Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
Lindsay A. Skinner
There are many ne books on singular perturbation theory and how to solve
singular perturbation problems. Many readers of this book will already be
familiar with one or more of them. What distinguishes this book from the
others is its rigorous development and rigorous application of the method of
matched asymptotic expansions. The point of view is that certain functions
have a certain structure for which this method is valid and these are precisely
the kinds of functions that arise in a wide variety of dierential equation and
integration problems.
This book is intended to serve primarily as a supplement or follow-up to
a typical rst year graduate course in asymptotic and perturbation analysis.
Hopefully it will also prove to be a valuable companion to all those who do,
or wish to do, rigorous work in the eld. The basic theory for the book is
presented in Chapter 1. Then there are four chapters in which this theory is
applied to a sequence of ordinary dierential equation problems. There are
a number of previously unpublished results. One of these is the uniformly
valid expansion at the end of Chapter 4 for a problem involving logarithms
once studied by L. E. Fraenkel. Another is the unexpectedly simple uniformly
valid Bessel function expansion established at the end of Chapter 3. All the
dierential equations chosen for study in the text are linear, but this is not
because of any limitation of the theory. Indeed, much has been done, and
much more can be done, in applying the theory to nonlinear problems, as
noted in Exercise 3.2, for example.
Another unique feature of this book is its inclusion of several Maple pro-
grams for computing the terms of the various asymptotic expansions that
arise in solving the problems. Developing these was a nice break from the
otherwise sometimes tedious hard analysis in the book. They could well have
just been designated as exercises, indeed a few have been, but at the same
time, until one gets familiar with symbolic programming, there is nothing like
having a few model programs around to show the way, or at least a possible
way. Studying the programs will also reveal some mathematical manipula-
tions that are not fully developed within the text itself.
v
vi Preface
2.1. Finally they should read Section 3.1, maybe work Exercise 3.1, and get
some experience with the Maple program for Problem C.
Lindsay A. Skinner
San Diego, California
Contents
4 Logarithm Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.1 Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.2 Problem F . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3 Problem G, Part 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.4 Problem G, Part 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.5 Problem G, Part 3 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.6 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
ix
x Contents
5 Oscillation Problems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.1 Problem H . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
5.2 Problem I, Part 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
5.3 Problem I, Part 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
5.4 Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Chapter 1
Uniform Expansion Theory
1.1 Introduction
10
Fig. 1.1 Graph of z(x, ) together with the approximations 1/(x + ), 1/(x + 2x3 ) and
c(x, ) when = 0.1.
N1
f (x) = f [n] (0)xn + O(xN ) (1.5)
n=0
1
d n
on the interval [0, 1], where f [n] (x) = n! dx f (x). Indeed, f (x) C N ([0, 1])
implies
1.2 Uniform Expansion Theorem 3
N 1
(x) = xN f (x) f [n] (0)xn (1.6)
n=0
N1
f (X) = f[n] (0)X n + O(X N ) (1.7)
n=0
and we say f (x, X) C N ([0, 1] [0, ]) if both f (x, X) and f(x, X) are in
C N ([0, 1] [0, 1]), where f(x, X) = f (x, 1/X).
We are going to be dealing with functions y(x, ) that are, for some o > 0,
expressible in the form y(x, ) = f (x, x/) for all (x, ) [0, 1] (0, o ], where
f (x, X) C ([0, 1] [0, ]). If y(x, ) is one of these functions, then for any
(0, 1) and any N 1,
N1
ON y(x, ) = (/x)n f [0,n] (x, ). (1.10)
n=0
f (x, X) X n f [0,n] (x, )
BN X N (1.11)
n=0
N 1
f (x, x/) (/x)n f [0,n] (x, )
CN N (1.12)
n=0
4 1 Uniform Expansion Theory
as x 0+ , where
M 1
IM y(x, ) = xm f [m,0] (0, x/). (1.14)
m=0
We call IM y(x, ) the M -term inner expansion of y(x, ). Note that it is the
rst M terms in the power series expansion of Y (X, ) = y(X, ) as 0+ ,
assuming X > 0, followed by the substitution X = x/. Applying this idea
to ON y(x, ) leads to
M 1
N1
IM ON y(x, ) = xm (/x)n f [m,n] (0, ). (1.15)
m=0 n=0
N 1
M 1
ON IM y(x, ) = (/x)n xm f [m,n] (0, ). (1.16)
n=0 m=0
N 1
N 1
N1
m
(/x)n xm f [m,n] (0, ) = m (x/)m (/x)n f [m,n] (0, ).
n=0 m=n m=0 n=0
(1.17)
Thus it is clear that
N1
CN y(x, ) = n [un (x) + vn (x/)], (1.18)
n=0
where
n1
un (x) = xn f [0,n] (x, ) xm f [m,n] (0, ) , (1.19)
m=0
1.3 Some Calculations 5
m
vm (X) = X m f [m,0] (0, X) X n f [m,n] (0, ) . (1.20)
n=0
Also, f (x, X) C ([0, 1] [0, ]) implies un (x) C ([0, 1]) and vm (X)
C ([0, ]). We can now state the theorem that is the basis for all our work
in this text.
Theorem 1. If for some o > 0, y(x, ) = f (x, x/) for all (x, ) [0, 1]
(0, o ], where f (x, X) C 2N ([0, 1] [0, ]), then
N1
y(x, ) = n [un (x) + vn (x/)] + O(N ) (1.21)
n=0
Before turning to the proof of Theorem 1, let us see how it applies to the
problem of Section 1.1. If we let y(x, ) = z(x, ), then y(x, ) = f (x, x/),
where f (x, X) = 1/[1 + X(1 + 2x2 )], and obviously f (x, X) C ([0, 1]
[0, ]). From
/(x + 2x3 )
y(x, ) = (1.22)
1 + /(x + 2x3 )
for x = 0, it is apparent that
N1
n
ON y(x, ) = (1)n+1 . (1.23)
n=1
(x + 2x3 )n
In particular,
O1 y(x, ) = 0, O2 y(x, ) = . (1.24)
x + 2x3
Similarly,
1/(1 + X)
y(X, ) = , (1.25)
1 + 22 X 3 /(1 + X)
so we have
1
I1 y(x, ) = I2 y(x, ) = . (1.26)
1 + x/
Next, from either (1.24) or (1.26) we get
Proof. Let > 0 be given. Then there exists > 0 such that |f (x, X)
f (0, X)| < whenever (x, X) [0, ] [0, ]. Therefore
for all (x, ) [0, ](0, ]. Similarly, there exists > 0 such that |f (x, X)
f (x, )| < whenever (x, X) [0, 1] [, ], and therefore
for all (x, ) [, 1] (0, d], where d = /. Together, (1.30) and (1.31) show
that |f (x, x/) (x, x/)| < 2 whenever (x, ) [0, 1] (0, d].
N
f (x, x/) = n [un (x) + vn (x/)] + o(N ) (1.32)
n=0
M
h(x, x/) = n [un (x) + vn (x/)] + o(M ) (1.33)
n=0
m
vm (X) = X m h[m,0] (0, X) X n h[m,n] (0, ) , (1.35)
n=0
M
f (x, x/) = (x, x/) + n [un (x) + vn (x/)] + o(M +1 ). (1.37)
n=0
Similarly,
Thus we see un (x) = un+1 (x) and vm (X) = vm+1 (X), and therefore (1.37)
is the same as (1.32) with N = M + 1.
Calculating more than a few terms in the uniform expansion (1.21) for most
any y(x, ) = f (x, x/) quickly becomes tedious work. One of our objectives
in this book is to encourage the use of computer algebra software to do cal-
culations like this for us. We will be doing this using the well-known product
called Maple. As our rst example, starting on a blank Maple worksheet, for
the function y(x, ) = /( + x + 2x3 ) of Section 1.3, by entering the sequence
of commands
y := ( + x + 2x3 )1 ; N := 4;
8 1 Uniform Expansion Theory
series(y, = 0, N );
ON y := convert(%, polynom);
series(subs(x = X, y), = 0, N );
IN y := convert(%, polynom);
series(subs(x = X, ON y), = 0, N );
IN ON y := convert(%, polynom);
series(subs(X = 1 x, IN y), = 0, N + 1);
ON IN y := convert(%, polynom);
for n from 0 to N 1 do
expand(xn coe (ON IN y, , n));
series(%, x = 0, n);
coe (ON y, , n) xn convert(%, polynom);
un := simplify(%);
end do;
The idea in this second step is to capture the sum in (1.19) from the coe-
cient of n in ON IN y(x, ) as it appears in (1.16). We used Maples expand
command to convert xn times this coecient into a polynomial and thus
avoid another removable singularity problem. By similar reasoning, looking
at (1.15) and the sum in (1.20), if we add
for m from 0 to N 1 do
expand(X m coe (IN ON y, , m));
series(%, X = , m + 1);
coe (IN y, , m) X m convert(%, polynom);
vm := simplify(%);
end do;
we immediately get
1 2(3X + 2)
v0 (X) = , v1 (X) = 0, v2 (X) = , v3 (X) = 0. (1.41)
1+X (1 + X)2
1.6 Two Corollaries 9
yields
N 1
f (x, X, ) = n f [0,0,n] (x, X, 0) + O(N ) (1.43)
n=0
N 1
y(x, ) = n f [0,0,n] (x, x/, 0) + O(N ) (1.44)
n=0
N1
y(x, ) = n [un (x) + vn (x/)] + O(N ) (1.45)
n=0
N 1
ON y(x, ) = n yn (x), (1.46)
n=0
N 1
IN y(x, ) = n Yn (x/), (1.47)
n=0
where, Yn (X) = Y [0,n] (X, 0), Y (X, ) = y(X, ). Also, for any M, N 1,
ON IM y(x, ) = IM ON y(x, ).
Proof. From (1.44) and Theorem 1, it is easy to see (1.45) holds with
k [m,n]
n1
[0,n]
uk (x) = xn gkn (x, ) gkn (0, )xm , (1.48)
n=0 m=0
k
m
[m,0] [m,n]
vk (X) = X gkm (0, X)
m
gkm (0, )X n , (1.49)
m=0 n=0
where gk (x, X) = f [0,0,k] (x, X, 0), and it is clear that uk (x) C ([0, 1]),
vk (X) C ([0, ]) and vk () = 0. For the (independent) determination
of these functions by outer and inner expansions, we defer to Exercise 1.3 at
the end of the chapter.
