of


the state variables is described as a function of time. For a system capable of exhibiting chaos, there will be a range of
values of the variables within which the
behavior of the system is classical, that is,
oscillating in a period-one orbit characterized by a single closed loop in its state
space. Then, as some system parameter
is changed (e.g., the electrode potential
for an electrochemical cell), a point may
be reached at which the solution splits
into two, that is, it bifurcates into a periodtwo orbit characterized by a double loop
in its state space. Each branch eventually
bifurcates again as the parameter is
varied still further, and the system finally becomes chaotic. This is called the
period-doubling route to chaos. Other
routes to chaos do exist, but discussion of
these is beyond the scope of this brief
article.
It is important to recognize that the
behavior just described is strictly that of
the set of equations. It may or may not
describe the actual behavior of the dy-

namical system the equations were selected to describe. In addition, the


method used to discretize the equations
so that they can be numerically integrated may affect the qualitative nature
of the solution, a potentially serious
problem which is discussed further in
the next section.
The second approach to looking at a
dynamical system is to study the experimental behavior of the system in question, to determine whether or not chaos,
or its precursor (bifurcation), is observed. This is far from simple. The most
common approach is to use what is
called the time-series method. The value
of a single dynamical variable w (or perhaps some function of a number of the
dynamical variables) is measured at
equidistant intervals of duration At. This
yields a time series of the form
Wo, U>i, W2,

Wn

(2)

where n is typically the order of several


tens of thousands and includes many
"peak-to-peak" oscillations of the sys-

tern variables. The manner in which w


changes with time actually contains information regarding the other dynamical
variables. One way such information can
be extracted is to construct a set of points
in a d-dimensional Cartesian space using
the time-series data given in Expression 2. These points have the form
{Wi, Wi + m, W, + 2w,

, WiUd-\)m\ W i t h

/ = 0,1, 2

(3)

where both d and m are positive integers.


As long as d and m are appropriately chosen (neither choice being a simple task),
the set of points in Expression 3 has characteristics that are similar to those of the
N independent dynamical variables of
the system. Without going into detail
here, this method, although powerful, is
full of traps and has been considerably
misused in the recent past. Nevertheless,
it is often used to test real systems for
chaotic behavior.
Another useful approach is to determine the magnitude and arithmetic sign
of the largest Lyapunov exponent of the
system, checking first to see that it does
indeed have a positive value. Its magnitude is a quantitative measure of just
how chaotic the system is.

Pitfalls in Studying Chaos

1.5

Both of the methods just described for


studying chaos have their pitfalls. The
trap in the first case, as already indicated, is that the calculated chaotic behavior may be a property of the set of
equations but not of the physical process
that the equations were chosen to describe. This problem is not necessarily
easy to detect. The chosen set of equations may be quite successful in describing the behavior of the system in the
classical regime but may predict bifurcation and chaos which are not observed in
practice. The second case is also fraught
with traps, as already noted.
As mentioned in the previous section,
a common example of one pitfall is the
discretization of continuous differential
equations to finite-difference equations
in order to obtain the solution numerically. To illustrate, consider the following
continuous differential equation for
some variable y:
dy/dt = ry{\ - y)

Figure 4. A Poincare section of the chaotic attractor for the Duffing oscillator
described by Equation 1. After initial transients have died away, values of position x
and velocity v are plotted each time the phase of the driving force is an integer
multiple of 2% that is, at times t n = n2vr where n is a positive integer. About
15,000 points are shown, so that a much larger interval is covered than that used to
generate Figures 1-3.

(4)

where y(0) = y0 with 0 < y0 <1 and


where r is a positive constant. The solution of Equation 4 can be obtained in
closed form; it is well-behaved and
nonoscillatory over the range of y of interest: yti y <\.\n particular, it does

Downloaded from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms of use, available at
MRS BULLETIN/JULY 1995 http://dx.doi.org/10.1557/S0883769400037131
http:/www.cambridge.org/core/terms.

