Professional Documents
Culture Documents
OF ORDINARY
DIFFERENTIAL
EQUATIONS
Universitext
Editorial Board
(North America):
S. Axler
F.W. Gehring
K.A. Ribet
Springer
New York
Berlin
Heidelberg
Barcelona
Hong Kong
London
Milan
Paris
Singapore
Tokyo
Universitext
Springer
Po-Fang Hsieh Yasutaka Sibuya
Department of Mathematics School of Mathematics
and Statistics University of Minnesota
Western Michigan University 206 Church Street SE
Kalamazoo, MI 49008 Minneapolis, MN 55455
USA USA
philip.hsieh@wmich.edu sibuya 0 math. umn.edu
Editorial Board
(North America):
S. Axler F.W. Gehring K.A. Ribet
Mathematics Department Mathematics Department Department of
San Francisco State East Hall Mathematics
University University of Michigan University of California
San Francisco, CA 94132 Ann Arbor. Ml 48109- at Berkeley
USA 1109 Berkeley, CA 94720-3840
USA USA
vii
viii PREFACE
larities of linear differential equations is given. In Chapter VI, the main topics are
the basic results concerning boundary-value problems of the second-order linear
differential equations. The comparison theorems, oscillation and nonoscillation of
solutions, eigenvalue problems for the Sturm-Liouville boundary conditions, scat-
tering problems (in the case of reflectionless potentials), and periodic potentials are
studied. The authors learned much about the scattering problems from the book
by S. Tanaka and E. Date [TD]. In Chapter VII, asymptotic behaviors of solu-
tions of linear systems as the independent variable approaches infinity are treated.
Topics include the Liapounoff numbers and the Levinson theorem together with
its various improvements. In Chapter VIII, some fundamental theorems concern-
ing stability, asymptotic stability, and perturbations of 2 x 2 linear systems are
explained, whereas in Chapter IX, results on autonomous systems which include
the LaSalle-Lefschetz theorem concerning behavior of solutions (or orbits) as the
independent variable tends to infinity, the basic properties of limit-invariant sets
including the Poincar6-Bendixson theorem, and applications of indices of simple
closed curves are studied. Those theorems are applied to some nonlinear oscillation
problems in Chapter X. In particular, the van der Pot equation is treated as both
a problem of regular perturbations and a problem of singular perturbations. In
Chapters XII and XIII, asymptotic solutions of nonlinear differential equations as
a parameter or the independent variable tends to its singularity are explained. In
these chapters, the asymptotic expansions in the sense of Poincare are used most
of time. However, asymptotic solutions in the sense of the Gevrey asymptotics are
explained briefly. The basic properties of asymptotic expansions in the sense of
Poincare as well as of the Gevrey asymptotics are explained in Chapter XI.
At the beginning of each chapter, the contents and their history are discussed
briefly. Also, at the end of each chapter, many problems are given as exercises. The
purposes of the exercises are (i) to help the reader to understand the materials in
each chapter, (ii) to encourage the reader to read research papers, and (iii) to help
the reader to develop his (or her) ability to do research. Hints and comments for
many exercises are provided.
The authors are indebted to many colleagues and former students for their valu-
able suggestions, corrections, and assistance at the various stages of writing this
book. In particular, the authors express their sincere gratitude to Mrs. Susan Cod-
dington and Mrs. Zipporah Levinson for allowing the authors to use the materials
in the book Theory of Differential Equations by E. A. Coddington and Norman
Levinson.
Finally, the authors could not have carried out their work all these years without
the support of their wives and children. Their contribution is immeasurable. We
thank them wholeheartedly.
PFH
YS
March, 1999
CONTENTS
Preface vii
ix
x CONTENTS
References 453
Index 462
CHAPTER I
FUNDAMENTAL THEOREMS OF
ORDINARY DIFFERENTIAL EQUATIONS
In this chapter, we explain the fundamental problems of the existence and unique-
ness of the initial-value problem
dy
(P)
dt
= y(to) _ 4o
in the case when the entries of f (t, y) are real-valued and continuous in the variable
(t, y), where t is a real independent variable and y is an unknown quantity in Rn.
Here, R is the real line and R" is the set of all n-column vectors with real entries.
In §I-1, we treat the problem when f (t, y) satisfies the Lipschitz condition in W. The
main tools are successive approximations and Gronwall's inequality (Lemma
In §1-2, we treat the problem without the Lipschitz condition. In this case, ap-
proximating f (t, y) by smooth functions, f-approximate solutions are constructed.
In order to find a convergent sequence of approximate solutions, we use Arzeli -
Ascoli's lemma concerning a bounded and equicontinuous set of functions (Lemma
1-2-3). The existence Theorem 1-2-5 is due to A. L. Cauchy and G. Peano [Peal]
and the existence and uniqueness Theorem 1-1-4 is due to E. Picard [Pi1 and E.
Lindelof [Lindl, Lind2J. The extension of these local solutions to a larger interval
is explained in §1-3, assuming some basic requirement.,, for such an extension. In
§I-4, using successive approximations, we explain the power series expansion of a
solution in the case when f (t, y) is analytic in (t, y). lit each section, examples and
remarks are given for the benefit of the reader. In particular, remarks concerning
other methods of proving these fundamental theorems are given at the end of §1-2.
Y1, f,
y2 f2
y= !y = max{jy,j 1Y21-.. ItYn11.
yn fn
In §§I-1, 1-2, and 1-3 we consider problem (P) under the following assumption.
Assumption 1. The entries of f (t, y) are real-valued and continuous on a rectan-
gular region:
I
2 I. FUNDAMENTAL THEOREMS OF ODES
Since f (t, yam) is continuous on R, I f (t, yul is bounded. Let us denote by M the
maximum value of I f (t, y-) I on R, i.e., M = mac I f (t, y) I. We define a positive
number a by
a if M = 0,
a
min Ca, M if M > 0.
l
To locate a solution in a neighborhood of its initial point, the number or plays an
important role as shown in the following lemma.
Lemma I-1-1. If ff = ¢(t) is a solution of problem (P) in an interval it - tol <
a < a, then I'(t) - 41 < b in It - tol < a, i.e., (t,5(t)) E R((to,co),a,b) for.
It - tol <a.
Proof
Assume that Lemma 1-1-1 is not true. Then, by the continuity of fi(t), there
exists a positive number $ such that
(i) < a,
(ii)
1¢(t) - 41 < b for It - tol <'6,
I4(to+0) -QoI =b or 10(to-fl)-qol
The assumption a < a < a implies (3 < a. Hence, (t,j(t)) E R for It - tot < p.
Therefore, I f (t, (t))I 5 M for it - to! < 0. From 4'(t) = f (t, ¢(t)) and 4(to) = 4o,
it follows that
Hence,
I4(t) - 4)1 = I1 f(s, i(s))ds < MIt - tol for It - tot < P.
r
This implies that
Set v(t) =
Jio g(s)ds. Then, by (1.1.5), ddtt) < K+Lv(t) and v(to) = 0. Hence,
(1.1.7) {exp[-L(t - to)]v(t)} < Kexp[-L(t - to)]
and, consequently,
Step 1. Each function 0k is well defined and (t,45k(t)) E R for It - tol < a
(k = 0,1,2,...).
Proof.
MLk
Ilk+1(t) - mk(t)I < (k 1)1It - toIk+1 for It - tol < a, k =0,1,2,... .
Proof.
For k = 0, we have 1m1(t) - y5o(t)I = J r f(s,co)ds` < MIt - tol. Assume that
I
co I
ML
I&(t)(t)I < k1-1 It - tolk for it - tol < a. Then,
<L I J r
to
t k
AkL
I r It - toIkdsl = (ML1)r It - tolk+1.
+oo
Step 3. The series E (&(t) - 4k-1(t)) is uniformly convergent on the interval
k=1
It - tol<a.
Proof.
By virtue of the result of Step 2, for a given c > 0, there exists a positive integer
N such that
°° M °o (LIt - tol)k M °O (La)k
/ <
E I&(t) - 0k-I(t)I < L k1 <L k!
k=N k=N k=N
1. EXISTENCE AND UNIQUENESS WITH THE LIPSCHITZ CONDITION 5
Use the identity ¢N (t) = do (t) + E ($k (t) - $k_1(t)) and the result of Step 3.
k=1
L f Id(s) - +L(s)Ids
o
on to < t < to+&, we conclude that I¢(t) -tG(t)I = 0 on the interval to < t < to+&.