It often happens too that we have y(x, ) = f (x, x/, ) and f (x, X, ) =
o(X ) as X . That is, f [0,n,0] (x, , ) = 0 for all n 0, which means
ON y(x, ) = 0 for all N 1. Also, additional parameters may be present and
of course we are not restricted to the interval 0 x 1. We will need the
following, for example, in which x is the additional parameter, in the next
chapter.
Corollary 2. If y(x, t, ) = f (x, t, t/, ) , where f (x, t, T, ) C ([0, xo ]
[0, x] [0, ] [0, o ]) for some o , xo > 0, and if f [0,0,n,0] (x, t, , ) = 0
for all n 0, then for any N 0,
N
y(x, t, ) = n [0,0,n] (x, t/, 0) + O(N +1 ) (1.50)
n=0
1.7 Exercises
1.1. We noted in Section 1.2 that y(x, ) = f (x, x/), where f (x, X)
C ([0, 1] [0, ]), implies y(x, ) = ON y(x, ) + O(N ) uniformly on
[, 1] for any (0, 1). In light of Theorem 1 this also means [IN
ON IN ]y(x, ) = O(N ) on [, 1]. Show directly from(1.14) and (1.15) that
in fact [IN ON IN ]y(x, ) = O(N ) uniformly on [, c] for any c > > 0.
+ x2
y(x, ) = (1.51)
+ 2 + x + 2x3
and that
u0 (x) = y0 (x), u1 (x) = y1 (x) 1/x, u2 (x) = y2 (x) 1/x + 1/x2 , (1.53)
and
Complete the calculation of un (x) and vn (X) for 0 n 2, and show that
1
y(x, ) dx = a+ ln(1/)b+2 ln(1/)+c2 43 ln(1/)+O(3 ), (1.55)
0
where a = 14 ln 3, b = 1
6 + 1
2 ln 3 + 4
2 7
arctan 2, c = 18 1
2 ln 3 +
5 2
4
arctan 2.
N 1
N 1
y1 (x, ) = n un (x), y2 (x, ) = n vn (x/) (1.56)
n=0 n=0
where = 2 , f (s, S) = S m eSu(s) , u(s) = as + b(1 s). Since u(s) > 0 for
0 s 1, it follows, as in Corollary 2, that for any N 0,
N 1
f (s, s/) = n [0,n] (s/, 0) + O(N ) (1.59)
n=0
as , where
(m + 2n)! (a b)n
cn = (1)n . (1.62)
n! bm+2n
What if m is not an integer?
where C is the straight line from the origin to a point zo = ro eio and g(z),
h(z) are analytic on C. Assume Re[h(z)] > 0 on C, except h[k] (0) = 0 for
0 k m 1 and o is such that = Re[eimo h[m] (0)] > 0, where m 1.
Observe that these assumptions imply the real part of u(r) = r m h(rei0 )
exceeds zero for 0 r ro , and therefore
ro
F () = f (r, r/) dr, (1.64)
0
N
f (r, r/) = n [0,n] (r/, 0) + O(N+1 ), (1.65)
n=0
1.7 Exercises 13
where (R, ) = f (R, R). Also as in Exercise 1.4, [0,n] (R, 0) is bounded by
a polynomial times a decreasing exponential, in this case exp(Rm ), and
therefore
[0,n] (r/, 0) dr = o( ). (1.66)
ro
N1
F () = n [0,n] (R, 0) dR + O(N+1 ). (1.67)
n=0 0
m
In other words, if (Z, ) = g(Z)eh(Z)/ , then
N1 eio
F () = n
[0,n] (Z, 0) dZ + O(N +1 ), (1.68)
n=0 0
[m]
(0)Z m
and [0,n] (Z, 0) is a polynomial times eh .
N 1 ei/3
J () = n [0,n] (Z, 0) dZ + O(N +1 ), (1.71)
n=0 ei/3
1.7. Assume a(x, ) C ([0, 1] [0, o ]). Show there exists b(x, )
C ([0, 1] [0, o ]) and c() C ([0, o ]) such that, for > 0,
a(x, ) c()
= b(x, ) + . (1.73)
x+ x+
Of course, this is trivial if a(x, ) is a polynomial.
where
N1
n (x) = xN f [0,n] (x, ) xm f [m,n] (0, ) , (1.75)
m=0
2.1 Problem A
The plan for this book is to apply the theory developed in Chapter 1 to a
sequence of more or less increasingly complex dierential equation problems.
We begin with
y + a(x, )y = b(x, ), (2.1)
where a(x, ), b(x, ) C ([0, 1][0, o ]) for some o > 0, and a(x, ) > 0. We
also assume y(0, ) = () C ([0, o ]). The problem is to asymptotically
approximate y(x, ) uniformly on 0 x 1 as 0+ . If we put z(x, ) =
y(x, ) (), then
x
1
z(x, ) = b(x t, )e[k(x,)k(xt,)]/ dt, (2.2)
0
where X
n (x, X) = [0,0,n] (x, T, 0) dT. (2.7)
0
Furthermore,
[0,0,n] (x, T, 0) = pn (x, T )ea(x,0)T , (2.8)
where pn (x, T ) is a polynomial in T , and therefore n (x, X) C ([0, 1]
[0, ]). Hence, in accordance with Corollary 1, applied to the sum in (2.6),
we see y(x, ) = z(x, ) + () has a uniformly valid asymptotic expansion of
the form
N 1
y(x, ) = n [un (x) + vn (x/)] + O(N ), (2.9)
n=0
where un (x) C ([0, 1]), vn (X) C ([0, ]), vn () = 0, and these terms
Also C1 y(x, ) = C1 z(x, ) + (0). Hence, the rst terms of (2.9) are
u0 (x) = b(x, 0)/a(x, 0), v0 (X) = [(0) b(0, 0)/a(0, 0)]ea(0,0)X . (2.13)
(2.1) implies
n
yn1 (x) + ank (x)yk (x) = bn (x), (2.15)
k=0
where an (x) = a[0,n] (x, 0), bn (x) = b[0,n] (x, 0). Similarly, for the N -term inner
expansion,
N1
IN y(x, ) = n Yn (x/), (2.16)
n=0
2.1 Problem A 17
Therefore,
n
Yn (X) + Ank (X)Yk (X) = Bn (X), (2.18)
k=0
where An (X) = A[0,n] (X, 0), Bn (X) = B [0,n] (X, 0). Also, y(0, ) = () im-
plies Yn (0) = [n] (0). For the actual calculations we oer our rst Maple
program.
This program solves for the N -term outer expansion of y(x, ) using (2.15)
and then computes the N -term inner expansion using (2.18). At the end, the
program computes IN ON y(x, ). Normally there would be the matter of sep-
arating IN ON y(x, ) into two parts to form the functions un (x) and vn (X).
18 2 First Order Dierential Equations
yields
1 x
u0 (x) = , u1 (x) = , (2.19)
(2 + x)2 (2 + x)4
and
1 1
v0 (X) = e2X , v1 (X) = (8 + 2X + X 2 )e2X . (2.20)
4 8
2.2 Problem B
Again we assume a(x, ), b(x, ) C ([0, 1][0, o ]) for some o > 0. We also
assume a(x, ) > 0 on [0, 1][0, o ], y(0, ) = (), where () C ([0, o ]),
and, without loss of generality, we assume a(0, 0) = 1. The essential dierence
here from Problem A is the factor of x multiplying a(x, ).
In place of (2.3), we now have
x
k(x, ) = ta(t, ) dt (2.22)
0
and therefore, as t 0+ ,
1
k(x t, 0) = k(x, 0) txa(x, 0) + t2 [a(x, 0) + xa[1,0] (x, 0)] + O(t3 ). (2.23)
2
If we put
then u(0, 0, 0) = 1/2 and therefore u(x, t, ) > 0 on [0, xo ] [0, x] [0, o ]
for some xo > 0, and possibly a smaller o > 0. Thus if we let z(x, ) =
y(x, ) () and b(x, ) = b(x, ) x()a(x, ), then
2.2 Problem B 19
x
2
z(x, ) = 1 f (x, t, t/, )et(xt)a(x,0)/ dt, (2.25)
0
where,
2
f (x, t, T, ) = b(x t, )eT u(x,t,)
, (2.26)
and we have f (x, t, T, ) C ([0, xo ] [0, x] [0, ] [0, o ]). Also
2
0 < et(xt)a(x,0)/ 1 (2.27)
for all (x, t, ) [0, 1] [0, x] (0, o ]. Hence, applying Corollary 2 with
(x, T, ) = f (x, T, T, ), from (2.25) we get
N
z(x, ) = n n (x, x/) + O(N ) (2.28)
n=0
Also,
2
[0,0,n] (x, T, 0) = pn (x, T )eT u(x,0,0)
, (2.30)
where pn (x, T ) is a polynomial in T .
From Exercise 1.4, it is clear that
X
2
F (x, X) = T m eT u(x,0,0) T (XT )a(x,0)
e dT, (2.31)
0
N 1
y(x, ) = n [un (x) + vn (x/)] + O(N ). (2.32)
n=0
But also, from Problem A, we know that y(x, ) = ON y(x, ) + O(N ) uni-
formly as 0+ for xo x 1, since xa(x, ) > 0 for xo x 1. Further-
more, by Exercise 1.1, we know [IN ON IN ]y(x, ) = O(N ) for xo x 1.
Hence, in conclusion, y(x, ) has a uniformly valid expansion of the form
(2.32) as 0+ on the full interval 0 x 1, where un (x) C ([0, 1]),
vn (X) C ([0, ]), vn () = 0, and these functions can be determined by
computing the corresponding outer and inner expansions for y(x, ) directly
from the dierential equation (2.21).
