Introduction
When one uses matched asymptotic expansions, the solution is constructed
in differrent regions that are then patched together to form a composite
expansion. The method of multiple scales differs from this approach in that
it essentially starts with a generalized version of a composite expansion. In
doing this, one introduces coordinates for each region (or layer); those new
variables are considered to be independent of one another. A consequense
of this is that what may start out as an ordinary differential equation is
transformed into a partial differential equation. Exactly why this helps to
solve the problem, rather than make it harder, will be discussed as the
method is developed in this chapter.
The history of multiple scales is more difficult to delineate than, say,
boundary-layer theory. This is because the method is so general that many
apparently unrelated approximation procedures are special cases of it. One
might argue that the procedure got its start in the first half of the 19th
century. For example, Stokes (1843) used a type of coordinate expansion in
his calculations of fluid flow around an elliptic cylinder. Most of these early
efforts were limited, and it was not until the latter half of the 19ht century
that Poincar (1886), based on the work of Lindstedt (1882), made more
extensive use of the ideas underlying mutiple scales in his investigations
into the periodic motion of planets. He found that the approximation obtained from a regular
expansion accurately described the motion for only
a few revolutions of the planet, after which the approximation was way
off. The error was due, in part, to the contributions of the second term
of the expansion. He referred to this difficulty as the presence of a secular
term. To remedy the situation, he expanded the independent variable with
the intention of making the approximation uniformly valid by removing
the secular term. This idea is also at the heart of the modern version of
the method. What Poincar was missing was the introduction of multiple
independent variables based on the expansion parameter. This stop carne
much later; the most influential work in this regard has been by Kuzmak
(1959) and Cole and Kevorkian (1963).
Introductory Example
As in the last chapter, we will introduce the ideas underlying the method by
going through a relatively simple example. The problem to be considered
is to find the y(t) that satisfies.
Where
This equation corresponds to an oscillator (i.e., a mass-spring-dashpot)
with weak damping, where the time variable has been scaled by the period of the undamped
systern (see Exercise 1.6.10). This is the classical
example used to illustrate multiple scales. Our approach to constructing
the approximation comes from the work of Cole and Kevorkian (1963) and
Reiss (1971).
Regular Expansion
It appears that this is not a layer type of problem since E is not multiplying
the highest derivative in (3.1a). It seems, therefore, that the solution has a
regular power series expansion. So, assuming
These functions are plotted in Fig. 3.1. It is apparent that we have not been particularly successful
in finding a very accurate approximation of the
solution for 0 < t < infinite. This is because the second term in the expansion
is as large as the first term once Et ~ 1. For this reason, the second term is
said to be a secular term, and for the expansion to be well ordered we must
have Et << 1. The problem with our approximation is that the solution
(3.4) decays but the first term in (3.3) does not. The second term tries to
compensate for this, and in the process it eventually becomes as large as
the first term.
Another way to look at what is happening in the problem is to consider
the energy. By multiplying (3.1a) by the velocity y, integrating, and then
using the initial conditions (3.1b), one obtains the equation
Where:
The function H is the Hamiltonian and consists of the sum of the kinetic
and potential energies for the system. Thus, (3.5) shows that the energy
equals the amount included at the start minus whatever is lost due to damping.
To relate this to the multiple-scales expansion, recall that a system is
said to be conservative, or energy is conserved, if H is independent of t. On
the other hand, if d/d H < 0, so that energy is lost, then the system is dissipative.
Using these ideas, we have that (3.1) is dissipative but the problem
for Yo(t) is conservative. Therefore, the first term in the approximation in
(3.3) is a conservative approximation to a dissipative problem. This is one
of many situations where the multiple-scales method is generally needed.
Multiple-scales expansion
This problem has an oscillatory component of the solution that occurs on
a time scale that is 0(1) and also has a slow variation in the solution that
takes place on a time scales 0(1/E). To incorporate two time scales into the
problem, we introduce the variables.
and
where.
To simplify the notation, we use the symbol dt1 in place of d/dt1 ( and similarly
for dt2). The benefits of introducing these two time variables are not yet
apparent. In fact, one might argue that we have made the problem harder
since the original ordinary differential equation has been turned into a
partial differential equation. The reasons for doing this will become evident
in the discussion that follows. It should be pointed out that the solution
of (3.9) is not unique and that we need to impose more conditions for
uniqueness on the solution. This freedom will enable us to prevent secular
terms from appearing in the expansion (at least over the tine scales we are
using).
where
For the moment, the coefficients a0(t2) and b0(t2) are arbitrary functions
of t2 except that they are required to satisfy the initial conditions given
in (3.12b).
To go any farther, we need to determine infinite in the slow time scale (3.8).
To do this, note that in the 0(E) equation obtained from (3.11) we are
going to have the term dt1 y0. It is not hard to check that this will cause
a secular term to appear in the expansion. Howevwer, this can be prevented
if we make use of the dependence of Y0 on the time scale t2. In (3.11) the
only term available to us that can be used to prevent a secular term is .....