23

Deterministic Chaos Theory and Its Applications to Materials Science

not, indeed it cannot, exhibit chaos. Sup-

pose, however, that a solution is obtained


numerically using the following approximation method:
y(t + St) = y(t) + ry(t)(l-y(t))8t

(5)

where 8t is the increment of time selected


for this finite-difference calculation.
Now, if we denote y(t) as yn and y(t + 8t)
as y,,H where n = 0,1,2...., we obtain
y,, +1 = yn + (rSt)yn (1 -

yn)

(6)

which actually would be the result of applying Euler's method to obtain the solution of Equation 4. Equation 6 has the
form of a map, in the sense that the value
of a function at some particular time is
used to generate its value at some discrete time interval later. In fact, this particular map is a form of what is called the
logistic map, which is used to describe
changes in the populations of animals
for which the generations do not overlap.
The important thing about this particular map is that it does exhibit chaos for a
certain range of values for rSt. Thus, if
the behavior of the system is accurately
described by the continuous differential
equation, its transformation to this
finite-difference equation leads to the
apparent result that the variation of y
with time may become chaotic, when in
fact this is not the case.
Equations for which this transformation is possible are common in materials
science. Kirkaldy11 has discussed the
equation that relates the entropy production rate per unit volume of the isothermal lamellar Fe/Fe3C eutectoid (pearlite)
reaction by volume diffusion in terms of
the lamellar spacing and the transformation-front velocity. He shows that this
can be expressed in a manner that is
essentially the same as the equations described here, and that by a transformation to a finite-difference form, a
transition to chaos is predicted that
Kirkaldy suggests has a form consistent
with the change from pearlite to upper
bainite. Kirkaldy is aware of this trap
and remarks that "It is a normal reaction
to suspect such proceedings of theoretical illegality," but argues on physical
grounds that, for transformations occurring at sufficiently high supersaturations,
the finite-difference form seems more
reasonable, and that the predictions of
bifurcation and chaos are consistent with
experiment. Akuezue and Stringer 12
have shown that the equations for solidstate diffusion can also be transformed
in the same way, and that bifurcations
and a transition to chaos are formally

networks, which appear to work well


even with considerable amounts of
superimposed noise.14
Exactly what one would want to predict, relative to a chaotic system, depends
upon the application at hand. Much of
the work carried out to date (e.g., Brawley
et al.14) has focused on the following
problem: Given any k consecutive points
in a chaotic time series w,, a>, + i, ,
Wit(i-i), predict the value of the next
point w,, k. The magnitude selected for k
depends upon such factors as the size of
the time interval between adjacent
points and the magnitude of the largest
Lyapunov exponent.
Other types of predictions wouid
also be of interest. For example, some
chaotic systems (e.g., McCoy et al.,13
Parmananda et al.16) are characterized by
Current Areas of Interest in
"mixed-mode" oscillations, that is, oscilNonlinear Dynamics
lations that have a mixture of large and
small amplitudes. In some instances, the
Research areas currently of interest in
chaotic dynamics can be broadly sepa- large-amplitude oscillation may be detrimental to the integrity of the system.
rated into three general categories: the
analysis of chaotic time series, prediction Therefore, prediction of the time of onset
of future chaotic behavior based on the for the next such excursion would allow
past, and exertion of control over a cha- for steps to be taken to prevent its occurrence. Another example consists of a meotic system.
chanical chaotic system for which the
potential energy is characterized by a
Analysis
"double-well" function, that is, a funcThe analysis of a chaotic time series
tion having two stable potential-energy
can involve a variety of things, such as
minima. For sufficiently large amplithe determination of appropriate values
for the parameters m and d in Expres- tudes of oscillation, the trajectory may
chaotically cross back and forth from one
sion 3. As fundamental as this may
potential well to the other. Suppose, for
seem, there still exists considerable conwhatever reason, that one of these wells
troversy13 over what is the best way to
needs to be avoided. Then, a useful premake such determinations. Added to
diction would be the time that the next
these and other problems are some
crossover to that well will occur so that
"real-life" factors, such as data sets of
some measure can be initiated to prevent
limited size and the presence of a
its occurrence.
stochastically varying signal superimposed on the deterministic chaos. In the
latter case, one task is to extract as much
Control
information as possible regarding the
The general idea of somehow controlchaotic component.
ling chaos is a topic that has generated
intense interest since having first been
Prediction
introduced by Ott et al.17 several years
Prediction of future behavior, given
ago. This interest is illustrated by the fact
the past, is one of the classic problems of
that a relatively complete, continuously
science. As we have already indicated,
updated bibliography on control and
prediction of chaotic behavior is imsynchronization of chaotic systems is acpossible in the long term. However, cessible via the Internet.18 It presently has
short-term predictions are possible, the
over 280 listings, almost all the articles
fundamental limit being (,/k where A is listed having been published after 1990.
the largest Lyapunov exponent and is a
The original intent was to identify an
positive real number of modest magniapproximately periodic orbit within the
tude. Actually, the magnitude of dechaotic attractor and then to periodically
pends on how well the current state of
apply a small stimulus that would enthe system is known and how accurate a
courage the system to remain on that orprediction is desired. Several approaches bit. The concept of control has since been
have been used to make these pre- extended to stabilize relatively complex
dictions. One that seems particularly
orbits (period-two or higher) or to elimipromising is the application of neural
nate oscillations altogether by using a
impossible for binary diffusion couples,
but such behaviors may arise in ternary
diffusion systems under special conditions which may be satisfied near phase
boundaries. Again, it appears that substantial departures from equilibrium
are required for these behaviors to be
observed.
From experience to date, one can only
conclude that great caution must be exercised in the construction and use of
mathematical models that purport to describe chaotic dynamics in physical systems. One must be certain that chaos
predicted using a set of equations is indeed physically realistic and not simply
an artifact of the equations themselves
nor of the numerical approach used to
integrate the equations.