20 2 First Order Dierential Equations
The solution to this equation has the same form as its right hand side. That
is,
1 2
Y (X) = p(X) + q(X)P (X) + r(X)e 2 X , (2.35)
where p(X), q(X), r(X) are polynomials. Indeed, if we substitute (2.35) into
(2.33), we nd
Therefore
X X
q(X) = q0 + (T ) dT, r(X) = r0 + (T ) dT, (2.37)
0 0
d1
d
p(X) = pn X n , s(X) = sn X n (2.38)
n=0 n=0
In this program, the letters I, P , and E are used to denote 1, P (X), and
exp( 12 X 2 ), respectively. Also, we have used the fact that IN ON y(x, ) is at
most O(X N1 ) as X to split it into the two parts necessary to form
un (x) and vn (X) for n = 0 to N 1.
As an example, if we set
with
c c2 x(3 + cx2 )
v2 (X) = [X + X 3 + (3 X 4 )P (X)], u3 (x) = . (2.42)
4 (1 + cx2 )3
As another, if
then
k gh
u0 (x) = 0, u1 (x) = , v0 (X) = hP (X), (2.44)
1 + gx
and
1 1 1 1 2
v1 (X) = gh(1 + X 2 ) ghX 3 P (X) + (2gh 3k)e 2 X . (2.45)
3 3 3
For
a graph of
C2 y(x, ) = P (x/) + [u1 (x) + v1 (x/)] (2.47)
when = 0.2 is shown in Figure 2.1, along with a portion of O2 y(x, ) and
Maples numerical solution of the dierential equation. In this last example,
0.8
0.6
0.4
0.2
Fig. 2.1 Numerical solution of (2.21) and asymptotic approximations of the solution when
a(x, ) = cos(x) + x, b(x, ) = cos(x), y(0, ) = 1 and = 0.2.
2.3 Exercises 23
1 1 1 5 1 2
u1 (x) = , v1 (X) = (1 + X 2 ) X 3 P (X) + e 2 X . (2.48)
cos(x) + x 3 3 3
2.3 Exercises
y + x(1 + x + )y = 2 + x2 , 1 x 2, (2.49)
2 + x2
y(x, ) = + e2(x1)/ + [u1 (x) + v1 ((x 1)/)] + O(2 ) (2.50)
x(1 + x)
for 1 x 2, where
2 4x + 3x2 + 2x3 + x4 + x5
u1 (x) = , (2.51)
x3 (1 + x)3
1
v1 (X) = (1 + 2X + 3X 2 )e2X . (2.52)
2
2.2. Note that An (X) and Bn (X) in (2.18) are polynomials and that therefore
Yn (X) = pn (X) + qn (X)exp[a(0, 0)X], where pn (X), qn (X) are polynomi-
als. Therefore vn (X) = qn (X)exp[a(0, 0)X], since vn () = 0. Furthermore,
n
vn (X) + Ank (X)vk (X) = 0, (2.53)
k=0
and vn (0) = [n] (0) un (0). Write a new, shorter Maple program for com-
puting the terms of (2.9).
2.3. Show that we could add 2 times c(x, ) C ([0, 1] [0, o ]), to xa(x, )
in (2.21) without upsetting the basic analysis, and modify the program P robB
appropriately to include it.
2 + x2 1 2
y(x, ) = e 2 (x/) + [u1 (x) + v1 (x/)] + O(2 ), (2.55)
1+x
24 2 First Order Dierential Equations
where
2 + x2 1 1 2
u1 (x) = , v1 (X) = 2P (X) + (12 + 3X 2 + 2X 3 )e 2 X . (2.56)
(1 + x)2 6
2.5. Suppose s(x) C ([0, 1]), s (x) > 0 and g(x, ) C ([0, 1] [0, o ])
for some o > 0. Let
x
F (x, ) = 1 e[s(x)s(t)]/ g(x, t) dt, (2.57)
0
1
G(x, ) = 1 e[s(x)s(t)]/ g(x, ) dt. (2.58)
x
From the analysis at the beginning of Section 2.1, we know F (x, ) has a
uniformly valid expansion of the form
N1
F (x, ) = n [un (x) + vn (x/)] + O(N ) (2.59)
n=0
N1
G(x, ) = n [un (x) + wn ((1 x)/)] (2.60)
n=0
2.6. Suppose g(x, t), h(x, t) C ([0, 1] [0, x]), h(x, t) > 0 for 0 < t
x 1, h(x, 0) = 0 for 0 x 1, and h[0,1] (x, 0) > 0 for 0 < x 1 but
h[0,1] (0, 0) = 0. Let
x
F (x, ) = eh(x,t) g(x, t) dt. (2.62)
0
where we have begun using hij = h[i,j] (0, 0), gij = g [i,j] (0, 0), so it must be
that h11 x + h02 t 0 for 0 < t x 1.
Assume h11 > 0 and h02 > 0. If we let
u(x, t) = t2 [h(x, t) th[0,1] (x, 0)], a(x) = x1 h[0,1] (x, 0), (2.65)
then u(0, 0) > 0 and a(0) > 0, and hence there exists xo > 0 such that
u(x, t) > 0 on [0, xo ] [0, x] and a(x) > 0 on [0, xo ]. Therefore
x
F (x, ) = f (x, t, t/)exta(x) dt, (2.66)
0
where = 1/2 ,
2
f (x, t, T ) = eT u(x,t)
g(x, t) (2.67)
and f (x, t, T ) C ([0, xo ] [0, x] [0, ]). In addition, 0 < exta(x) 1
and thus, applying Corollary 2, with (x, T, ) = f (x, T, T ), we get
N 1
F (x, ) = n n (x, x/) + O(N ) (2.68)
n=1
g(x, 0) g00
u2 (x) = , (2.72)
h[0,1] (x, 0) h11 x
and X
2
v2 (X) = p2 (X, T )eh02 T h11 XT
dT, (2.73)
0
26 2 First Order Dierential Equations
where
p2 (X, T ) = g10 + (g01 g00 h21 X 2 )T g00 h12 XT 2 g00 h03 T 3 . (2.74)
Note, furthermore, that for x > 0 the right side of (2.70) is asymptotically
equal to 2 [g(x, 0)/h[0,1] (x, 0)] + O(3 ), the beginning of its outer expansion,
and therefore, in view of (2.63), equation (2.70) actually holds uniformly for
0 x 1.
The result (2.32) for the solution to Problem B is an example of an ex-
pansion for a case of F (x, ) with h11 > 0 and h02 < 0.
Chapter 3
Second Order Dierential Equations
3.1 Problem C
N 1
y(x, ) = n [un (x) + vn (x/)] + N RN (x, ), (3.2)
n=0
where x
f (x, ) = () + a(0, )() + c(t, ) dt. (3.4)
0
Therefore, introducing
where x
1
(x, ) = () + e[k(x,)k(t,)]/ f (t, ) dt (3.7)
0
and x
K(x, t, ) = 1 g(t, ) e[k(x,)k(s,)]/ ds. (3.8)
t
Again as in Problem A, we know by Corollary 2 that
x
1
e[k(x,)k(t,)]/ dt = O(1) (3.9)
0
N 1
N1
n
a(x, ) n un (x) = n ak (x)unk (x) + O(N ) (3.11)
n=0 n=0 k=0
N
N
n
a(x, ) n vn (x/) = n
Ak (x/)vnk (x/) + O(N +1 ) (3.12)
n=0 n=0 k=0
3.1 Problem C 29
for 0 n N 1 and
n
n1
vn (X) +
Ak (X)vnk (X) + Bk (X)vn1k (X) = 0, (3.14)
k=0 k=0
N 1
ON y(x, ) = n un (x) (3.16)
n=0
P robC := proc(a, b, c, , , N )
N u := sum(n un , n = 0..N 1); N du := sum(n dun , n = 0..N 1);
N ddu := sum(n ddun , n = 0..N 1);
N ude := series( N ddu + a N du + b N u c, = 0, N );
N := series(, , 0, N ); N := series(, = 0, N );
for k from 0 to N 1 do
temp := coe (N ude, , k);
de := subs(uk = z(x), duk = di (z(x), x), dduk = di (z(x), x, x), temp);
bc := z(1) = coe (N , , k);
dsolve({de, bc}); uk := rhs(%); duk := di (uk, x); dduk := di (duk, x);
N ude := subs(uk = uk, duk = duk, dduk = dduk, N ude);
print(uk = simplify(uk)); uk := uk;
end do;
A := subs(x = X, a); B := subs(x = X, b);
N v := sum(n vn , n = 0..N 1); N dv := sum(n dvn , n = 0..N 1);
N ddv := sum(n ddvn , n = 0..N 1);
N vde := series(N ddv + A N dv + B N v, = 0, N );
for k from 0 to N 1 do
temp := coe (N vde, , k);
de := subs(vk = z(X), dvk = di (z(X), X), ddvk = di (z(X), X, X), temp);
bc := z(0) = coe (N , , k) subs(x = 0, uk ), z() = 0;
dsolve({de, bc}); vk := rhs(%); dvk := di (vk, X); ddvk := di (dvk, X);
N vde := subs(vk = vk, dvk = dvk, ddvk = ddvk, N vde);
print(vk = simplify(vk));
end do;
end proc:
3.2 Problem D
If we take s (x) to be the positive square root of b(x, 0), then (3.19) is a
case of Problem C and, as noted at the end of Section 3.1, then (3.19) has a
solution with a uniformly valid expansion as 0+ of the form
N1
z(x, ) = n zn (x) + O(N ), (3.21)
n=0
where zn (x) C ([0, 1]). If we also require z(0, 0) = 1, then z(0, ) > 0 for
> 0 suciently small.
Actually, a solution to (3.19) of the form (3.21) also exists if s (x) =
[b(x, 0)]1/2 . Indeed, in this case, simply replacing x by 1 x changes (3.19)
into another case of Problem C. Hence, to summarize, if we let
x
s(x) = [b(t, 0)]1/2 dt, (3.22)
0
N1
z (k) (x, ) = n zn(k) (x) + O(N ) (3.25)
n=0
uniformly as 0+ for 0 x 1.