Similarly, the same conclusion is obtained on the interval to - & < t < to. Steps 1
through 6 complete the proof of Theorem 1-1-4. 0
The following lemma gives a sufficient condition that f (t, y") satisfies Assumption
2 (i.e., the Lipschitz condition).
6 I. FUNDAMENTAL THEOREMS OF ODES
Lemma I-1-6. If Of (t' y-) (j = 1, 2, ... , n) exist, are continuous, and satisfy the
8yj
condition: I of (t, y j < Lo (j = 1, 2, .. - , n) for (t, y) E I x D, where I is an
1
Iy' -Y21
If(t,Yi) - f(t,y2)1 < 5
1. EXISTENCE AND UNIQUENESS WITH THE LIPSCHITZ CONDITION 7
on the region R= {(t, y) : It-11 < f, Iy-51 < 2} (cf. Exercise I-2). Furthermore,
If(t,y)I <- 1
(3_(/)2)(9_22) = 1
5
1
Set a = V25, b = 2, and M = . Then,
5
Therefore, the given initial-value problem has one and only one solution on the
interval It - 11 < f.
Problem 2. Using Theorem I-1-4, show that the initial-value problem
(P) dy = 1 y()
4 = 3
dt (1 + (t - 4)2)(5 + (y - 3)2)'
has one and only one solution on the interval -oo < t < +oc.
Answer.
The function
1
,y
(1 + (t - 4)2)(5 + (y - 3)2)
is continuous and satisfies the Lipschitz condition everywhere in the (t, y)-plane.
Also, If (t, y)I < everywhere. This implies that a = min(a, 5b), if Problem (P)
is considered in a region
Hence, if a < 5b, Problem (P) has one and only one solution on the interval it - 41 <
a. Now, this solution is independent of a due to the uniqueness. Therefore, we can
conclude that (P) has one and only one solution on the interval -oo < t < +oo.
Remark 1-1-7. It is usually claimed that the general solution of the system
dt = f (t, y-) depends on n independent arbitrary constants. Theorem 1-1-4 verifies
this claim if we define the general solution very carefully, since the initial data (to, Cl
represent n arbitrary constants for a fixed to. However, if the reader looks into this
definition a little deeper, he will find various complicated situations. It is important
for the reader to know that the number of independent arbitrary constants con-
tained in the general solution depends strongly on the space of functions to which
the general solution should belong. For example, if the differential equation
1
(E) tae =
8 I. FUNDAMENTAL THEOREMS OF ODES
is considered on the set j = {t E R : t A 0), the general solution is y(t) = cl +In Jtj
for t > 0 and y(t) = c2 + In JtJ for t < 0, where ct and c2 are independent arbitrary
constants. If the Heaviside function
H(t) =
{1 if t > 0,
-I if t<0
is used, this general solution is given by
c1 - c2 H(t)
y(t) = + c' + c2 + In JtI.
2 2
Equation (E) may be looked at in terms of the generalized functions (or the distri-
butions of L. Schwartz [Sc]). The transformation y = u + In Jti, changes equation
(E) to
(E') t - = 0.
As a differential equation on the generalized functions, (E') is reduced to
(E") j = cb(t),
where c is an arbitrary constant and b(t) is the Dirac delta function. Integrating
(E"), we obtain u(t) = cH(t) + c, where c is another arbitrary constant. Therefore,
the general solution of (E) is y(t) = cH(t) + c + in Jtt. For more information on
differential equations on the generalized functions, see [Ko] for examples.
The following example might be a little strange. Let us construct solutions of
the differential equation
(E) L + 2y(cost + cos(2t)) = 0
Proof.
Choose from F a sequence of subsequences F. 1, 2.... }, j=
1, 2, ... , such that
(i) .Fj +1 C .Fj for every j,
(ii) t lim o f',,e(rl) = c, exists for every j.
To do this, first look at functions in F at t = r1. Using the boundedness of the
set {f(r1) : f E .F}, we choose an infinite subsequence F1 = {f 1.t : t = 1, 2,... }
from .F so that lim f1,t(r1) = cl exists. Next, look at the sequence F1 at t = r2.
1 +00
Using the boundedness of the set { f l,t(r2) : t = 1, 2.... }, we choose an infinite
subsequence F2 = {j 2 t : t = 1,2.... }from F1 so that lim h,t(r2) = c2 exists.
Continuing this way, we can choose subsequences .F,+1 from Y. successively.
Set fh = fh,h (h = 1, 2, 3.... ). Then, ft, E fl for every h >) since fh = fh,h E
.Fh C .Fj if h > j. Therefore, lim fh(rj) = c, exists for every j, i.e., lira fh(r)
- h-+oo h-+oo
exists for every rational number r E Q.
10 I. FUNDAMENTAL THEOREMS OF ODES
choose a rational number rt(e) E 1t f1 Q. For all t E It, set r t,b1 3) I = rt(E)
/E \
and set
1
N(e) = maxE N (r,(.E), 3 )
1
Proof.
We prove this lemma in three steps.
2. EXISTENCE WITHOUT THE LIPSCHITZ CONDITION 11
{f(t on R,
F(t, yJ =
0 for It-tot <a and If-cal>b+1.
The construction of such an f is left to the reader as an exercise. Since f is
uniformly continuous on R, properties (1), (2), and (3) of f imply that
(4) for every positive number e, there exists a positive number b(e) such that
Step 2. For every positive number b, there exists a real-valued function p(t, 6) of
a real variable t such that
(a) p has continuous derivatives of all orders with respect to for -cc < t < +oo,
(b) p(4, b) > 0 for -co < t < +oc,
(c) p(.,b)=0forI{I>b,
(d) f(.5)de = 1.
,o
The construction of such a function p is left to the reader as an exercise.
Set 0(g,6) = p(yi, b)p(y2, b) ... p(y,,, b). Then
(a') has continuous partial derivatives of all orders with respect to y1, y2,.,. ,y
for ally",
(b') t(#, 6) 0 for all
(c') 0(f, b) = 0 for Iyj > 6,
(d')
f
r +00 ... f +00
'(y, 6)dy1dy2 ... d1M = 1.
Step 3. Set
+- ...
4(t, YJ = 1-00
J+oo (y - 1, b(E))F(t, ff)d ... dnn,
00
where b(e) is defined in (4) of Step 1. Then, F,(t, y-) satisfies all of the requirements
(i), (ii), (iii), and (iv) of Lemma 1-2-4.
Proof
Properties (i) and (ii) are evident. Property (iii) follows from the estimate
< iTMaExRIf(T,nl1.
12 I. FUNDAMENTAL THEOREMS OF ODES
=l
r+oo
j + (y-;),b(E)){F(t,n7)-F(t,)}dg1...dg
00 l
Step 3. Using Lemma 1-2-3 (Arzeli -Ascoli) choose a sequence {ee : I = 1, 2,. .. } of
positive numbers such that lim e, = 0 and that the sequence l = 1, 2.... }
r+oo
converges uniformly on It - tot < or as t -- +oo. Then, set
,(t) = lim
+oo
,,(t) on It - to, < a.
and that
However, Theorem I-1-4 does not apply to (P). In fact y(x) = 0 is a solution of
Problem (P) and also
0 for x < 3,
y(x) _ (2(x2_ 9)154
for x > 3
5
is a solution of Problem (P). Note that the right-hand side of the differential
equation (P) does not satisfy the Lipschitz condition on any y-interval containing
y=0.
Two other methods of proving Theorem 1-2-5 are summarized in the following
two remarks.
Remark 1-2-7. Let every entry of an n-dimensional vector f (t, y-) be a real-valued
continuous function of n + 1 independent variables t and y = (yl, ... , y,,) on a
rectangular region R = {(t, y-) : It - tot < a, Iy"-cod < b}. Assume that I f(t,yj 1 < M
14 I. FUNDAMENTAL THEOREMS OF ODES
(1.2.2)
dt = f (t, J- , i(to) = 0o
uniformly on to < t < to + a as j -+ +oo.
For more details, see [CL, pp. 3-5].