Downloaded
from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms
of use,
available at 1995
MRS
BULLETIN/JULY
24
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1557/S0883769400037131

Deterministic Chaos Theory and Its Applications to Materials Science

0.1180

0.1170

00 0.1160

0.1150

0.1140

0.1

0.2

1
0.3

1
,
0.4

'
0.5

0.6

0OH
Figure 5. A state-space plot of 0OH versus 0O for chaotic oscillations of the corrosion
model.15 As in Figure 3, this trajectory lies on the chaotic attractor. The x marks the
location of an unstable steady state, as projected onto the 6OH-0O plane.

small control signal to stabilize an otherwise unstable steady state, causing the
system to move to this steady state. On
the other hand, if chaos is beneficial, one
would be interested in a control that
would ensure that the state-space trajectory would remain on the attractor and
would compensate for the slow "drift" of
parameters that might eventually take
the system to a nonchaotic regime.
It is worth realizing that a chaotic system plus control is fundamentally different from a system that is free or
uncontrolled. Be it a map-based or continuous control, the attractor for the controlled system is a simple or complex
periodic orbit or a steady state. Moreover,
the control is generally chosen such that
its magnitude decreases as the new attractor is approached and actually vanishes when the system reaches the
attractor. At that point, the system is free
again, its tendency being to leave the
controlled attractor and return to the
chaotic attractor. As soon as it starts to
do so, however, the control is reinitiated
and the system is restored to the controlled configuration.
Some examples of chaos control, pertaining specifically to an electrochemical
model and an actual electrochemical

cell, are presented in the section, An Example: Nonlinear Dynamics and Chaos
in Metal Corrosion.
To close this section, we draw attention to a comprehensive review by Abarbanel et al.19 on dealing with chaotic data
in physical systems. Tools available for
obtaining physically interesting information from such data are discussed.

Prevalence of Nonlinear
Dynamical Effects in
Materials Behavior
How can we expect the "revolution" in
nonlinear science to impact people who
are trying to understand materials and
their performance characteristics? The
answer is that the impact is already being felt, which is not surprising since so
many aspects of materials behavior are
inherently nonlinear. Actually, a survey
of literature on the subject indicates that
"nonlinear science" first had a major influence on the study of materials during
the late 1980s. Some materials-related
subjects in which nonlinear effects are
presently known to be important are the
following: plastic deformation, fracture,
magnetism, superconducting and semiconducting properties, phase transformations, solid-state diffusion, fluid

dynamics, heat flow, solidification, surface reactions and catalysis, and general
and localized corrosion. These activities
have involved more than just the study of
chaotic dynamics. For example, sudden
changes in the behavior of a material as
one or more input parameters is varied
(such as the sudden appearance of a plastic instability as the applied stress is increased) can be described using the
well-established formalism of bifurcation theory. Likewise, the stability characteristics associated with dynamical
aspects of material behavior can be investigated using linear-stability theory.
This latter approach can also be used to
determine the conditions under which a
bifurcation will occur, as well as the nature of the bifurcation itself.