The Wronskian of these two homogeneous equation solutions is W (x, ) =
W (0, )(x, ), where
32 3 Second Order Dierential Equations
x
a(t,) dt
(x, ) = e 0 (3.26)
and W (0, ) = 1 es(1)/ (), where () has an expansion in powers of
as 0+ . In particular, (0) = 2[b(0, 0)]1/2 . By the method of variation of
parameters, if we let
where x
F (1) (x, ) = 1 e[s(x)s(t)]/ g (1) (x, t, ) dt, (3.30)
0
1
F (2) (x, ) = 1 e[s(t)s(x)]/ g (2) (x, t, ) dt. (3.31)
x
From Exercise 2.5, we know F (1) (x, ), F (2) (x, ) have uniformly valid
asymptotic expansions for 0 x 1, expressible as
N1
F (1) (x, ) = n [u(1) (1) N
n (x) + vn (x/)] + O( ), (3.32)
n=0
N 1
F (2) (x, ) = n [u(2) (2)
n (x) + vn ((1 x)/)] + O( ),
N
(3.33)
n=0
(k) (k) (k)
where un (x) C ([0, 1]), vn (X) C ([0, ]) and vn (X) = o(X ) as
(k)
X . In addition, there exists vh,n (X) C ([0, ]) such that
(1)
N1
(1)
yh (x, ) = n vh,n (x/) + O(N ), (3.34)
n=0
N 1
(2) (2)
yh (x, ) = n vh,n ((1 x)/) + O(N ) (3.35)
n=0
(k)
uniformly for 0 x 1 as 0+ , and vh,n (X) = o(X ) as X .
Putting all this together we see the solution to (3.17), namely yp (x, ) plus
(1) (2)
an -dependent linear combination of yh (x, ) and yh (x, ), subject to the
boundary conditions y(0, ) = (), y(1, ) = (), has a uniformly valid
expansion for 0 x 1 of the form
3.2 Problem D 33
N 1
y(x, ) = n [un (x) + vn (x/) + wn ((1 x)/)] + O(N ) (3.36)
n=0
n2
n
un2 (x) + ak (x)un2k (x) bk (x)unk (x) = cn (x). (3.37)
k=0 k=0
Also, in terms of An (X) = A[0,n] (X, 0), Bn (X) = B [0,n] (X, 0), where
A(X, ) = a(X, ), B(X, ) = b(X, ), we have
n1
n
vn (X) +
Ak (X)vn1k (X) Bk (X)vnk (X) = 0, (3.38)
k=0 k=0
where, again, An (X) = A[0,n] (X, 0), Bn (X) = B [0,n] (X, 0). So, rst we
nd the outer expansion terms un (x) for 0 n N 1, beginning with
u0 (x) = c(x, 0)/b(x, 0), from (3.37). Then the (constant coecient) dier-
ential equations (3.38) and (3.39) can be solved successively, with the bound-
ary conditions vn (0) = [n] (0)un (0), wn (1) = [n] (0)un (1), together with
vn () = wn () = 0. A Maple program patterned after P robC is readily de-
vised to automate these calculations. To get more than a couple of terms,
however, Maple needs assistance like the assistance provided in P robB to
calculate vn (X) and wn (X). This issue comes up in the next section, too.
If a(x, ) = 1 + x, b(x, ) = 2 + x + , c(x, ) = 4x2 and () = 2,
() = 1/2, then
4x2 5
u0 (x) = , v0 (X) = 2e 2X
, w0 (X) = e 3X . (3.40)
2+x 6
2.0
1.5
1.0
0.5
0.0
Fig. 3.1 Numerical solution to (3.17) and asymptotic approximations of the solution when
a(x, ) = 1 + x, b(x, ) = 2 + x + , c(x, ) = 4x2 , y(0, ) = 2, y(1, ) = 1/2 and = 0.15.
2.0
1.5
1.0
0.5
0.0
0.0 0.2 0.4 0.6 0.8 1.0
The problem for this section is a variation of Problem D in the same way Prob-
lem B is a variation of A. Assuming the functions a(x, ), c(x, ), d(x, )
C ([0, 1] [0, o ]) for some o > 0, we also assume 0 < b(x) C ([0, 1])
and, without loss of generality, set b(0) = 1. The dierential equation is
2 y xy = d(x, ). (3.42)
For (3.42), the corresponding homogeneous equation has the two linearly
independent Airy function solutions
(1) (2)
yh (x, ) = Ai(x/2/3 ), yh (x, ) = e(2/3)/ Bi(x/2/3 ) (3.43)
and, knowing the Wronskian of Ai(X) and Bi(X) is 1/, it is readily veried
that
yp (x, ) = F (1) (x, ) + F (2) (x, ) (3.44)
is a particular solution to (3.42), where
x
F (1) (x, ) = Ai(x/2/3 ) d(t, )Bi(t/2/3 ) dt, (3.45)
0
1
F (2) (x, ) = Bi(x/2/3 ) d(t, )Ai(t/2/3 ) dt. (3.46)
x
If we let
3/2 3/2
g(X) = e(2/3)X Ai(X), h(X) = e(2/3)X Bi(X), (3.47)
and dene g(X) = g(X 4 ), h(X) = h(X 4 ), then g(X), h(X) C ([0, ]). If
we also let x = x1/4 , = 1/6 , and substitute t = t4 , then
x
F (1) (x, ) = g(x/) t, )h(t/)e(2/3)(x6 t6 )/6 dt,
d( (3.48)
0
1
F (2) (x, ) = h(x/) t, )g(t/)e(2/3)(x6 t6 )/6 dt,
d( (3.49)
x
1/6
t, ) = 4 t3 d(t4 , 6 ) C ([0, 1] [0, o ]). The asymptotic expan-
where d(
sion form of these two integrals is readily established.
By Corollary 1, there exists n (t) C ([0, 1]), n (T ) C ([0, ]) such
that for any N 0,
N1
t, )g(t/) =
d( n [n (t ) + n (t/)] + O(N ) (3.50)
n=0
uniformly for 0 t 1, and the same can be said for d( t, )h(t/). In turn,
the results of Exercise 3.5, applied with u(t) = n (t) and v(T ) = n (T ), show
(k) (k) (2)
there exists n (x) C ([0, 1]), n (X) C ([0, ]), and also n (X) =
36 3 Second Order Dierential Equations
N 1
F (1) (x, ) = g(x/) n [(1) (1) N
n (x) + n (x/)] + O( ), (3.51)
n=0
N1
F (2) (x, ) = h(x/) n [(2) (2) (2)
n (x) + n (x/) + n ((1 x)/)] + O( ).
N
n=0
(3.52)
From here, with a few additional applications of Theorem 1, it is clear that the
solution to (3.42) satisfying the boundary conditions y(0, ) = (), y(1, ) =
() has the asymptotic form
N 1
y(x, ) = n [un (x) + vn (x/) + wn ((1 x)/)] + O(N ), (3.53)
n=0
where un (x) C ([0, 1]), both vn (X), wn (X) C ([0, ]), and wn (X) =
(2)
o(X ) as X . For example, to deal with the contribution of yh (x, )
(2)
to (3.53), let t = 1 x and observe that yh (x, ) = f (t, t/, ), where
the sum in (3.55) are uniformly o( ) for 1/2 t 1. Hence (3.55) holds
uniformly for the full interval, 0 x 1.
As in Problem D, the homogeneous version of (3.41), the full dierential
equation for Problem E, has a pair of linearly independent solutions express-
ible as
(1)
yh (x, ) = es(x)/ z (1) (x, ), (3.56)
(2)
yh (x, ) = e[s(1)s(x)]/ z (2) (x, ), (3.57)
where x
s(x) = [tb(t)]1/2 dt. (3.58)
0
N1
z (k) (x, ) = n/6 [u(k) (k)
n (x) + vn (x/
2/3
)] + O(N/6 ), (3.59)
n=0
(k) (k)
where un (x4 ) C ([0, 1]) and vn (X 4 ) C ([0, ]). In Exercise 3.6,
we show that in fact this is true whether a(x, ) = 0 or not. Therefore, as
above for (3.42) and as in Problem D, we can use the method of variation
of parameters to obtain a particular solution of (3.41). It is also a straight-
forward matter to determine the asymptotic form of this solution, and the
homogeneous equation solutions (3.56) and (3.57).
Note that s(x) = x3/2 (x), where 0 < (x) C ([0, 1]), and again intro-
duce x = x1/4 to get s(x) = x6 (x), where (x) = (x4 ). Then, for example,
N1
f (x, x/) = n [0,n] (x/, 0) + O(N ), (3.62)
n=0
where u(t ) C ([0, 1]), let t = t[(t )]1/6 , let t = (t ) denote the inverse of
this transformation, and let x = x[(x)]1/6 to obtain
x
(x/)6 6
F (x, ) = e g(t)e(t/) dt, (3.64)
0
N 1
F (x, ) = n [n (x) + n (x/)] + O(N ), (3.65)
n=0
for any N 0, where n (x) C ([0, 1]) and n (X) C ([0, ]). Also,
we can apply Theorem 1 to f (x, X) = n (X[(x)]1/6 ). Thus, ultimately,
we see the solution to the boundary value problem for (3.41) has the same
asymptotic form as the solution for the special case (3.42). That is, there
exists un (x), vn (X) and wn (X) such that, for any N 0,
38 3 Second Order Dierential Equations
N 1
y(x, ) = n/6 [un (x) + vn (x/2/3 ) + wn ((1 x)/)] + O(N/6 ), (3.66)
n=0
N 1
ON y(x, ) = n y [0,n] (x, 0), (3.67)
n=0
and
N1
IN y(x, ) = n Y [0,n] (X, 0), (3.68)
n=0
N 1
ON y(x, ) = n zn (x) (3.69)
n=0
where zn (x) = z [0,n] (x, 0). If we also let Z(X, ) = y(4 X, 6 ), then Z(X 4 , ) =
Y (X, ), so
N1
IN y(x, ) = n Zn (X), (3.70)
n=0
[0,n]
where Zn (X) = Z (X, 0), and X = X 4 .
Upon substituting = 6 in (3.41), it is clear that zn (x) = 0 unless n 1
is a multiple of 6. In particular,
u6 (x) = z6 (x) + d00 /x, u12 (x) = z12 (x) + d01 /x (3.73)
v6 (X) = Z6 (X) + d10 d00 b1 , v12 (X) = Z12 (X) + d11 d01 b1 , (3.74)
v10 (X) = Z10 (X) + [d00 (b21 b2 ) d10 b1 + d20 ]X, (3.75)
and otherwise vn (X) = Zn (X) for 0 n 13.