Remark 1-2-8. We use the same assumption, M, and a as in Remark 1-2-7. Also,
let i1(t) be a continuously differentiable function of t such that #(to) = co and
(t, #(t)) E R on the interval to - r < t < to, where r is a positive number. For
every e such that 0 < e < r, define
q(t) for to - r < t < to,
c
1%c(t) =
clo+ff(s,il(s_6))ds for to < t < to + a.
o
Remark 1-2-9. If Ax, y-) is assumed to be measurable for each fixed y, continuous
in y for each fixed t, and I f (t, yj I is bounded by a Lebesgue-integrable function when
(t, y") E R, a result similar to Theorem 1-2-5 (Caratheodory's existence theorem)
[Ca, pp. 665-688) is obtained. Similarly, if, in addition, it satisfies an inequality
similar to Lipschitz condition with the Lipschitz constant replaced by a Lebesgue-
integrable function L(t) when (t, y-) E R, we obtain also a result similar to Theorem
1-1-4. (See [CL, p. 43), [Har2, p. 10], and [SC, p. 15).)
(ii) a function fi(t) satisfies the condition 4 (t) = f(t, ¢(t)) and (t, fi(t)) E D, in
an open interval I = {t : r1 < t < r2},
(iii) 4(t) cannot be extended to the left of rl(or, respectively, to the right of r2)
with property (ii),
(iv) jlimo(tj,0(ti)) = (TI,fl) (or, respectively, (r2, r))) exists for some sequence
{tj :j = 1,2....l of points in the interval Z,
then the limit point (Ti, n (or, respectively, (T2, i))) must be on the boundary of V.
Proof
We shall prove this result by deriving a contradiction from the assumption that
(r1,,7) E E) (or, respectively, (,r2, i-11 E D). Applying Lemma 1-3-1 to this situation,
we obtain cr
lim(t,¢(t)) = (rl,r) (or, respectively, (r2ir))). Hence, by applying
Theorem 1-2-5 to the initial-value problem
dy
dt
= f (t, y), y(rl) = n (or, respectively, ff(r2)
the solution 45(t) can be extended to the left of rl (or, respectively, to the right of
r2). This is a contradiction. 0
The following example illustrates how to use Lemma 1-3-1.
Problem 1-3-3. Show that the solution of the initial-value problem
d2 y
2
- 2xyL = y3 - y2 y(xo) = 17, Y (xo) = (,
exists at least on the interval 0 x < xo if xo > 0, q > 0, and (is any real number.
Answer.
Assume that the solution y = 4(x) of the given initial-value problem exists on
an interval I = {x : £ < x < xo} for some positive number . Observe that
on Z. Hence,
(O'(x))2
+ 14(x)3
3
+ 14(x)-2
2
< 1(0'(xo))2
2
+ 14(x0)3
3
+ 14(x0)-2
2
(= Al > 0)
on Z. Therefore, we have
Corollary 1-3-4. Assume that f (t, yam) is continuous fort I < t < t2 and all y E R".
Assume also that a function fi(t) satisfies the following conditions:
(a) and d+' are continuous in a subinterval I of the interval tl < t < t2,
(b) (t) = f (t, ¢(t)) in Z.
Then, either
(i) Qi(t) can be extended to the entire interval t1 < t < t2 as a solution of the
differential equation = f (t, y), or
j
(ii) limO(t) = oo for some r in the interval ti < r < t2.
t-r
Using Corollary I-3-4, we obtain the following important result concerning a
linear nonhomogeneous differential equation
dt = A(t)yy + bb(t),
where the entries of the n x n matrix A(t) and the entries of the R"-valued function
b(t) are continuous in an open interval I = It : ti < t < t2}.
Theorem 1-3-5. Every solution of differential equation (1.3.1) which is defined in
a subinterval of the interval I can be extended uniquely to the entire interval I as
a solution of (1.3.1).
Proof
Suppose that a solution y = d(t) of (1.3.1) exists in a subinterval Z' _ It : r1 <
t < r2} of the interval I such that tj < 71 < r2 < t2. Then,
we obtain
in T.
o
is derived by using Lemma 1-1-5. Hence, case (ii) of Corollary 1-3-4 is eliminated.
Since the right-hand side of (1.3.1) satisfies the Lipschitz condition, the extension
of the solution y5(t) is unique.
For nonlinear ordinary differential equations, we cannot expect to obtain the
same result as in Theorem 1-3-5. The following example shows the local blowup of
solutions of the differential equation of Problem 1-3-3.
18 I. FUNDAMENTAL THEOREMS OF ODES
Problem 1-3-6. Show that for any fixed t > 0, there exists a real-valued function
0(x) such that
(1) 0(x) is continuous on an interval 0 < C-x <- 6 for some small positive number
6 which depends on (,
(2) fix) is continuous on an interval 0 < - z < 6,
(3) the function
1 + (: - x)4(x)
y(x) -
F( - x)
satisfies the differential equation
y2
(Eq) dxy - 2xyL y3
ono<x<6.
Answer.
Step 1. Ifwesett=£-randy = , the given differential equation is changed
to t
(1) t2u" - 2tu' + 2u + 9(tu' - u)u = 2t(tu' - u)u + tsu-3 - tut,
where f =
dt
Step2. If wesetu= 1+u'1 and tu' differential equation (1) becomes
=w2,
(2)
_ - 2(1 + wl) + 3w2 - 2(w2 - (1 + wl))(1 + w1)
+ 2r1t(w2 - (1 +wl))(1 + w1)
+ t6 (1 + w1)-3 - (1 + w1)2.
td
!W--'
= W2,
dt
dw2
(3)
t
=2w1+w2-2w1(w2-wl)
dt
+ 2{-lt(w2 - (I + w1))(1 + w1)
+ t6t4(1 + w1)-3 - tt-1(I + wl)2.
3. SOME GLOBAL PROPERTIES OF SOLUTIONS 19
where
_ wl _ 1 0
w- [w2]' ['2 1] F(w) -[2w1(w2-wi)J'
with
Two eigenvalues of f2 are -1 and 2, which are distinct. Let us diagonalize SZ by the
transformation
w" = Pv", where P =
A= [ O1
(6) tv = Av + P-' F(PV) + tP-1G(t, Pv), } .
Set ii= ti. Then, t9 = ti+t2z and F(Pv-) = t2F(Pz). Therefore, (6) becomes
where Ao =
[_2 0I and f (t, z) = tP-'F(Pz) + P-'C(t, tPa). If t > 0 and
if tl1 and tll are small, there exist two positive constants Ko and L such that
If(t, z-) <- tLli - {I and If(t,z)1 <- Ko + tL17.
Step 4. We change (7) to the following system of integral equations:
where to 1is a sufficiently small positive number and c is an arbitrary constant, and
z' = (21 and f = 2
J.
We want to construct a bounded solution of (8) on
J
20 I. FUNDAMENTAL THEOREMS OF ODES
0 < t < to. To do this, first note that, if IZ(t)I < K on 0 < t < to, we obtain
t
Izt(t)I < t-2 fo rI f1(r, zlr))Idr < t-2 t r(Ko + rLK)dr
0
to
< Icit + t
f t
to
r-2(Ko + rLK )dr
t
For a positive number co, fix another positive number K so that co + Ko < 2 and
then fix t0 > 0 so that to < 1, toL < 1, and toKIPI < 1. Here, IPI = max{IP.kl
1 < j, k < 2}, where P = (Pjkj? k=1. Define successive approximations by
t_2Jo
i (t, C) = m > 0-
d +t JJJ( r-2f2(r,-im(r,c))dr
to
It can be shown that the successive approximations converges uniformly for Icl < Co
and 0 < t < to. This establishes the existence of a bounded solution of the integral
equation (8). Thus, we can complete the construction of solution (Sol) of equation
(Eq).
Theorem 1-4-1. Assume that each entry of a Cn-valued function f (z, yam) is given
by a power series in (z, yl, y2, ... , yn) which converges in a polydisk:
where z, yl, y2, , y" are complex quantities. Assume also that there exist positive
numbers M and L such that
If(z,-)I < /M in D,
l 0z, 91) - 1(z, g2)1 5 Lull - y21 whenever (z,y,) E D (j = 1,2).