Balancing this with.... gives us infinite = 1. With this, and the
initial conditions in (3.9b), we get the following problem:
where
Substituting the solutions for y0 and y1 given in (3.12a) and (3.14a) shows
that to prevent secular terms in the expansion we must require that
Escalas mltiples
Introduccin
Cuando se utiliza expansiones asintticas, la solucin se construye
en regiones diferentes que son entonces parcheadas juntas para formar un compuesto
expansin. El mtodo de escalas mltiples difiere de este enfoque en que
esencialmente comienza con una versin generalizada de una expansin compuesta. En
haciendo esto, se introducen coordenadas para cada regin (o capa); esos nuevos
se considera que las variables son independientes entre s. Una consecuencia
de esto es que lo que puede comenzar como una ecuacin diferencial ordinaria es
transformada en una ecuacin diferencial parcial. Exactamente por qu esto ayuda a
resolver el problema, en lugar de hacerlo ms difcil, ser discutido como el
se desarrolla en este captulo.
La historia de las escalas mltiples es ms difcil de delinear que, digamos,
teora de la capa lmite. Esto se debe a que el mtodo es tan general que muchos
los procedimientos de aproximacin aparentemente no relacionados son casos especiales de la
misma. Uno
podra argumentar que el procedimiento tuvo su inicio en la primera mitad del 19
siglo. Por ejemplo, Stokes (1843) utiliz un tipo de expansin de coordenadas en
sus clculos de flujo de fluido alrededor de un cilindro elptico. La mayora de estos
los esfuerzos fueron limitados, y no fue hasta la segunda mitad del siglo XIX
que Poincar (1886), basado en el trabajo de Lindstedt (1882), hizo ms
uso extensivo de las ideas subyacentes a las escalas mltiples en sus investigaciones
en el movimiento peridico de los planetas. Encontr que la aproximacin obtenida a partir de
una expansin regular describe con exactitud el movimiento
algunas revoluciones del planeta, despus de lo cual la aproximacin fue
Ejemplo introductorio
Como en el ltimo captulo, presentaremos las ideas que subyacen al mtodo
pasando por un ejemplo relativamente simple. El problema a considerar
es encontrar el y (t) que satisface.
Dnde:
Esta ecuacin corresponde a un oscilador (es decir, un saliente de masa-resorte)
con amortiguacin dbil, donde la variable de tiempo ha sido escalonada por el perodo del
sistema sin amortiguar (vase el ejercicio 1.6.10). Este es el clsico
ejemplo utilizado para ilustrar escalas mltiples. Nuestro enfoque para construir
la aproximacin viene de la obra de Cole y Kevorkian (1963) y
Reiss (1971).
Expansin Regular
Parece que esto no es un tipo de capa de problema ya que E no se est multiplicando
la derivada ms alta en (3.1a). Parece, por lo tanto, que la solucin tiene un
ampliacin regular de la serie de potencia. Asumiendo
Estas funciones se representan en la figura 3.1. Es evidente que no hemos sido particularmente
exitosos en encontrar una aproximacin muy
solucin para 0 <t <infinito. Esto se debe a que el segundo trmino en la expansin
es tan grande como el primer trmino una vez Et ~ 1. Por esta razn, el segundo trmino es
ser un trmino secular, y para que la expansin sea bien ordenada, debemos
han Et << 1. El problema con nuestra aproximacin es que la solucin
(3.4) se desintegra pero el primer trmino en (3.3) no lo hace. El segundo trmino trata de
compensar esto, y en el proceso llega a ser tan grande como
el primer trmino.
Otra manera de ver lo que est sucediendo en el problema es considerar
la energa. Multiplicando (3.1a) por la velocidad y, integrando, y luego
utilizando las condiciones iniciales (3.1b), se obtiene la ecuacin
Dnde:
dnde.
Para simplificar la notacin, usamos el smbolo dt1 en lugar de d / dt1 (e igualmente para dt2). Los
beneficios de introducir estas dos variables de tiempo todava no
aparente. De hecho, se podra argumentar que hemos hecho el problema ms difcil
ya que la ecuacin diferencial ordinaria original se ha convertido en una
ecuacin diferencial parcial. Las razones para hacerlo se harn evidentes
en la discusin que sigue. Cabe sealar que la solucin
de (3.9) no es nico y que debemos imponer ms condiciones para
unicidad en la solucin. Esta libertad nos permitir
trminos de aparicin en la expansin (al menos sobre las escalas de
utilizando).
Al escribir la ecuacin (3.11), hicimos un esfuerzo para incluir slo los trminos
que podran contribuir a los dos primeros trminos de la expansin. De esto
y (3.9b), obtenemos el primer problema a resolver:
dnde
Por el momento, los coeficientes a0 (t2) yb0 (t2) son funciones arbitrarias
de t2, excepto que estn obligados a satisfacer las condiciones iniciales dadas
en (3.12b).
dnde
Esta es una aproximacin de primer trmino que es vlida hasta al menos Et = 0 (1).
Esta aproximacin y la solucin exacta se muestran en la figura 3.2, de
que es evidente que el resultado de escala mltiple es bastante bueno. De hecho, una
puede probar que se trata de una aproximacin asinttica uniformemente vlida de la
solucin para 0 <t <0 (1 / E) (vase el ejercicio 6).