From the dierential equation for Z(X, ) it is clear that Zn (X) = 0 if n
is odd. Also,
Z0 XZ0 = 0, Z0 (0) = 0 , (3.76)
Z2 XZ2 = d00 , Z2 (0) = 0, (3.77)
Z4 XZ4 = b1 X 2 Z0 a00 Z0 , Z4 (0) = 0, (3.78)
Z6 XZ6 = b1 X Z2 2
a00 Z2 d10 X, Z6 (0) = 1 , (3.79)
where n = [n] (0). In addition, vn () = 0 implies, in particular, Z0 () =
Z2 () = 0. Therefore
and, in terms of
X
P (X) = Ai(X) Bi(T ) dT Bi(X) Ai(T ) dT, (3.81)
0 X
where R(X) = Ai(X), S(X) = Ai (X), Q(X) = P (X), and (X), (X),
(X), (X) and (X) are polynomials. But the desired solution to an equa-
tion of this form has the same form as its right side. That is,
we get
2p2 + q0 = 0 , 6p3 + 3q1 = 1 , 5q2 = 2 , (3.88)
2q2 + 2p1 = 0 , 4p2 = 0 , 6p3 = 2 . (3.89)
Hence, in order,
1 1 1
q2 = 2 , p3 = 2 , q1 = (1 6p3 ), (3.90)
5 6 3
1 1
0 , q0 = 0 2p2 , p1 = (0 2q2 ).
p2 = (3.91)
4 2
The functions r(X), s(X), and t(X) are similarly determined once (X),
(X) and (X) are given and, in the end, p0 is determined by
Below is a Maple program that performs these calculations for us. The
program also computes wn (X), using the fact that wn (X) is a polynomial
times exp[b(1)X], and
N 1
y(1, ) = ON y(1, ) + n wn (0) + O(N ), (3.93)
n=0
P robE := proc(a, b, c, d, , , N )
ON y := sum(n yn , n = 0..N 1);
ON dy := sum(n dyn , n = 0..N 1);
ON ddy := sum(n ddyn , n = 0..N 1);
eq := subs( = 6 , 2 ON ddy + 2 a ON dy (x b + 2 c) ON y d);
ON eq := series(eq, = 0, N );
for k from 0 to N 1 do
3.4 Problem E, Part 2 41
3.5 Exercises
3.1. Complete the analysis necessary to verify that N (x, ) in (3.15) is uni-
formly O(1) for 0 x 1 as 0+ . In particular, using (3.11) and (3.13),
show that
N1
n [un (x) + a(x, )un (x) + b(x, )un (x)] c(x, ) = O(N ) (3.94)
n=0
N
n [vn (x/) + a(x, )vn (x/) + b(x, )vn1 (x/)] = O(N ) (3.95)
n=0
uniformly for 0 x 1.
3.5 Exercises 43
N 1
f (x, x/, ) = n [un (x) + vn (x/)] + O(N ) (3.96)
n=0
N 1
g(x, x/, ) = n [n (x) + n (x/)] + N N (x, ), (3.97)
n=0
where
A3 (X) = [h[0,0,1] (0, u0 (0) + v0 (X), 0) h[0,0,1] (0, u0 (0), 0)], (3.104)
A4 (X) = u1 (0)[h[0,1,0] (0, u0 (0) + v0 (X), 0) h[0,1,0] (0, u0 (0), 0)]. (3.105)
3.3. Another way to calculate the terms of (3.2) when the initial conditions
y(0, ) = (), y [1,0] (0, ) = 1 () are given is to begin by rst calculating
the terms of the inner expansion,
N1
IN y(x, ) = n Yn (X). (3.106)
n=0
These satisfy Yn (0) = n and Yn (0) = n and just as noted for Problem A
in Exercise 2.2, they are of the form Yn (X) = pn (X) + qn (X)exp[a(0, 0)X],
so vn (X) = Yn (X) pn (X) for 0 n N 1. Once this calculation is
done, the terms of ON y(x, ) which satisfy (3.13), can be computed using
un (0) = n vn (0). Write a Maple program to do these calculations.
3.4. There also is another way to calculate the terms of (3.2) when the bound-
ary conditions y(0, ) = (), y(1, ) = () are given. The idea is to rst
solve (3.1) for y(x, , ), subject to the initial conditions y(0, , ) = () and
y [1,0,0] (0, , ) = /. This yields
1
N
y(x, , ) = n un (x, ) + vn (x/, ) + O(N ), (3.107)
n=0
say, and then, to satisfy y(1, , ) = (), we just have to solve the (linear)
equation
N1
() = n un (1, ) + O(N ) (3.108)
n=0
for
N 1
= n n + O(N ), (3.109)
n=0
then substitute into (3.107) and reduce the result to the nal form (3.2),
altogether a simple task for Maple. Try this out with your own choice of
parameters, using your program from the previous exercise. You can check
your results using the program at the end of Section 3.1.
k2
p(t) = u[m] (0)tm , q(t) = tk+1 [u(t) p(t)]. (3.111)
m=0
Show that
k2
1 k1 k
F (x, ) = m u[m] (0)Pm (x/) + [q(x) q(0)e(x/) ]
m=0
k
x
1 k k
+ k1 e(x/) q (t)e(t/) dt, (3.112)
k 0
where X
X k k
Pm (X) = e T m eT dT, (3.113)
0
then there exists un (x) C ([0, 1]), and vn (X), wn (X) C ([0, ]) such
that
N 1
F (x, ) = n [un (x) + vn (x/) + wn ((1 x)/)] + O(N ) (3.116)
n=0
N 1
k
+(1/)k
e(x/) = n n (t/)e(t/) + O(N ), (3.117)
n=0
46 3 Second Order Dierential Equations
Show that G(x, ) = V (x/), where V (X) C ([0, ]), and show there
exists vn (X), wn (X) C ([0, ]) such that
N1
H(x, ) = n [vn (x/) + wn ((1 x)/)], (3.120)
n=0
where, as above, = k .
where c(x, ) C ([0, 1] [0, o ]). Show further that if u(x4 ) C ([0, 1])
and v(X 4 ) C ([0, ]), then there exists un (x), vn (X)) such that un (x4 )
C ([0, 1]), vn (X 4 ) C ([0, ]) and, for any N 0,
N1
[u(x) + v(x/2/3 )](x, ) = n/6 [un (x) + vn (x/2/3 )] + O(N/6 ) (3.123)
n=0
2 y (x + 2x2 + 2 )y = 5 3x (3.124)
where = 2/3 ,
3.5 Exercises 47
10
Fig. 3.3 Numerical solution to (3.124) and asymptotic approximations of the solution
when = 0.1.
5P (0) 13 2
v1 (X) = Ai(X) 5P (X), u(x) = , w(X) = e 3X
Ai(0) 1 + 2x 3
(3.126)
and
11 2P (0)
v2 (X) = Ai(X) [XAi(X) X 2 Ai (X)]
Ai(0) Ai(0)
+2XP (X) 2X 2 P (X) + 4. (3.127)
Figure 3.3 shows this approximation and Maples numerical solution for
y(x, ) when = 0.1. It also shows c(x, ) plus v3 (x/), the next term after
c(x, ) in the uniformly valid asymptotic expansion of y(x, ), which is signif-
icantly closer to the numerical solution.
N 1
(/2)1/3 J ((1 x)) = n vn (x/) + O(N ) (3.129)
n=0
48 3 Second Order Dierential Equations
3 2 1 17
q0 (X) = 0, q1 (X) = X , q2 (X) = + X 3. (3.132)
10 35 70
In addition,
2 173 3 957
p3 (X) = + X + X 6, (3.133)
225 3150 14000
37 611 4 9
q3 (X) = X+ X + X 7, (3.135)
1575 3150 2000
3.9. Write a Maple program for Problem D and use it to verify (3.40). Also
verify that
4x2 1
u1 (x) = , v1 (X) = (5X 2 2X 2X 2 )e 2X , (3.137)
(2 + x)2 4
1
w1 (X) = (32 55X + 20 3X + 5 3X 2 )e 3X (3.138)
72
for this example.
Chapter 4
Logarithm Problems
4.1 Preliminaries
N 1
f (t, t/, ) = n [un (t) + vn (t/)] + O(N ) (4.2)
n=0
and X
Vn (X) = X 1 [vn (T ) cn (1 + T )1 ] dT, (4.4)
0
[1]
where cn = vn (), then Un (x) C ([0, 1]) and Vn (X) C ([0, ]).
Therefore
N1
y(x, ) = n [Un (x) + Vn (x/) + cn L1 (x/)] + O(N ), (4.5)
n=0
then X
1
F (X) = cL2 (X) + X 1 (T ) ln(1 + T ) dT, (4.7)
2 0
By induction, if
X
F (X) = X 1 v(T ) lnn (1 + T ) dT, (4.10)
0
for > 0, where u(x, ) C ([0, 1] [0, o ]) for some o > 0. If we integrate
by parts and dene t
a(t, ) = t3 s2 u(s, ) ds, (4.13)
0
t3 3
= t2 t + 2 , (4.14)
t+ t+
4.1 Preliminaries 51
we have
x x
a(t, )
y(x, ) = x1 (t2 + t 2 )a(t, ) dt + 3 x1 dt. (4.15)
0 0 t+
Furthermore, from Exercise 1.7, there exist g(t, ) C ([0, 1] [0, o ]) and
c() C ([0, o ]) such that
a(t, ) c()
= g(t, ) + . (4.16)
t+ t+
Therefore
for > 0, where u(x, ) C ([0, 1] [0, o ]), then there exists ak (x, )
C ([0, 1] [0, o ]) for 0 k n and ck () C ([0, o ]) for 1 k m such
that
n
m
y(x, ) = k xnk ak (x, ) + n ck ()Lk (x/) (4.21)
k=0 k=1
for all (x, ) [0, 1] (0, o ]. We will use this result to help establish the
asymptotic form of the solution to Problem F in the next section.