An Example: Nonlinear
Dynamics and Chaos in
Metal Corrosion
As an illustration of materials-related
chaos, we consider one example: an investigation of the corrosion of a metal
surface exposed to an aqueous medium.
This particular work is selected for discussion because it involves both of the approaches to studying chaos that were
described earlier: (1) use of a set of equations to describe a physical process, and
(2) execution of actual experiments. This
example can be regarded as a prototype
of a nonlinear system. The techniques
that have been developed to study or alter its dynamical behavior, some of which
are described here, are broadly applicable
to chaotic systems in general.
For the modeling studies, certain aspects of two earlier aqueous corrosion
models were combined to yield a single,
highly nonlinear model of anodic dynamics (details having been presented
elsewhere15). The model has three independent state variables: the fractional
coverage on the metal surface of a hydroxide film and a passivating oxide film
0OH and 0O, respectively, and the concentration Y of metal (in dimensionless
units) dissolved in the ambient, wellstirred solution. We note immediately
that the model satisfies the necessary but
not sufficient conditions for chaos. By
changing certain of the rate constants
that characterize the various chemical
reactions, it was found that the model exhibited a range of behaviors that were
amazingly complex. Included among
these were the existence of multiple
steady states, hysteresis, spontaneous
mixed-mode oscillations of the state
variables (for constant anodic potential),
period doubling, coexisting attractors,
and chaos attained by several different

Downloaded from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms of use, available at
MRS BULLETIN/JULY 1995 http://dx.doi.org/10.1557/S0883769400037131
http:/www.cambridge.org/core/terms.

25

Deterministic Chaos Theory and Its Applications to Materials Science

routes. The chaotic trajectory, corresponding to a particular set of rateconstant values, is shown in Figure 5 as
projected onto the 9OH-9O plane of the
three-dimensional state space. This trajectory runs in a counterclockwise sense
for this particular projection and lies on
the chaotic attractor, the calculated fractal dimension of which is 2.035. (The
chaos-prediction studies of Brawley
et al.14 mentioned previously were carried out using this model, although the
prediction strategy itself is broadly applicable to other nonlinear systems.)
Experimental studies were carried out
using a rotating copper-disk anode exposed to a sodium-acetate/acetic-acid
buffer solution. Parameters that could be
varied, to seek different behaviors, included the anodic potential and the diskrotation speed. Among the behaviors
observed were periodic oscillations,20
mixed-mode oscillations,16 and deterministic chaos. ",,21
An algorithm with which to control
chaos in either a theoretical model or an
actual experiment was derived and then
applied to both the corrosion model22
and the experimental cell.23 The algorithm is a map-based strategy called "recursive proportional feedback" (RPF).
This approach to chaos control was
found to be quite powerful and was successful in achieving control even in cases
in which other methods failed. It consists
of periodically applying very small perturbations to the anodic potential. This
causes an otherwise chaotic system to
oscillate periodically.
An illustration of RPF-induced control
in the electrochemical cell is presented
in Figure 6. Plotted here are the minima of
the measured anodic-current oscillations, both with and without the control
being applied. The system oscillates chaotically when no control is being applied.
The perturbations applied to the otherwise constant anodic potential to achieve
and maintain control are also shown.
Clearly, the control acts quickly and reproducibly to eliminate the chaotic oscillations and maintain the system in a
periodically varying manner. Moreover,
the system quickly moves back to its chaotic oscillations when the control is
turned off.
This same team of investigators has
developed other strategies for controlling
chaos. Included among these is the use
of adaptive learning,24 in which a neural

network is used to continually update


the method of control for cases in which
certain of the system parameters undergo a slow drift. Another control strategy involves feedback to the anodic

5.0

4.0
<

E
3.0
.*"

on

H^-

off

+ 1.0
-0.5

100

200
Time (sec)

300

400

Figure 6. Minima of anodic current oscillations in an electrochemical cell, with and


without applied control. Control is achieved using the RPF algorithm.22'23 The perturb ations added to the anodic potential to achieve and maintain control are also shown.