Define successive approximations by
1m0(z) = 4o,
where the path of integration may be taken to 1be the line segment zaz joining zo to
z in the complex z-plane. Set a = min I a, }. Then,
M
(i) For every m, the entries of the C"-valued))) function 4m(z) are power series in
z - zo which are convergent in the open disk A = {z : iz - zoj < a},
(ii) the sequence { ,,,(z) m = 0,1, ... } satisfies the estimates
!5
m
II ,c
,4n+l(Z) - om(Z)I IZ - zojm+l on 0, m = 0, 1, ... ,
(nl + 1),
(iii) the sequence {¢m(z) m = 0, 1,... } converges to
uniformly in 0 as m -+ +oo,
(iv) the entries of the C"-valued function b(z) are power series in z - zo which
are convergent in 0,
(v) the function Q(z) is the unique solution to the initial-value problem
d = f(z,y), y(zo)
The proof of this theorem is left to the reader as an exercise. The reader must notice
that even if a sequence of analytic functions fn(t) of a real variable t converges to a
function f (t) uniformly on an interval Z, the limit function f (t) may not be analytic
at all. For example, any continuous function can be uniformly approximated by
polynomials on a bounded closed interval. In order to derive analyticity of f (t),
uniform convergence must be proved on a domain in the complex t-plane.
Observation 1-4-2. The estimates
MLm ZOlm+1
m+1(Z) - & (Z) 1 :5 (m+ 1)I IZ - in A, n1 = 0, 1,...
22 I. FUNDAMENTAL THEOREMS OF ODES
+00
imply that there exists a power series Eck(z - zo)k such that
k=0
M
&(z) = E ck(z - z,)' +O(Iz - zol"'+1), m=0,1,2,....
k=O
00
we obtain (z) - ¢,,,(z) = O(Iz - zol'"+1) and, hence, (z) _ >2Ck(Z - ZO)k.
k=0
Observation 1-4-4. Set
m 00
1Cm,kI
a (k = m + 1, m + 2,...).
Note that
1 0(z) - ca
4.,k = 2?ri Izo1=P (z - zo) k+1
for any positive number p < a. This, in turn, implies that
,n oo
Ck(Z-ZO)kl :5 b IZ aZOlk
k=O k=m+1
=blzzD 1
I
a
1-IZ
ZIDI
in the disk A.
Example 1-4-5. In the case of the initial-value problem dzy = (1 - z)y2, y(0) = 2,
we obtain
2
O(z) = 2 E(k + 1)zk =
(1 - Z)2
k=O
EXERCISES I 23
Oo(z) = 2 ,
01(z) = 2 + 4z - 2z2,
83 23
02(x) = 2 + 4z + 6x2 - - 6x4 + 4x5
2x4 2825 82zs
¢3(z) = 2 + 4z + 6z2 + 8x3 _
3 3 9
256z7 124z8 68z9 46z10 56z"
+ 63 + 9 9 9 + 9
4z13 2z14
22z12
9 + 9 - 63
Combining Theorem 1-3-5 and Theorem 1-4-1, we obtain the following theorem:
Theorem I-4-6. If all the entries of an n x n matrix A(t) and C"-valued function
b(t) are analytic on an interval Z, then every solution of linear differential equation
(1.3.1) is analytic on Z.
EXERCISES I
The reader must consider various cases concerning the initial data (r, i , m)1).
1
I-2 Show that the function f (x y) _ satisfies the
' (3 - (x - 1)2)(9 - (y - 5)2)
Lipschitz condition
If(x,yl)-f(x,y2)I < IY1Y2I
if
Ix-1I 5 f, Iyi-5I < 2, lye-51 < 2.
1-3. Show that the initial-value problem
dy _ x3 + 3x + 1
sm y(5) = 3,
(P)
dx - 101-y2)
has one and only one solution on the interval Ix - 51 < 7.
24 I. FUNDAMENTAL THEOREMS OF ODES
1-4. Suppose that u(x) is continuous and satisfies the integral equation
x
u(x) = J sin(u(t))u(t)pdt
0
on the interval 0 < x < 1. Show that u(x) = 0 on this interval if p > 0. What
would happen if p < 0?
u -p
Hint. In case p < 0, the problem is reduced to the initial-value problem du =
sin(u)
dx, u(O) = 0. Thus, the solution u(x) is not identically zero and has a branch point
at x = 0.
1-5. Show that if real-valued continuous functions f (x), g(x), and h(x) satisfy the
inequalities
f(x) ? 0, g(x) < h(x) + f
J0*'C
1-7. Let f (t, y) be a real-valued function of two independent variables t and y which
is continuous on a region:
D = {(t, y) : 91(t) 5 y <- 92(t), r1 < t < r2}.
Suppose that
(i) 91, g'1, 92, and g2 are continuous on rl < t < T2,
(ii) 91(t) < 92(t) on rl < t < T2,
(iii)
gi(t) < f(t,9i(t)), rl < t < rz,
f 9t> f(t, 92(t),
z
(IP)
d = f (t, x), X(r) _ .
where (r, {) is a given point in R x B. Assume that f satisfies the following condi-
tions:
(i) f(t,x)iscontinuous onaregion R={(t,x)ERxB:It-rl <a, IIx-t:ll<b},
where a and b are positive numbers and II II is the norm of B,
(ii) f(t.x) also satisfies the Lipschitz condition IIf(t,xl) -f(t,x2)II 5 LIIX1 -X211
whenever (t, xl) E R and (t, x2) E R, where L is a positive constant.
Show that
(1) there exists a positive number Af such that Ilf(t,x)II < M for (t,x) E R,
(2) problem (IP) has one and only one solution on the interval It - rI < or, where
b \
a= min a, I.
b
Hint. See [Huk7].
26 1. FUNDAMENTAL THEOREMS OF ODES
I-10. Let B be a Banach space over the field R of real numbers and let C be a
nonempty compact subset of B. Also, on a bounded interval Z, let F be an infinite
and equicontinuous set of C-valued functions. Show that F contains an infinite
sequence which is uniformly convergent on Z.
I-11. Let B be a Banach space over the field R of real numbers and let C be
a nonempty compact and convex subset of B. Assume that a C-valued function
f (t, x) is continuous on a region
R={(t,x)ERxB:It-rl <-a,llx-t1l <-b},
where r E R, t E B, a > 0, b > 0, and 11 11 is the norm of B. Show that
(1) there exists a positive number M such that IIf(t,z)115 M for (t,x) E R,
(2) the initial-value problem - = f (t, x), x(r) has a solution on the interval
Hint. Use a method similar to that of Remark 1-2-7. See, also, [Huk7l.
1-12. Let f (t, x") be Re-valued and continuous for (t, x") E Rr+t. Assume also that
E is a nonempty closed set in R"+t. Show that in order that for every (r,)) E E,
there exists a real number a(r, depending on (r, ) such that r < a(-r, {) and
that the intial-value problem
di = f (t, xF), x(r) _ has a solution s = ¢(t, r,
satisfying the condition (t, ¢(t, r, ) E E on an interval r < t < o(r), it is necessary
and sufficient that for every (r, f E E and every positive number e, there exists a
point (a, S) E E such that
< AI e +BluF+C
I
on D, where B and C are positive numbers. Set M = max(Iu(x,0)(: a < r < b}.
Show that
u(x, y) 1 5 Me 11 + (eBV - 1)
on D. B
Hint. The last inequality is the Haar inequality. (See [Haa1, [Na5, pp. 51-561, and
[Har2, pp. 140-141]).
1-15. Let H(u, t, x, q) be a function defined on an open set in R4 containing the
point (u, t, x, q) = (0, 0, 0, 0) and satisfy a uniform Lipschitz condition with respect
to (u, q). Let O(x) be a function of class C' satisfying 0(0) = 0 and 0=(0) = 0.
Show that the initial-value problem
DEPENDENCE ON DATA
The right-hand side of (E) is regarded as a function of (t, to, ca, f, yj. In this chap-
ter, we explain some basic properties of solutions with respect to (t, to, ca, f). A
parameter a is used to represent the variable f. In §11-1, we explain the continuity
of the solution of (P) with respect to (t, to, c"o, e) at (t, r, e, co) when the solution of
(P) is unique for the initial data (r , c', ea) (Theorem 11-1-2). Theorem 11-1-2 was
first proved by I. Bendixson [Benl] for the scalar equation and later by G. Peano
(Pea2] and E. Lindelof [Lind2] for a system of equations. In §11-2, we explain the
differentiability with respect to initial data (to,ca) (Theorem 11-2-1) and with re-
spect to a parameter a (Theorem 11-2-2). The discussion of these topics can be
found in [CL, pp. 22-28, 57-60] and [Har2, pp. 94-100].