Another source of logarithms is exponentiation. For example, if y(x, ) =
1 + x/ for 0 x 1 and > 0, then
N 1
1
y (x, ) = e ln(1+x/) = 1 + x n Ln+1 (x/) + O(N ) (4.22)
n=0
(n + 1)!
N 1
n
y d() (x, ) = 1 + x n dnk Lk+1 (x/) + O(N ) (4.23)
n=0 k=0
52 4 Logarithm Problems
where f (x, X, ) C ([0, 1] [0, ] [0, o ]), then there exists unm (x)
C ([0, 1]) and vnm (X) C ([0, ]) such that
N1
n+1
y(x, ) = n [unm (x) + vnm (x/)]Lm (x/) + O(N ) (4.26)
n=0 m=0
4.2 Problem F
satisfying y(0, ) = 0.
We know there exists d() C ([0, o ]) and g(x, ) C ([0, 1] [0, o ])
such that
a(x, ) I + d()
= g(x, ) + (4.36)
x+ x+
for > 0. Therefore
x d()
h(x, ) I1 1 + t/
y(x, ) = (1 + t/) (t, ) dt, (4.37)
(1 + x/)I 0 1 + x/
where
x
h(x, ) = exp g(t, ) dt , (x, ) = b(x, )/h(x, ). (4.38)
0
for 0 t x 1 and 0+ .
Let
x
h(x, ) 1 + t/
wm (x, ) = (1 + t/)I1 (t, ) lnm dt. (4.40)
(1 + x/)I 0 1 + x/
Since x
x1 (t/)k (t, ) dt = (x/)k k (x, ), (4.41)
0
where k (x, ) C ([0, 1] [0, o ]), we have w0 (x, ) = f00 (x, x/, ), where
I1
Xh(x, ) I 1
f00 (x, X, ) = X k k (x, ), (4.42)
(1 + X)I k
k=0
and clearly f00 (x, X, ) C ([0, 1] [0, ] [0, o ]). Similarly, for m 1,
in view of the expression (4.21) for the integral (4.20), there exist k (x, )
C ([0, 1] [0, o ]) and ck () C ([0, o ]) such that
I1
(x/)h(x, ) m
k
wm (x, ) = (x/) k (x, ) + ck ()Lk (x/) . (4.43)
(1 + x/)I
k=0 k=1
m
wm (x, ) = fmk (x, x/, )Lk (x/). (4.44)
k=0
N 1
n
y(x, ) = w0 (x, ) + n dmn wm (x, ) + O(N ), (4.45)
n=0 m=1
it is clear there exists nk (x, X, ) C ([0, 1] [0, ] [0, o ]), just linear
combinations of the fmk (x, X, ), such that
N 1
n
y(x, ) = n nk (x, x/, )Lk (x/) + O(N ). (4.46)
n=0 k=0
N1
n
y(x, ) = n [unm (x) + vnm (x/)]Lm (x/) + O(N ) (4.47)
n=0 m=0
uniformly as 0+ for 0 x 1.
Of course, we want to calculate the terms of (4.47) from the dierential
equation (4.35). It is apparent from (4.47) that y(x, ) has an inner expansion
of the familiar form,
N1
IN y(x, ) = n Yn (X), (4.48)
n=0
[k]
where unmk = unm (0). In untangled form, we nd, with help from Maple,
n
Yn (X) = Znm (X)Lm (X), (4.50)
m=0
where,
Z00 (X) = u000 + v00 (X), (4.51)
Z10 (X) = u100 + u001 X + v10 (X), (4.52)
4.2 Problem F 55
nm
rnm (X) = rnmk X k , (4.61)
k=0
N1
n1
ON y(x, ) = z00 (x) + n lnm ()znm (x), (4.63)
n=1 m=0
The rst part of the Maple program P robF , given below, calculates Yk (X)
for 0 k N 1, once a(x, ), b(x, ) and N are prescribed, in the usual
way from the dierential equation for Y (X, ). Then each Zij (X) for 0 i
N 1 and 0 j i is isolated and split into rij (X) and vij (X). Note that
this involves a procedure for expanding Zij (X) as X out to precisely
O(X N ). The resulting series expansions for uij (x) as x 0 and vij (X) as
X are used in the outer expansion calculations. For example, if we take
a(x, ) = 2 + x, b(x, ) = 1 and N = 4, P robF yields
1 6X 1
v00 (X) = , v10 (X) = , (4.71)
2(1 + X)2 12(1 + X)2
X 21X + 8
v11 (X) = , v20 (X) = , (4.72)
2(1 + X)2 36(1 + X)2
13X 6 X
v21 (X) = , v22 (X) = , (4.73)
12(1 + X)2 4(1 + X)2
and in accordance with (4.62), P robF also yields
1 1 1 1 3
u00 (x) = x + x2 x + O(x4 ), u11 (x) = O(x3 ), (4.74)
2 6 24 120
1 1 29 2
u10 (x) = x+ x + O(x3 ), u22 (x) = O(x2 ), (4.75)
12 18 144
1 1 2 131
u21 (x) = x + O(x2 ), u20 (x) = + x + O(x2 ). (4.76)
2 4 9 2160
In addition, the P robF substitution of (4.63) into (4.35), still with a(x, ) =
2 + x, b(x, ) = 1 and N = 4, yields the sequence of dierential equations
xz00 + (2 + x)z00 = 1, xz10 + (2 + x)z10 = z00 (x), (4.77)
xz20 + (2 + x)z20 = z10 (x), xz21 + (2 + x)z21 = 0, (4.78)
xz31 + (2 + x)z31 = z21 (x), xz32 + (2 + x)z32 = 0, (4.79)
and from the last lines of P robF we get
To complete the calculations for this example, from (4.80a) and (4.74a)
we see z00 (0) = 1/2 and therefore the solution to (4.77a) is
1 x
u00 (x) = (e 1 x) (4.84)
x2
Next, the solution to (4.77b) such that z10 (0) = u10 (0) = 1/12 is
ex 2
u10 (x) = [E(x) + (ex 1 x)], (4.85)
x2 x
x
where E(x) = 0
t1 (et 1) dt. In addition, we readily nd
1
u11 (x) = 0, u22 (x) = 0, u21 (x) = (1 ex ), (4.86)
2x
and nally, the (convergent) series solution to (4.78a) yields
2 131 977 2 4367 3
u20 (x) = + x x + x + O(x4 ). (4.87)
9 2160 8640 302400
Thus we have computed all the terms of (4.47) with N = 3, assuming
a(x, ) = 2 + x, b(x, ) = 1.
P robF := proc(a, b, N )
A := subs(x = X, a); B := subs(x = X, b);
IN y := sum(n Yn , n = 0..N 1);
IN dy := sum(n dYn , n = 0..N 1);
IN de := series((1 + X) IN dy + A IN y B, = 0, N );
IN de := convert(%, polynom);
for k from 0 to N 1 do
temp := coe (IN de, , k);
de := subs(Yk = z(X),dYk = di (z(X), X), temp) = 0;
dsolve({de, z(0) = 0});
Y k := rhs(%); dY k := di (Y k, X);
IN de := subs(Yk = Y k, dYk = dY k, IN de);
Yk := Y k; end do;
for i from 0 to N 1 do
Zi,0 := coe (Yi , ln(1 + X), 0);
for j from 1 to i do
Zi,j := X coe (Yi , ln(1 + X), j);
end do; end do;
for i from 0 to N 1 do
for j from 0 to i do
if Zi,j = 0 then sZij := 0 else
temp := 0; k := N 1;
while order(temp) < N do
58 4 Logarithm Problems
where, for n 1,
First, there exists B > 0 such that |b(x, )|, |c(x, )|, |d(x, )| B for all
(x, ) [0, 1] [0, o ]. Therefore |w0 (x, )| BF (x, ) on [0, 1] (0, o ],
where x
2 2
F (x, ) = 1 e(x/) (x)
e(t/) (t)
dt, (4.95)
0
x
2
(x) = x ta(t) dt. (4.96)
0
/ x/
2
= f (x, x/), (4.98)
1 + (/) 1 + (x/)2
where
(1 + X 2 )[(x)]1/2
f (x, X) = , (4.99)
1 + X 2 (x)
so f (x, x/) = O(1) on [0, 1] (0, o ], by Theorem 1. Therefore F (x, )
A(x/)/[1 + (x/)2 ] for some constant A > 0. Furthermore, X/(1 + X 2 )
2/(1 + X) for 0 X . Therefore
x
1 dt
|w0 (x, )| 2AB 2AB ln(1 + 1/) (4.100)
0 1 + t/
|b(x, )w0 (x, ) + c(x, )w0 (x, )| 4AB 2 ln(1 + 1/) (4.101)
we know the rst two converge absolutely and uniformly for all (x, )
[0, 1] (0, o ], assuming o is set suciently small, and therefore the second
sum is the derivative of the rst. The third sum also converges absolutely
and uniformly, and therefore is the second derivative of the rst sum. This is
seen by summing both sides of (4.94) times n , and this shows simultaneously
that the rst sum in (4.102) is indeed the desired solution of (4.89).
To ascertain the asymptotic form of w(x, ), we need to examine the terms
of (4.93) in more detail. Let m (x, ) = xLm (x/) for m 1 and 0 (x, ) = 1.
Proposition 1. Assume f (x, X, ) C ([0, 1] [0, ] [0, o ]) and let
x
1 (x/)2 (x) 2
gm (x, ) = e e(t/) (t) f (t, t/, )m (t, ) dt, (4.103)
0
4.3 Problem G, Part 1 61
where 0 < (x) C ([0, 1]). There exists mnk (x) C ([0, 1]) and
mnk (X) C ([0, ]) with mnk () = 0 such that for any N 0,
N 1
m
n
gm (x, ) = [mnk (x) + mnk (x/)]k (x, ) + O(N ) (4.104)
n=0 k=0
Proof. We know this is true for m = 0 from our work on Problem E. For
m 1, we readily get (4.104) from (4.103) using integration by parts. In
particular,
g1 (x, ) = g0 (x, )1 (x, ) g0 (x, ), (4.105)
where g0 (x, ) is the same as g0 (x, ), except g0 (t, )/(1 + t/) occurs in place
of f (t, t/, ). Therefore (4.104) holds for m = 1 with 1,0,1 (x) = 0 and
1,0,0 (x) = 1,0,0 (X) = 1,1,0 (x) = 0. For m 2,
where gm1 (x, ) is the same as gm1 (x, ), except mg0 (t, )/(1 + t/) oc-
curs in place of f (t, t/, ). Therefore m,0,m (x) = 0 and if we assume
m1,0,k (x) = 0 for 0 k m 1, m1,0,0 (X) = 0 and m1,1,0 (x) = 0,
then m,0,k (x) = 0 for 0 k m 1, m,0,0 (X) = 0 and m,1,0 (x) = 0.