Figure 7. State-space trajectory of corrosion model,15 with and without derivative


control.25 The sequence of events goes from the thin solid line to the thick solid line
to the dashed line. Thin solid line: trajectory on chaotic attractor with no control.
Thick solid line: derivative control is applied so system leaves chaotic attractor and
goes to the now-stabilized steady state shown in Figure 5. Dashed line: control is
switched off so system leaves steady state, which is once again unstable, and
spirals its way back to the chaotic attractor.

Downloaded
from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms
of use,
available at 1995
26
MRS
BULLETIN/JULY
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1557/S0883769400037131

Deterministic Chaos Theory and Its Applications to Materials Science

potential of the time derivative of a measurable quantity (for our example, the
anodic current). This approach is used to
eliminate oscillatory dynamics altogether by stabilizing an otherwise unstable steady state of the nonlinear
system. Its use may necessitate the addition of a new state variable to the system.
To date, this "derivative-control" strategy has been used to control chaos in the
electrochemical model and to eliminate
periodic oscillations in the electrochemical cell.25 The state-space trajectory,
shown in Figure 7, vividly demonstrates
the result of applying this control strategy to the corrosion model. Other forms
of control are also being investigated, including the application of classic methods of linear and nonlinear feedback.
Practical Applications
These strategies for predicting and
controlling chaotic dynamics are all very
interesting, but of what practical use can
they be? Answers to this question are
just beginning to emerge and the future
is promising. For example, it is now recognized that properly engineered chaos
control can be used to obtain dynamical
behavior that heretofore had been impossible to achieve. We are just beginning to
explore the possibilities that are offered
by these new findings.
Consider one example: chaos in a
model of pulse combustion, for which it
was found26 that control could indeed be
achieved. One approach that was used
was to combine an adaptive learning
strategy with the RPF algorithm,
whereby chaotic oscillations were suppressed and the useful operating range
of the pulse combustor was extended
well beyond that for a system having no
applied control. In addition, a derivative
control strategy was used to eliminate
oscillations entirely by stabilizing an
otherwise unstable steady state.26 A major next step would be to apply these
methods to a "real" combustor to ascertain exactly what benefits would accrue.
It is in applications such as these that
prediction and control of chaos will become practical and important tools, not
only in materials-related dynamical behavior, but throughout the many subject
areas in which nonlinear effects play a
major role.
Conclusion
In closing, we note some excellent advice for novices that was offered by the
distinguished chemist E. Bright Wilson:
"If there be any readers who are not already familiar with this set of phenomena, somewhat recently discovered, I