28
1. CONTINUITY WRT INITITIAL DATA AND PARAMETERS 29
N N
= U,&(r)), Uo = nU(r)), M = 1+Mo.
3=1 J=1
Then, all the requirements of Lemma II-1-1 are satisfied by 0, Uo, and M. 0
The main purpose of this section is to prove the following theorem under As-
sumption 1.
Theorem II-1-2. Let E be an open neighborhood of to and let Z = {t : s1 < t < s2}
be a subinterval of Zo. Assume that
Since
we obtain
I
-
I!(t(E),E)I+MIt-t(E)I onIx(£fU) and Ii(t,E) AT, E)I
MI t - rI if (t, c) and (r, E) E Z x (£ n U), where M is the positive constant which
was determined by Lemma II-1-1. Therefore, the set F _ E) : E E £ n U} is
bounded and equicontinuous on the interval Z.
Let us derive a contradiction from the assumption that ,JJ(t, e) does not con-
verge to fi(t) uniformly on the interval I as e -+ co. This assumption means that
max I does not tend to zero as e --+ co; i.e., there exists a positive
number p and a sequence (E. : j = 1,2,. .. } such that
Ej E£nU and lim E; = Eo,
(II.1.3)
m Zx 1 (t,Ej) - v(t)I > p.
Here, use was made of the assumption that the topology of the E-space is separable.
Observe that the sequence { (-, j) : j = 1, 2.... } is a subset of Y. Hence,
this sequence is bounded and equicontinuous on Z. Therefore, b y Lemma 1-2-
3 (Arzela-Ascoli), there exists a subsequence v = 1, 2, ... } such that
j., -+ +oo as v - +oo and that tp(t, Ej converges uniformly on I as v -+ +oo.
Set l(t) = lim r%(t, on Z. Since (11.1.2) holds for c = Ef for all v, we obtain
Hence, f (t, J(t), Eo) on I and 1J(to) = ¢(to). Now, condition (iii) of As-
sumption 1 implies that j(t) = fi(t) identically on the interval Z. This, in turn,
implies that lim max I (t, E j ) - $(t) I = 0. This contradicts (1I.1.3).
v-+oo LET
Step 2. We shall prove that there exists an open neighborhood U of co such that
UClloand (t,1i(t,E))EAon Ix(EnU).
Choose no > 0 so that the rectangular region
Ro(r) = {(t.: It - rl < ao, (y - (r)I <-Mao}
is contained in A for every r E I (cf. Figure 1).
FIGURE 1.
1. CONTINUITY WRT INITITIAL DATA AND PARAMETERS 31
has on the region lo x (E f1 Ll) a solution y = J(t, e) such that (t, 111(t, E), E) E Do.
Theorem II-1-4 can be proved by using an argument similar to Step 2 of the proof
of Theorem 11-1-2 and the local existence theorem (cf. Theorem 1-2-5). Details are
left to the reader as an exercise.
11-2. Differentiability
In this section, we explain the differentiability of solutions of an initial-value
problem with respect to the initial data under the following assumption.
Assumption 2. For the initial-value problem (P), assume that
(i) f (t, y-) and Lf' (t, y-) (j = 1, 2, ... , n) are continuous in (t, y-) on an open set
A in the (t, yj-space,
(ii) Q'(t) = f (t, and (t, ¢(t)) E ., on an interval To = {t : tl < t < t2}.
Using Lemma 1-1-6, Theorems 1-1-4, 11-1-2, II-1-4, and Remark 11-1-3, we can
show that there exists a positive number p such that the initial-value problem
has a unique solution y' = O(t, r, q') on the interval lo if r E Zo and fr) - ¢(r)j <
p. The solution y = m(t, r, ri") is continuous in (t, r, q-) and satisfies the condition
(t. (t, r, E 0 if t E Zo, r E lo, and (r)I < p. The solution t(t, r, III
satisfies the integral equation
- 1('r, q) +
e) +
8f
f t (s,( s, r, ))
I" &R1
where q, is the j-th entry of i), eJ is the vector in IR" with entries 0 except for I at
the j-th entry, and Lf is the n x n matrix whose j-th column is . From this
0Yj
speculation, we obtain the following result.
2. DIFFERENTIABILITY 33
Furthermore,
8r
respectively 0is the unique solution of the initial-value
BRA
problem
=
(t,(t, r, n)) z,
(11.2.3)
Proof.
We prove the existence and continuity of by showing that they are the unique
solution of initial-value problem (11.2.3) with the initial-value z-(-r) f(7-, r'). The
existence and continuity of
L can be proved similarly.
d7),
Let +;1(t, r, ij) be the unique solution of (11.2.3) with the initial condition z(r)
- f (r, r)). From Theorems 1-3-5 and 11-1-2 and Remark 11-1-3, it follows that
1i(t, r, n) exists and is continuous in (t, T,i) on the closure D of domain (11.2.2).
Furthermore,
(t,r+h,ff) = ri + f t f(s,G(s,r+h,i))ds,
- rh
f
(11.2.5)
d(t, r, i) = n + f (s, (s, r, i))ds,
r
from (11.2.1) on D assuming that h 54 0 is sufficiently small. Using (11.2.4) and
(H.2-5), we compute
as follows.
34 II. DEPENDENCE ON DATA
T j) = I rt I
l`
f (s, $(s, r + h, AS- J(s, 7,70)] ds
- 17,+h
f ( s, $6, r + h,
if h 96 0 is sufficiently small.
Observation 2. Using an idea similar to the proof of Lemma 1-1-6, for a sufficiently
small h, we obtain
f(s,j(s,r+h,n)) -
1 L(s,0 (s,r+h,fl)+(1 -9)¢(s,r,n1)dOl (t,r,rr))
[10
f o r s E 7 o and a fixed (r, i f ) such that (s, r, n E D. Note that (s, O (s, r + h, r') +
(1- 0)4(s, r, n ) ) E A for s E 10 and a f i x e d (r, n) such that (s, r, n') E D, if h 0 0
is sufficiently small.
Finally, applying Observations I and 2 to (11.2.4), (11.2.5), and (I1.2.6), we obtain
g(t, r, h)
1 rr+h
_- h rr
Jl f(r,qj
r
+JtIJ1
(11.2.7)
-p.:+
i9f
Observation 3. The first and the third terms of the right-hand side of (11.2.7)
8
tend to zero as h - 0. Also, since 4r, r, J-7) = n and (s, d(s, r, rte} is bounded
09
on D, there exist a positive constant L and a non-negative function K(h) such that
limh_o K(h) = 0 and that
1g'(t,r,if,h){ <K(h) + LI
I J1g(s,r,il,h)Ids
t for tEZo.
Thus, we conclude that g(t, r, il, h) - 0 ash -+ 0, i.e., (t, T, t) = rfi(t, r,'.
Upon applying Theorem 11-2-1 to the initial-value problem
f (t, du
d= y, u), = 0, y{ r) = i1, u(r) = e,
1, 2, ... , n), and 8e (t, y', e) are continuous on an open set Do in the (t, y, e)-space.
Let y = y(t, T, i), e) be the unique solution of the initial-value problem
dy
dt = f (t, y, e),
y(r) _ *!
Then, there exists an open set 11 in the (t, T, 77, e)-space such that the function
00 00
(t, T, ii, e) and are continuous on fl. Furthermore, i = (t, r, 4J, e) is the
unique solution of the initial-value problem
di Of
Z(r) = 0.
dt = zi (t, (t, r, 1/, e), e)l + 8E (t, 4(t, r, n, E). E),
EXERCISES II
II-1. Given an interval I = {x : a < x < b}, show that if f ei is sufficiently small,
the initial-value problem
dy
e cos(x(y - x)), y(a) = a + e
dx
has the unique solution y = O(x, e) on Z. Show also that Ui m O(x, e) = x uniformly
on Z.