N 1
w0 (x, ) = n [u0,n,0 (x) + v0,n,0 (x/)] + O(N ), (4.107)
n=0
where u0,n,0 (x) C ([0, 1]), v0,n,0 (X) C ([0, ]) and u0,0,0 (x) = 0.
Therefore, as in the rst paragraph of Section 4.1,
N 1 1
w0 (x, ) = n [U0,n,k (x) + V0,n,k (x/)]k (x, ) + O(N ) (4.108)
n=0 k=0
There exists mnk (x) C ([0, 1]) and mnk (X) C ([0, ]) such that
N1
m+1
gm (x, ) = n [mnk (x) + mnk (x/)]k (x, ) + O(N ) (4.110)
n=0 k=0
Proof. This follows from the second paragraph of Section 4.1, in particular,
from our analysis of (4.20) with n = 0, or, independently, using integration
by parts. We leave the details as an exercise.
At this point, from (4.107) and (4.108), and Corollary 1, we can see
where nk (x) C ([0, 1]), nk (X) C ([0, ]) and 0,0 (x) = 0,0 (X) =
0. Therefore, by Proposition 1, there exists u1,n,k (x) C ([0, 1]) and
v1,n,k (X) C ([0, ]) such that, after canceling an factor,
N1 1
w1 (x, ) = n [u1,n,k (x) + v1,n,k (x/)] lnk (1 + x/) + O(N ). (4.112)
n=0 k=0
N1 2
w1 (x, ) = n [U1,n,k (x) + V1,n,k (x/)]k (x, ) + O(N ), (4.113)
n=0 k=0
where, in particular, U1,0,0 (x) = V1,0,0 (X) = 0. Obviously, we can iterate this
process indenitely. Hence we conclude, there exists Umnk (x) C ([0, 1])
and Vnmk (X) C ([0, ]) such that for every m 0,
N 1
m+1
wm (x, ) = n [Umnk (x) + Vnmk (x/)]k (x, ) (4.114)
n=0 k=0
We can see upon substituting (4.114) into (4.93) that if we let w(x, ) =
2 w(x, ), then there exists unm (x) C ([0, 1]) and vnm (X) C ([0, ])
such that, for any N 0,
N1
n
) =
w(x, n [unm (x) + vnm (x/)]Lm (x/) + O(N ) (4.116)
n=0 m=0
N1
) =
IN w(x, n (X),
n W (4.117)
n=0
n
n (X) =
W Znm (X)Lm (X) (4.118)
m=0
and for the leading terms we have (4.51)-(4.60). Likewise, for the N-term
outer expansion
N1
n1
) = z00 (x) +
ON w(x, n lnm ()znm (x), (4.119)
n=1 m=0
) is
we have (4.64)-(4.70). The dierential equation for w(x,
2 w
+ [xa(x) + 2 b(x, )]w
+ c(x, )w
= 2 d(x, ), (4.120)
xa(x)z32 = 0, (4.125)
xa(x)z31 = c0 (x)z21 (x), (4.126)
where b0 (x) = b(x, 0), c0 (x) = c(x, 0), d0 (x) = d(x, 0) and c1 (x) = c[0,1] (x, 0).
From the dierential equation for W (X, ), it is apparent that
+ 2X W
W = 0 (4.127)
0 0
n (0) = W
and therefore, since W n (0) = 0 for all n 0, W
0 (X) = 0. From
(4.51), rst with X = , this means
u10 (0) = u00 (0) = u11 (0) = 0, v10 (X) = v11 (X) = 0. (4.130)
u20 (0) = d00 Q(0), u10 (0) = u00 (0) = 0, v20 (X) = d00 Q(X). (4.134)
1
u11 (0) = d00 , u21 (0) = u22 (0) = 0, v21 (X) = v22 (X) = 0. (4.135)
2
), from (4.121), (4.64) and
Turning to the outer expansion of w(x,
(4.128a), we see
z00 (x) = u00 (x) = 0. (4.136)
Therefore from (4.122), along with (4.65) and (4.128b), we also have
4.4 Problem G, Part 2 65
+ 2X W
W = d01 + d10 X c00 W
2 (X) (b00 + a1 X 2 )W
(X), (4.145)
3 3 2
2 (X) is
where b00 = b(0, 0), c00 = c(0, 0), a1 = a (0), dij = d[i,j] (0, 0) and W
given by (4.133). If we denote the right side of (4.145) by f (X), then
X
3 (X) = eX 2
W
2
eT f (T ) dT (4.146)
0
and
1
f (X) = + X c00 d00 ln(1 + X) + (X), (4.147)
2
where
1
= d01 c00 d00 Q(0), = (2d10 d00 a1 )X (4.148)
2
and (X) C ([0, ]). Indeed,
1
(X) = c00 d00 Q(X) b00 d00 P (X) a1 d00 [2X 2 P (X) X], (4.149)
2
66 4 Logarithm Problems
Also
X
2 2
eX eT ln(1 + T ) dT = P (X) ln(1 + X) + (X), (4.151)
0
where X
2 2 dT
(X) = eX eT P (T ) , (4.152)
0 1+T
so (X) = O(X 3 ) and (X) C ([0, ]). Therefore
When we substitute the results of the last section, beginning with (4.141),
into (4.116), and also note that ln(1 + 1/) = ln(1/) + O(), we get, for the
solution of (4.89) with w(0, ) = w (0, ) = 0, the uniformly valid expansion
for 0 x 1, where
1 ln(1 + x/)
w11 (x, ) = d00 , (4.162)
2 ln(1 + 1/)
1 ln(1 + x/)
z21 (x, ) = c00 . (4.168)
2 ln(1 + 1/)
Returning now to (4.91), to determine (), we have w(1, )+()z(1, ) =
() and a straightforward calculation, utilizing 1/z(1, ) = 1+ 12 c00 ln(1/)+
O(), reveals
68 4 Logarithm Problems
0.8
0.4
0.0
0.4
Fig. 4.1 Numerical solution to (4.88) and an asymptotic approximation to the solution
when a(x, 0) = 2 + x, b(x, = 0, c(x, ) = cos(x), d(x, ) = 1 + x + , y(0, ) = 0, y(1, ) = 1
and = 0.1
.
1
() = d00 ln(1/) + [0 d00 Q() D(1)]
2
1
c00 d00 ln2 (1/) + O( ln(1/), (4.169)
8
where 0 = (0). When we substitute this, along with the above expansions
for w(x, ) and z(x, ) into (4.91), we get, nally, for the solution to Problem
G,
where
1 ln(1 + x/)
y11 (x, ) = d00 erf(x/) , (4.171)
2 ln(1 + 1/)
y20 (x, ) = [0 d00 Q() D(1)]erf(x/) + d00 Q(x/) + D(x), (4.172)
1 ln(1 + x/) ln2 (1 + x/)
y22 (x, ) = c00 d00 2 2 erf(x/) , (4.173)
8 ln(1 + 1/) ln (1 + 1/)
and this holds uniformly for 0 x 1 as 0+ .
A comparison of the numerical solution of (4.88) and the asymptotic result
(4.170)-(4.173), assuming a(x) = 2 + x, b(x, ) = 0, c(x, ) = cos(x), d(x, ) =
1+x+ and () = 1, in which case D(x) = 12 ln(1+x/2), is shown for = 0.1
4.6 Exercises 69
y(x, ) = ln(1/)w11 (x, ) + [w20 (x, ) + 0 z20 (x, )] + ln2 (1/)w22 (x, )
+ ln(1/)[w21 (x, ) + 0 z21 (x, )] + O(). (4.174)
4.6 Exercises
n1
m+1
y(x, ) = k xnk ak (x, ) + n ck () lnk (1 + x/) (4.176)
k=0 k=1
for all (x, ) [0, 1] (0, o ]. Do this rst for m = 0 and then, by induction,
for arbitrary m 0. Now use this to show that (4.21) follows from (4.20),
where n 0, m 1.
4.2. Use the Maple routine of Section 1.5 to conrm the expansion results
stated at the end of Section 4.1 for y(x, ) given by (4.27).
11 11 17 169 3 388 4 97
u20 (x) = + x+ x2 + x + x + x5 +O(x6 ), (4.181)
12 48 90 5760 302400 120690
1 1 1
u21 (x) = x2 x4 + O(x6 ), (4.182)
2 8 96
and evidently u11 (x) = u22 (x) = 0. At least, u11 (x) = O(x6 ), u22 (x) = O(x6 ).