9. J.M.T. Thompson and H.B. Stewart, Nonlinear Dynamics and Chaos Geometrical
Methods for Engineers and Scientists (Wiley,
Chichester, U.K., 1986) p. 3.
10. E.N. Lorenz, "Deterministic Nonperiodic
Flow," /. Atmos. Sciences 20 (1963) p. 130.
11. J.S. Kirkaldy, "Deterministic Chaos and
Eutectoid Phase Transformations," Scr.
Metall. Mater. 24 (1990) p. 179.
12. H.C. Akuezue and J. Stringer (unpublished manuscript).
13. G.W. Frank, T. Lookman, and M.A.H.
Nerenberg, "Recovering the Attractor: A Review of Time-Series Analysis," Can. ]. Phys. 68
(1990) p. 711.
14. G.H. Brawley, A.J. Markworth, and P. Parmananda, "Use of Neural Networks to Predict
the Short-Term Behavior of Chaotic Time Series, Including Effects of Superimposed
Noise," in Proc. 26th Southeastern Symp. on
System Theory (IEEE Computer Society Press,
Los Alamitos, CA 1994) p. 643.
15. J.K. McCoy, P. Parmananda, R.W. Rollins,
and A.J. Markworth, "Chaotic Dynamics in a
Model of Metal Passivation," /. Mater. Res. 8
(1993) p. 1,858.
Acknowledgments
16. P. Parmananda, H.D. Dewald, and R.W.
Much of the research referred to in
Rollins, "Mixed-Mode Oscillations in the
this article (specifically, References 14-16
Electrodissolution of Copper in Acetic Acid/
Acetate Buffer," Electrochim. Acta 39 (1994)
and 20-26) was carried out with the supp. 917.
port of the Electric Power Research In17. E. Ott, C. Grebogi, and J.A. Yorke, "Constitute (EPRI) of Palo Alto, California
trolling Chaos," Phys. Rev. Lett. 64 (1990)
under EPRI Contract RP2426-25 and
p. 1,196.
with the guidance of one of the authors
18. G. Chen, "Control and Synchronization of
(J.S.). This work is still in progress and is
Chaotic Systems (A Bibliography)," Electrical
being conducted jointly by investigators
Engineering Department, University of Housat The Ohio State University in Columton, Texas; available by ftp: "uhoop.egr.
bus and Ohio University in Athens. Most
uh.edu/pub/TeX/chaos.tex" (login name and
of the work summarized here was compassword: both "anonymous").
pleted while one of the authors (A.J.M.)
19. H.D.I. Abarbanel, R. Brown, J.J. Sidorowich, and L.Sh. Tsimring, "The Analysis of
was a full-time staff member at the BatObserved Chaotic Data in Physical Systems,"
telle Memorial Institute in Columbus.
Rev. Mod. Phys. 65 (1993) p. 1,331.
20. H.D. Dewald, P. Parmananda, and R.W.
References
Rollins, "Periodic Current Oscillations in the
1. A.D. Beyerchen, "Nonlinear Science and
Anodic Dissolution of Copper in Acetate
the Unfolding of a New Intellectual Vision,"
Buffer," /. Electroanal. Chem. 306 (1991) p. 297.
Papers in Contemporary Studies 6 (1989) p. 25. 21. H.D. Dewald, P. Parmananda, and R.W.
2. A.L. Kawczyriski, W. Raczyriski, and B.
Rollins, "Periodic and Chaotic Current OscilBaranowski, "Analysis of Chaotic Oscillations
lations at a Copper Electrode in an Acetate
in a Simple Electrochemical System," Z. Phys. Electrolyte," /. Electrochem. Soc. 140 (1993)
Chemie Leipzig 269 (1988) p. 596.
p. 1,969.
3. J. Wisdom, "Is the Solar System Stable? and
22. R.W. Rollins, P. Parmananda, and P. SherCan We Use Chaos to Make Measurements?",
ard, "Controlling Chaos in Highly Dissipative
in CHAOS/XAOC Soviet-American PerSystems: A Simple Recursive Algorithm,"
spectives on Nonlinear Science, edited by D.K. Phys. Rev. E 47 (1993) p. R780.
Campbell (American Institute of Physics, New
23. P. Parmananda, P. Sherard, and R.W.
York, 1990) p. 275.
Rollins, "Control of Chaos in an Electrochem4. J. Gleick, ChaosMaking a New Science
ical Cell," Phys Rev. E 47 (1993) p. R3,003.
(Viking, New York, 1987).
24. M.A. Rhode, R.W. Rollins, and CA. Vas5. J.M.T. Thompson and H.B. Stewart, Nonlinsiliadis, in "Adaptive Learning to Control
ear Dynamics and ChaosGeometrical Meth- Chaos," Proc. 26th Southeastern Symp. on Sysods for Engineers and Scientists (Wiley,
tem Theory (IEEE Computer Society Press, Los
Chichester, U.K., 1986).
Alamitos, CA, 1994) p. 638.
6. F.C. Moon, Chaotic VibrationsAn Intro25. P. Parmananda, M.A. Rhode, G.A. Johnduction for Applied Scientists and Engineers
son, R.W. Rollins, H.D. Dewald, and A.J.
(Wiley, New York, 1987).
Markworth, "Stabilization of Unstable Steady
7. F.C. Moon, Chaotic and Fractal DynamStates in an Electrochemical System Using
icsAn Introduction for Applied Scientists
Derivative Control," Phys. Rev. E 49 (1994)
and Engineers (Wiley, New York, 1992).
p. 5,007.
8. J.H. Kim and J. Stringer, eds., Applied Chaos 26. M.A. Rhode, R.W. Rollins, A.J. Mark(Wiley, New York, 1992).

hope that these few sentences will inspire them to rush out and learn more
because chaos is very likely to turn up in
whatever field they work in."27
Indeed, we have already discovered
that chaotic dynamics can occur in a
plethora of materials-related phenomena.
The challenge that lies before us is to
learn, in each application, how to take
advantage of its occurrence to enhance
the desired dynamical behavior of the
system in question. Depending upon the
specific application, this may involve the
use of control strategies to maintain
chaos, to force an otherwise chaotic system to oscillate periodically in judiciously chosen simple or complex orbits,
or to eliminate oscillatory behavior altogether. In addition, prediction of chaotic
behavior, even though it is limited to the
short term, can be used in a variety of
ways to enhance system performance.