Hint. The function cos(x(y - x)) is continuously differentiable on the entire (x, y)-
plane. Furthermore, I oos(x(y - x))I < 1. This shows the existence of the unique
solution. Note that y = x is the unique solution in case e = 0. Complete the proof
by using Theorem 11-1-2.
36 II. DEPENDENCE ON DATA
11-2. Let y = 4 1(x, A) and y =¢2(x,.1) be tvm solutions of the differential equation
2 + A(1 + x2)y = 0
determined respectively by the initial conditions
Show that
(i) 01 and 02 are analytic in (x, A) everywhere in the complex (x, A)-space (i.e.,
in C2),
(ii) aA (1, A0) # 0, if 02(1, 1\0) = 0.
Hint. Theorem II-2-2 implies that z (x, A) is the solution of the initial-value
problem
8 (x, A)
01(X"\)
J0 o
Hence, we obtain
11-4. Let Ax, y-) be an RI-valued function whose entries are continuously differ-
entiable with respect to (x, yam) in a domain D C IIPn+1 Also, let y = (x, 171 be a
solution of the system fy = f (x, y) such that p(0) and that (x, 5(x, rtj) E D
for 0 < x < 1 and it E Ao, where Do is a domain in R. Note that we must have
(0, rl E D for rj E Ao. Set A = {(1, 1 : if E Do}. Show that
(a) Al is open in lR',
(b) the mapping 4(1,r) : Ao - Al is one-to-one, onto, and differentiable with
respect to i in Do,
(c) if we denote the inverse of the mapping ¢(1,i) by AI -. Ao, then
is also differentiable with respect to S.
11-5. For the differential equation
find
(i) the unique solution y = O(x, e, ri) satisfying the initial-condition y(£) = q;
(ii) the partial derivative 84(x, , n) ,
of
am(x, £, r7)
(iii) the artial derivative
8 t , n) at x =
(iv) 190(x,
(I+ x2) dz
dz
- 2xz = 4xO(x, {, q)z .
II-6. Assume that f (t, x1, x2i ... , xn) is a real-valued, continuous, and continu-
ously differentiable function of (t, xl,... , x,) on an open set A in the (t, x1.... , xn)-
space. Assume also that o(t) is a real-valued solution of the n-th order differential
,0On-1)(t))
equation x(") = f(t,x,x',... ,x(n-1)) and (t,Oo(t),e0(t),... E A on an
interval Zo = It : t1 < t < t2}. Show that there exists a positive number p such
that
(i) the initial-value problem
x(n) f (t, x, x , ... , x(n-1) ), x(r) = SI, x (r) = 52, ... , x(n-1) (r) = n
=
has a unique solution x = {n) on the interval 20 if r E Zo and
t 3-O0-1)(r)I<p (j=1,2,...,n);
isfies the condition t t 'r, 6, )
(ii) the solution x = qi(t, r, ti, ... , {n) is continuous in (t, r, t1, ... , n) and sat-
W) e(t r O("-1)(t r
(j=1,2,...,n);
38 II. DEPENDENCE ON DATA
and if 8 &2 (t, A) is continuous on the region A = {(t, A) : 0 < t < 1, a <
A < b}, then d(t, A) is identically equal to zero on A, where a and b are real
numbers.
(2) Does the same conclusion hold if 0(t,,\) is merely continuous on A?
11-9. Let a(x, y) and b(x, y) be two continuously differentiable functions of two
variables (x, y) in a domain Do = {(x, y) : IxI < a:, IyI < Q} and let F(x, y, z)
be a continuously differentiable function of three variables (x, y, z) in a domain
Do = {(x, y, z) : Ixl < a:, IyI < 3, Izl < 7}. Also, let x = f (t, i7), y = 9(t, ij), and
z = h(t,17) be the unique solution of the initial-value problem
dx OF
- (x+ y, z, p, q),
dy _ OF
flq(T, y, z, p, q),
dt
dz OF OF
dt = pij (x,y,z,P,q) + gaq(x,y,z,P,q),
dp OF OF
dt 8x (T, y, z, P, q) - P 8z (x, y, z, p, q),
dq OF OF
(x, y, z, P, q) - q az (x, y, z. P, q)
dt 8y
where x0 (s), yo(s), zo(s), po(s), and qo(s) are differentiable functions of s on R such
that
dzo(s)=Po(s)- (s)+go(s)dyo(s)
F(xo(s),yo(s),zo(s),Po(s),go(s))=0, ds ds ds
on R. Show that
F(x(t, s), y(t, s), z(t, s), p(t, s), q(t, s)) = 0,
at (t, s) = At, s) 5 (t, s) + q(t, s) (t, s),
flz(t,s)
fls = P(t,s)L(t,s)
as
+ q(t,s)ay(t,8)
as
Oz Z 8z\ =0
+ H t, x, y,-,
dp 8H (LL dq _ _ 8H
dt 8x ' dt W,
dz 8H 8H du OH
= u + p-p + q8q, _
= --6T.
dt
CHAPTER III
NONUNIQUENESS
dy
dt = f(t,y1,
(P)
111-1. Examples
In this section, four examples are given to illustrate the nonuniqueness of so-
lutions of initial-value problems. As already known, problem (P) has the unique
solution if f'(t, y) satisfies a Lipschitz condition (cf. Theorem I-1-4). Therefore, in
order to create nonuniqueness, f (t, y-) must be chosen so that the Lipschitz condi-
tion is not satisfied.
Example III-1-1. The initial-value problem
dy
(III.1.1) = yt/3 y(to) = 0
dt
has at least three solutions
3/2
(S.1.2) y(t) _ [3 t - to) ]I
, t > to,
0, t < to,
and
3/2
[3(t - to),
2
(S.1.3) y(t) t > to,
0, t < to
41
42 111. NONUNIQUENESS
(cf. Figure 1). Actually, the region bounded by two solution curves (S.1.2) and
(S.1.3) is covered by solution curves of problem (I11.1.1). Note that, in this case,
solution (S.1.1) is the unique solution of problem (P) for t < to. Solutions are not
unique only fort ? to.
Example 111-1-2. Consider a curve defined by
and translate (111.1.2) along a straight line of slope 1. In other words, consider a
family of curves
(111.1.3) y = sin(t - c) + c,
dy
(II1.1.4)
dt = 1OS[G(y-t)-(y-t)1.
Since G(2n7r) = 0 and cos(-2n7r) = 1 for every integer n, differential equation
(111. 1.4) has singular solutions y = t + 2mr, where n is an arbitrary integer. These
lines are envelopes of family (111.1.3) (cf. Figure 2).
FIGURE 1. FIGURE 2.
Example 111-1-3. The initial-value problem
(111.1.5) y(to) = 0
dt =
and
t,0)2,
4 (t - t > to ,
(S.3.2) y(t)
- 4(t - to)2, t < to
(cf. Figure 3). The region bounded by two solution curves (S.3.1) and (S.3.2) is
covered by solution curves of problem (111.1.5).
FIGURE 3.
dy
= + E, y(to) = 0
dy _ y2
y2 + C2 lyl y(to) = 0,
where a is a real positive parameter. Each of these two differential equations satisfies
the Lipschitz condition. In particular, the unique solution of problem (111. 1.6) is
given by
(cf. Figure 4). On the other hand, (S.3.1) is the unique solution of problem (111.1.7).
Figure 5 shows shapes of solution curves of differential equation (111.1.7). Note that
nontrivial solution of (111.1.7) is an increasing function of t, but it does not reach
y = 0 due to the uniqueness.
44 III. NONUNIQUENESS
FIGURE 4. FIGURE 5.
Generally speaking, starting from a differential equation which does not satisfy
any uniqueness condition, we can create two drastically different families of curves
by utilizing two different smooth perturbations. In other words, a differential equa-
tion without uniqueness condition can be regarded as a branch point in the space
of differential equations (cf. [KS]).
Example 111-1-4. The general solution of the differential equation
2
(1II.1.8)
d) +y2=1
is given by
(III.1.10) dy = 1 - yz and
dy=- l-y2.
dt dt
Each of these two differential equations satisfies the Lipschitz condition for lyI <
1. Figures 6-A and 6-B show solution curves of these two differential equations,
respectively.
y=I
y=-I
FIGURE 6. FIGURE 6-A. FIGURE 6-B.