5.1 Problem H
where a(x, ), b(x, ) C ([0, 1] [0, o ]) for some o > 0. Assume also that
a(x, ) > 0 and y(0, ) = 0. We know from Problem A that
x
y(x, ) = 1 f (x, t, t/, ) cos((x t)/) dt, (5.2)
0
where
f (x, t, T, ) = b(x t, )eT u(x,t,) , (5.3)
with x
u(x, t, ) = t1 a(s, ) ds, (5.4)
xt
and we could expand f (x, t, t/, ) in accordance with Corollary 2, but this
would not be immediately useful. Instead, assuming b(x, ) is real valued, like
a(x, ) > 0, we note from (5.2) that y(x, ) = Re[w(x, )], where w(x, ) =
z(x, )eix/ and, if we let
where
v(x, t, ) = u(x, t, ) + i, (5.6)
then x
z(x, ) = 1 (x, t, t/, ) dt. (5.7)
0
Now, since Re[v(x, t, )] = u(x, t, ) > 0 for all (x, t, ) [0, 1] [0, x] [0, o ],
we have (x, t, T, ) C ([0, 1] [0, x] [0, ] [0, o ]), and therefore, by
Corollary 2, as in Problem A, there exists un (x) C ([0, 1]) and vn (X)
C ([0, ]) with vn () = 0 such that for any N 0,
N 1
z(x, ) = n [un (x) + vn (x/)] + O(N ) (5.8)
n=0
b(x, 0)
O1 z(x, ) = (5.10)
a(x, 0) + i
b(0, 0)
I1 z(x, ) = 1 e[a(0,0)+i]x/ . (5.11)
a(0, 0) + i
Consequently,
where
a(x, 0)b(x, 0) b(x, 0)
p0 (x) = , q0 (x) = , (5.14)
1 + [a(x, 0)]2 1 + [a(x, 0)]2
and
r0 (X) = p0 (0)ea(0,0)X . (5.15)
In general, if y(x, ) is the solution to (5.1), where a(x, ), b(x, ) are real
valued, a(x, ) > 0 and y(0, ) = 0, then
where
N 1
N 1
p(x, ) = n pn (x) + O(N ), q(x, ) = n qn (x) + O(N ), (5.17)
n=0 n=0
5.1 Problem H 73
and
N1
r(x, ) = n rn (x/) + O(N ) (5.18)
n=0
Proposition 3. Assume g(x), h(x) C ([0, 1]) are real valued, h(0) = 0,
and x
F (x, ) = 1 eih(t)/ g(t) dt. (5.19)
0
(a). If h (x) = 0 for 0 x 1, then there exists c(), un (x) C ([0, 1])
such that
F (x, ) = c() + (x, )eih(x)/ , (5.20)
where, for any N 0,
N1
(x, ) = n un (x) + O(N ) (5.21)
n=0
uniformly as 0+ for 0 x 1.
(b). If h (x) > 0 for 0 < x 1, h (0) = 0 and h (0) > 0, then there ex-
ists c(), un (x) C ([0, 1]) and vn (X) C ([0, ]) such that
N 1
(x, ) = n [un (x) + vn (x/)] + O(N ) (5.23)
n=0
uniformly as 0+ for 0 x 1.
where x
F1 (x, ) = 1 eih(x)/ u0 (t) dt. (5.25)
0
74 5 Oscillation Problems
Obviously we can repeat this process indenitely, and this proves Part (a) of
the proposition. Indeed, we have un (x) = iun1 (x)/h (x) for n 1 and we
can choose c() to be any function in C ([0, 1]) such that c[n] (0) = un (0).
To prove Part (b), let s = [h(x)]1/2 , let x = (s) denote the inverse of this
transformation, and let G(s) = g((s)) (s). Then
[h(x)]1/2
2
F (x, ) = 1 ei(s/) [G(0) + 2isG1 (s)] ds, (5.28)
0
where 2
H(x, ) = eix /
[ (x/)G(0) + G1 (x)], (5.30)
ix2 2
(x) = e eis ds, (5.31)
x
and
[h(x)]1/2
2
1
F1 (x, ) = ei(s/) [G1 (0) + 2isG2 (s)] ds. (5.32)
0
where G2 (s) = (2is)1 [G1 (s) G1 (0)]. So again we have a process we can
iterate indenitely. Furthermore, (x) C ([0, ]), as we see in Exercise
5.2, and therefore we can expand ([h(x)]1/2 /) = ((x/)[h(x)/x2 ]1/2 ) ac-
cording to Theorem 1. This leads directly to (5.22)-(5.23).
For our nal problem, we seek an asymptotic expansion for the solution to
satisfying the initial conditions y(0, ) = (), y (0, ) = 1 (). The func-
tions a(x, ), b(x, ) C ([0, 1] [0, o ]) and (), () C ([0, o ]) for
some o > 0 are assumed to be real, and we assume a(x, ) > 0. We further
assume, for some xo (0, 1), that a00 (x) < 1 for 0 x < xo and a00 (x) > 1
for xo < x 1, where, in general, aij (x) = a[i,j] (x, 0), bij (x) = b[i,j] (x, 0).
Finally, we assume a10 (xo ) = 0.
If (instead) a(x, ) = (1 + c)2 , where c is a constant, and b(x, ) = 1/2 ,
then
()
y(x, ) = () cos[(1 + c)(x/)] + sin[(1 + c)(x/)] + yp (x, ), (5.34)
1 + c
where, in the particular solution
sin(cx) 1 cos(cx)
M (x, ) = , N (x, ) = (5.36)
c(2 + c) c(2 + c)
N1
u(x, ) = n un (x) + O(N ), (5.39)
n=0
where x
1
r(x) = [a00 (t)]1/2 a01 (t) dt. (5.41)
2 0
N 1
(x, ) = n [un (x) + vn (x/)] + O(N ) (5.46)
n=0
where A (x, ) has a uniformly valid expansion of the same form as (x, ).
Similarly, by Part (a) of Proposition 5.1, we nd
(2) (2)
B(x, )wh (x, ) = cB ()wh (x, ) + B (x, )eix/ , (5.48)
where B (x, ) also has a uniformly valid expansion like (x, ), although
there are no vn (x/) terms. Consequently, when we substitute (5.47) and
(5.48) into (5.42) we see there is a particular solution w(+) (x, ) = z (+) (x, )eix/
of the dierential equation for w(x, ), in addition to wp (x, ), such that, for
(+) (+)
certain un (x) C ([0, 1 xo ]), vn (X) C ([0, ]),
5.3 Problem I, Part 2 77
N 1
z (+) (x, ) = n [u(+) (+) N
n (x) + vn (x/)] + O( ) (5.49)
n=0
uniformly as 0+ for 0 x 1 xo .
The dierential equation for z (+) (x, ) is
N1
ON z (+) (x, ) = n zn(+) (x), (5.51)
n=0
(+)
whose existence is implied by (5.49), it is apparent that z0 (x) = 0. Also,
for the leading term of the N -term inner expansion of z (+) (x, ),
N1
IN z (+) (x, ) = n Zn(+) (X), (5.52)
n=0
N1
z () (x, ) = n [u() () N
n (x) + vn (x/)] + O( ), (5.58)
n=0
To solve now for y(x, ) = Re[w(x, )], rst we note that, since the dif-
ference between any two particular solutions of the dierential equation for
w(x, ) is a solution to the corresponding homogeneous dierential equation,
there exists c1 (), c2 () such that
(1) (2)
w() (x, ) = w(+) (x, ) + c1 ()wh (x, ) + c2 ()wh (x, ). (5.60)
In addition,
(1)
wh (xo , ) = [u0 (xo ) + O()]eis(xo )/ , (5.63)
(1)
wh (xo , ) = 1 [iu0 (xo ) + O()]eis(xo )/ , (5.64)
(2) (1)
and, of course, wh (xo , ) = wh (xo , ). Therefore, from
(5.60) and its deriva-
tive evaluated at x = xo , and the fact that (0) = ( /2)ei/4 , what we nd
is
b00 (xo ) i{3/4+[xo s(xo )]/r(xo )}
c1 () = e + O() (5.65)
2mo [a00 (0)]1/4
and c2 () = O(). Consequently,
5.3 Problem I, Part 2 79
where
b00 (xo )
0 (x, ) = cos{3/4 [xo s(xo )]/ + r(xo )}, (5.67)
2mo [a00 (x)]1/4
b00 (xo )
0 (x, ) = sin{3/4 [xo s(xo )]/ + r(xo )}. (5.68)
2mo [a00 (x)]1/4
Similarly, the dierence between the particular solution w() (x, ) and the
one that satises the initial conditions w(0, ) = (), w (0, ) = 1 () also
is a solution to the corresponding homogeneous dierential equation. Thus
we nd, for the solution to Problem I,
To deal with Re[w() (x, )] and Re[w(+) (x, )], note that (x) = p(x) +
iq(x), where
p(x) = (cos x2 cos s2 + sin x2 sin s2 ) ds, (5.71)
x
q(x) = (sinx2 cos s2 cosx2 sin s2 ) ds. (5.72)
x
It follows that
b00 (xo )
Re[w() (x, )] = [q(mo (x0 x)/) + O()] cos(x/)
2mo
+ [p(mo (xo x)/) + O()] sin(x/) (5.73)
for 0 x xo , and
b00 (xo )
Re[w(+) (x, )] = [q(mo (x xo )/) + O()] cos(x/)
2mo
+ [p(mo (x xo )/) + O()] sin(x/) . (5.74)
N 1
A(x, ) = n An (x, ) + O(N ) (5.76)
n=0
stemming from (5.49), (5.58) and (5.39), in which each An (x, ) = O(1) for
0 x 1, and there are corresponding expansions for B(x, ), C(x, ) and
D(x, ). In particular, we have determined
2
0.0 0.2 0.4 0.6 0.8 1.0
Fig. 5.1 Uniformly valid asymptotic approximation of the solution to (5.33) when
a(x, ) = 1/4 + 2x + , b(x, ) = 1, y(0, ) = 1), y (0, ) = 0 and = 0.001.
5.4 Exercises 81
2
0.0 0.2 0.4 0.6 0.8 1.0
Fig. 5.2 Outer approximation of the approximate solution to (5.33) shown in Figure 5.1.
5.4 Exercises
a := 2 + x + + I; b := 1 + x2 (5.84)
1 + x2
u0 (x) = (5.85)
x+2+i
and therefore, in agreement with (5.14),
x3 + 2x2 + x + 2 x2 + 1
p0 (x) = , q0 (x) = . (5.86)
x2 + 4x + 5 x2 + 4x + 5
But then
x3 + 3x2 + 5x + 1 + i(x2 + 2x + 1)
u1 (x) = (5.87)
x3+ 6x2 + 9x + 2 + i(3x2 + 12x + 11)
and, with the help of Maples evalc command, this implies
6. F. W. J. Olver, Asymptotics and Special Functions, Academic Press, New York, 1974.
11. L. A. Skinner, Stationary phase theory and passage through resonance, J. Math.
Anal. Appl., 205 (1997), 186-196.
13. Milton Van Dyke, Perturbation Methods in Fluid Mechanics, Parabolic Press, Stanford
California, 1975.