Downloaded from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms of use, available at
http:/www.cambridge.org/core/terms.
MRS BULLETIN/JULY 1995 http://dx.doi.org/10.1557/S0883769400037131

27

Deterministic Chaos Theory and Its Applications to Materials Science

worth, K.D. Edwards, K. Nguyen, OS. Daw,


and J.F. Thomas, /. Appl. Phys. in press.
27. E. Bright Wilson, "One Hundred Years of
Physical Chemistry," Am. Scientist 74 (1986)
p. 70.
Alan J. Markworth is a professor in the
Department of Materials Science and Engineering at The Ohio State University in Columbus. Prior to pining the Ohio State
faculty in 1995, he was a member of the research staff at Battelle Memorial Institute in
Columbus for nearly three decades, working
primarily in the area of computational materials science. He holds a PhD in physics from
Ohio State. For the past several years, he has
been particularly active in the application of
nonlinear dynamics to materials problems
related to the electric utilities, concentrating
on applications in the areas of corrosion,
combustion, and fracture. Markworth can be
reached at the following phone number: 614688-3581.
John Stringer is a technical executive in applied science and technology at the Electric
Power Research Institute in Palo Alto, CA.
Prior to joining the Institute in 1977, he was

head of the Department of Metallurgy and


Materials Science at the University of Liverpool and also worked at Battelle Memorial
Institute in Columbus as a Fellow in the
Metal Science Group. He holds PhD and
DEng degrees from Liverpool. His research
interests have been primarily in the areas of
high-temperature oxidation and corrosion of
metals and alloys, galvanomagnetic effects in
alloys, and erosion and corrosion of components influidized bed combustors.
Roger W. Rollins is a professor of physics
and a member of the Condensed Matter and
Surface Science Program at Ohio University
in Athens. He holds a PhD in applied physics
from Cornell University. His research interests were centered on experimental aspects of
superconductivity until 1980, when he became involved in nonlinear dynamics and
chaos. His current studies include investigation of a variety of approaches to controlling
chaotic dynamics, with applications in several areas, including aqueous-corrosion and
combustion systems. Another interest involves the development of interactive software for the study of chaotic systems.

Advertisers in This Issue


Page No.
Academic Press, Inc.

49

Elchema
36
Epion Corp.
High Voltage Engineering
Europa BV
inside front cover
Huntington Labs.
outside back cover
MDC Vacuum Products Corp.
10
n&k Technology, Inc.
New Focus, Inc.
inside back cover
Oxford Instruments
Philips Analytical X-Ray
28
Virginia Semiconductors, Inc.
VLSI Standards, Inc.
Voltaix, Inc.
John Wiley & Sons, Inc.
47
For free information about the products and
services offered in this issue, fill out and mail
the Reader Service Card, or FAX it to (312)
922-3165.

Silicon for Research


think small quantities
think thin ( 2 ^ ) or thick

When it comes
to small
diameter
silicon
requirements
"If VSI can't
make them,
you don't
need them."

. think small diameter (1" to 4")


. think single or double side polishing
think on or off axis for any orientation
think Virginia Semiconductor, Inc.,
your one source for all of the above.
A 3" diameter 2-4 /y membrane
manufactured at
Virginia Semiconductor, Inc.

1501 Powhatan Street, Fredericksburg, VA 22401


PHONE (703) 373-2900

FAX (703) 371-0371

Circle No. 11 on Reader Service Card.


Downloaded from http:/www.cambridge.org/core. * DUPLICATE DO NOT USE * USE BID 79433, on 08 Sep 2016 at 08:29:31, subject to the Cambridge Core terms
of use,
available at
MRS
BULLETIN/JULY
28
http:/www.cambridge.org/core/terms. http://dx.doi.org/10.1557/S0883769400037131

1995

S-ar putea să vă placă și