Observe that each of these two pictures gives only a partial information of the
complete picture (Figure 6).
2. THE KNESER THEOREM 45
FIGURE 7.
In a case such as this example, a differential equation on a manifold would give
a better explanation. To study a differential equation on a manifold, we generally
use a covering of the manifold by open sets. We first study the differential equation
on each open set (locally). Putting those local informations together, we obtain a
global result. Each of Figures 6-A and 6-B is a local picture. If these two pictures
are put together, the complete picture (Figure 6) is obtained.
(III.2.1) dt = f(t,y-)
under the assumption that the R"-valued function f is continuous and bounded on
a region
(111.2.2) S2 = {(t, y-) : a:5 t < b , Iy1 < +oo }.
Under this assumption, every solution of differential equation (III.2.1) exists on the
interval Zfl = {t : a < t < b} if (to, y"(to)) E f2 for some to E Zo (cf. Theorem 1-3-2
and Corollary I-3-4). The main concern in this section is to investigate topologi-
cal properties of a set which is covered by solution curves of differential equation
(111.2.1).
2. THE KNESER THEOREM 47
Hence, (c) E $.(A) but * (c) 4 F1 U F2. This is a contradiction (cf. Figure 10).
48 III. NONUNIQUENESS
!=c t=c
(III.3.1) dt =
under the assumption that the entries of the R°-valued function fare continuous
and bounded on a region
(111.3.2) n = {(t, y-) : a < t < b , lyi < +oc}.
50 M. NONUNIQUENESS
FIGURE 13.
Choose a sequence {ok : k = 1, 2, ... } of subdivisions of the interval cl < t < c2
such that
Ok : C1 = Tk,O < Tk.1 < ... < Tk,2k-I < Tk,2" = C2,
f e
1 Tk,t = C1+2k(c2-cl), e=0,1,...,2k, k=1,2,....
Remark 111-3-3. In general, the conclusion of Theorem 111-3-2 does not hold
on an interval larger than cl < t < c2.
where t and y are real variables, and assume that f is real-valued, bounded, and
continuous on a region H = {(t,y) : a < t < b, -oo < y < +oo}. Choose c in
the interval I = (t : a < t < b}. Then, Sr ({ (c, r,) }) is compact and connected
for every r E I and every real number r) (cf. Theorem III-2-4). This implies
that there exist two numbers 41(r) and ¢2(r) such that ST({(c,17)}) = { y E
R : bl(r) < y < &2(r)}. Hence, R({(c,n)}) = {(t,y) E R2 : ¢1(t) < y <
02(t), a < t < b} (cf. Figure 14). The two boundary curves (t,4i(t)) and (t,02(t))
of R({(c,rl)}) are solution curves of differential equation (111.4.1) (cf. Theorem
111-3-2). Every solution ¢(t) of (111.4.1) such that 0(c) = n satisfies the inequalities
01(t) < ¢(t) < ¢2(t) on Z. The solution 02(t) (respectively 01(t)) is called the
maximal (respectively minimal) solution of the initial-value problem
dy
(111.4.2)
dt = f (t, y), y(c) = 77
on the interval Z. In this section, we explain the basic properties of the maximal
and minimal solutions. Before we define the maximal and minimal solutions more
precisely, let us make some observations.
FIGURE 14.
Proof.
For a fixed to E To, we prove that
If ¢1(to) > -02(to), then there exists a positive number 6 such that 4(t) = 01(t) for
It - tol < J. Therefore, (111.4.3) holds. Similarly, (111.4.3) holds if 02(to) > &(to).
Hence, we consider only the case when '(to) _ .01(to) = 02(to). In this case,
0(t) - 0(to) = 0'(t) - 0'(to),
f o r each fixed t E To, we have where j = 1 or
t - to t- to
j = 2, depending on t. Hence, by the Mean Value Theorem on O,(t), there exists
r E To such that r --r to as t - to and 0(t) - ¢(to) = f (r, 0,(r)). Also, 03 (r) _
t - to
¢j(to) + f (a,-Oj (a)) (7 - to) for some a E To such that a to as r - to. Since f is
bounded on the two curves (t, 01(t)) and (t, 02(t)), (111.4.3) follows immediately. 0
Observation 111-4-2. Let f (t, y), D, and To be the same as in Observation
111-4-1. Consider a set F of solutions of differential equation (111.4.1) such that
(t, 0(t)) E D on To for every ¢ E F. Assuming that there exists a real number K
such that 0(t) < K on lo for every 0 E F, set 00 (t) = sup{Q(t) : ¢ E F} for t E To.
Assume also that (t, 00(t)) e D on To. Then, 4'o(t) is a solution of differential
equation (111.4.1) on To.
Proof.
As in Observation 111-4-1, we prove that
for each fixed to E To. Choose three positive numbers po, p, and S so that
(i) A = {(t, y) : t E To, Iy - 4o(t)I < po} C D,
(ii) we have (t, 4(t)) E A on the interval t - to I < 6, if d(t) is a solution of
I
(III.4.1) such that 0 < 0o(r) - 4i(r) < p for some r in the interval It - tol < b.
There exists a positive number Af such that I f (t, y) I < M on A.
Let us fix a point r on the interval It - tol < b. First, we prove the existence of
a solution ty(t; r) of (III.4.1) such that
(111.4.5) P(r; r) = 0o (r) and yJ(t; r) < 4o(t) for it - tol < b.
To do this, select a sequence {4'k : k = 1, 2,... } from the family F so that
lirao0thk(r) = 40(r). We may assume that (t, hk(t)) E A on It - tol < S for
Y+
k (cf. (ii) above). Then, the sequence {4'k : k = 1, 2, ... )is bounded and
equicontinuous on It - tot < J. Hence, we may assume that lim 4'k(t) = 0(t; r)
k-»+oo
exists uniformly on the interval It - tol < 6. It is easy to show that V'(t; r) is a
solution of (111.4.1) and that (111.4.5) is satisfied.
Set ty(t) = max{t/,(t; r), tJ'(t; to)} for Jt - tol < b. Then, v is a solution of (111.4.1)
such that (1) V)(r) = oo(r) and ty(to) = 00(to), (2) iP(t) < 4'o(t) for It - tol < b,
and (3) (t, v/ (t)) E A for It - tol < 6 (cf. Figure I5).
54 III. NONUNIQUENESS
Y= #o(')
Y s V(1)
FIGURE 15.
O0(T) - ¢040) = t.ti(T) - 0(t0) = f(a,tp(a))
Using this solution tp(t), we obtain
T - to T- to
for some a such that Ja - tol < 6 and that or to as r -" to. Since 10(a) - tt'(td)J <
Mba - t01, (111.4.4) can be derived immediately. 0
Let us define the maximal (respectively minimal) solution of an initial-value
problem
where f is real-valued and continuous on a domain V in the (t, y)-plane and the
initial point (T, 1:) is fixed in D.
Definition 111-4-3. A solution tp(t) of problem (111.4.6-() is called the maximal
(respectively minimal) solution of problem (111.4.6-() on an interval I = it : T <
t < r' } if
(a) tp(t) is defined on I and (t, {1(t)) E D on Z,
(b) if 4(t) is a solution of problem (II14.6-{) on a subinterval r < t r" of Z,
then
0(t) < y(t) (respectively ¢(t) > P(t)) on r < t < r".
The following two theorems are stated in terms of maximal solutions. Similar
results can be stated also in terms of minimal solutions. Such details are left to
the reader as an exercise. In the first of the two theorems, we consider another
initial-value problem
(III.4.6-n) !LY
dt
= f(t,y), y(r) = r)
Proof
Set 0(p) = {(t, y) : t E Z, yt(t) < y < ot(t) + p}. For a sufficiently small
positive number p, we have A(p) C 1), and, hence, if (t, y)I is bounded on A(p) for
a sufficiently small positive number p.
First, we prove that for a given p > 0, there exists a positive number 6 such that,
if f < q < { + b, every solution 0(t) of problem (111.4.6-1?) defined on a subinterval
r < t < r" of I satisfies
(III.4.7) 0(t) < ,'(t)±p for r < t <7".
Otherwise, there exist two sequences {qk : k = 1 , 2, ... ) and {rk : k = 1, 2, ... } of
real numbers such that (1) qk > {, lim qk = t;, and r < rk < r', (2) 0&-(-r) = nk
k -+oo
and Ok(Tk) = v((rk) + p, and (3) Ok(t) < vf(t) + p for r < t < rk. Furthermore,
there exists a real number r(p) such that rk > r(p) > r (cf. Figure 16).
Y=W4+P
- 1 I
Y Wt
I
I i
t=r t= TA 1=',
FIGURE 16.
Set
max(0k(t), '' (t)), T<t<Tk,
f (t)+p
11)k (t )
,
Tk St ST.
It is easy to show that the sequence {0k W : k = 1, 2, ... } is bounded and equicon-
tinuous on the interval Z. Hence, we can assume without any loss of generality
that
(1) tim Tk = rp exists,
k-.+oo
(2) tim 10k(t) = 0(t) exists uniformly on Z.
Note that
(3) d(r) _ and O(ro) _ 'af(ro) + p,
(4) ro?r(p)>r.
It is also easy to show that 0(t) is a solution of (III.4.6-t) on the subinterval
r < t < ro of Z. Since 104 is the maximal solution of (III.4.6-1:) on 7, we must
have 0(t) < af(t) on the interval r < t < To. This is a contradiction, since
0(ro) = i (ro) + p. Thus, (111.4.7) holds.
Now, for a given positive number p, there exists another positive number 6(p)
such that (111.4.7) holds for any solution ¢(t) of problem (III.4.6-17) if < ' <
t; + 6(p). Hence, using max(4(t), 0((t)), we can extend 6(t) on the interval Z in
such a way that
(111.4.8) ?Gf(t) 5 y(t) < p on Z.
56 III. NONUNIQUENESS
Let F be the set of all solutions 6(t) of (III.4.6-71) which satisfy condition (111.4.8).
Set t[i,r(t) = sup{¢(t) : 0 E F j for t E I. Then, ty,, is the maximal solution of
(III.4.6-r7) on I. Furthermore, 104(t) < 1,1(t) < ot(t) + p if t: < rt < t + 6(p).
Letting p - 0, we complete the proof of the theorem.
Assuming again that f is continuous on a domain D in the (t, y)-plane, consider
initial-value problem (III.4.6-{) together with another initial-value problem
where (t, 1;) E V and f is a positive number. We prove the following theorem.
Theorem III-4-5. If the maximal solution tt'(t) of problem (111.4.6-,) on the in-
terval I = {t : r < t < r'} exists, then, for any positive number p, there exists
another positive number e(p) such that for 0 < E < E(p), every solution d(t) of
(11!.4.9-E) exists on I and tp(t) < d(t) < y',(t) + p on the interval I. In particu-
lar, for every sufficiently small positive number f, then exists the maximal solution
tf=, (t) of problem(111.4.9-E) on I and lim tL, = tV uniformly on I.
Proof
The maximal solution tI'(t) satisfies the condition (t,>l,(t)) E D on I (cf. Defini-
tion 111-4-3). Define a function 9(t, y) by
Then, g is continuous and bounded on the domain {t E I, -oo < y < +oo}. Hence,
every solution ¢(t, e) of the initial-value problem
= 9(t, y) + E, y(r) _
We prove that for a given positive number p, there exists another positive number
e(p) such that if 0 < e <_ E(p), we have 0(t, e) < t1(t)+p on I. Otherwise, there exist
a real number r(p) and two sequences {Ek : k = 1, 2.... ) and {rk : k = 1, 2, ... }
of real numbers such that
(1) ek > O and lim Ek = 0,
k +oo
4. MAXIMAL AND MINIMAL SOLUTIONS 57
exists on To and w_ (t) < 0(t) < w+(t) on 10 (i.e. (t, y5(t)) E Q). In particular, the
maximal solution 01(t) and the minimal solution 02(t) of problem (111.4.12) on lo
exist and w_ (t) < 02 (t) < ¢, (t) < w+ (t) on Zo
Proof of this lemma is left to the reader as an exercise (cf. Figure 18).
Y = vKW)
v= #(z)
Y=
and w_ (a) < 4 < w+(a). Then, there exists a solution Q(t) of initial-value problem
(111.4.12) on To such that
Prof.
Let us define a function g(t, y) by
Then, g is bounded and continuous on the region {(t, y) : t E 10, -00 < y < +oo).
Hence, every solution ¢(t) of the initial-value problem
exists on Za.
Set "I+(t,E) = W.}. (t) + E(t - a) and w_(t,e) = w_(t) - e(t - a), where c is a
positive number and t E To. Then,
+(t, E) dW+(t)
dt
+ ( > f(t,w?(t)} + E > g(t,w+(t,e)),
dt
dw_(t,E) _ d -(t)
- E < f(t,W-(t)) - E < g(t,W_(t,E))-
dt dt
Hence, w_ (t, c) < ¢(t) < w+ (t, e) on Zo (cf. Lemma 111-4-6). Letting e 0, we
complete the proof of Theorem 111-4-7.
Comment III-4-8. The maximal and minimal solutions given in this section are
also defined and explained in [CL, pp. 45-48] and f Har2, p. 251.
Assumption 1. Let t and u be two real variables and let g(t,u) be a real-valued
and continuous function of (t, u) on a domain D in the (t, u)-plane. Also let .y(t)
be the maximal solution of the initial-value problem
du
= g(t,u), u(a) = uo
on an interval lo = It : a < t < b}, where (a, uo) E D and (t, i,b(t)) E D on To.
We first prove the basic theorem given below.
Theorem 111-5-1. Assume that
(1) g(t, u) > 0 on D and uo > 0,
(2) an R"-valued function ¢'(t) is continuously differentiable on a subinterval I =
it: a< t <,r) of 4,
(3) uo,
(4) (t, 14(t) I) E V on Z,
(5)
do(t)
< g on Z.
dt
Then,
P(t) on 1.
Proof.
Let a be a positive number and let 0(t, e) be any solution of the initial-value
problem
du
dt = 9(t, u) + e, u(a) = u0.
If e > 0 is sufficiently small, i/i(t, e) exists on Zo and lim 0(t, e) = P(t) uniformly on
To (cf. Theorem III-4-5). Let us make the following observations:
where t > s. Suppose that l (s)I = r'(s, e) for some s E Z. Then, there exists a
positive number b(e) such that
FIGURE 19.
Note that
and
dj(t) a<t<b
< C + MI¢(t)t for and uo,
dt
we can apply Theorem 111-5-1 to fi(t) with g(u) = C + Mu. Note that the initial-
value problem = C + Mu, u(a) = uo has the unique solution
Hence,
du(t) = g(t,u(t)),
(t,u(t)) E A on Z,
dt
(111.6.1)
aim u(t) = 0
d9(t)
(t, y"(t)) E Q on Z,
dt
(111.6.2)
lim 19(01 = 0,
t-.a r(t)
where S1 = {(t, y) : jyj < w(t), t E Z}. The main result of this section is the
following theorem.
Proof
Suppose that problem (III.6.2) has a nontrivial solution j(t) on Z. This means
that (a) ,-a< 6. for some a E Z. Choose a positive number 6 so that f <
min aminbw(t) }.
on the interval a < t < b. Using Theorem 111-2-7, we can construct a nontrivial
solution uo(t) of (111.6.3) on the interval a < t < b so that (t, uo(t)) E A(f3) on
a<t<b(cf. Figure 20).
Observation 2. The nontrivial solution uo(t) of (111.6.3) which was constructed
in Observation 1 can be continued on the interval a < t < a so that
satisfies the condition tfi(t, e) < on any subinterval a < t < a of I if it exists
on this subinterval (cf. Figure 21).
r=a r=b
I I u= 1 ;(r)1
I I u ='Kr' E)
I 1 L -_Ju=P
U TV, El
u= u0(t) I u uo(r)
U=0 u=0
r=b r=a
FIGURE 20. FIGURE 21.
Now, define a real-valued function G(t, u) by
dylt) on 1,
dt
= f(t,b(t)), (t, y'(t)) E Do
(111.6.6)
lun y(t)r 0
t-a ()
has only the solution (t) itself.
Proof.
Set l(t) = z(t) + .(t). Then, problem (111-6.6) becomes
di(t)
dt = F(t, i(t)) = A t ' z"(t) + fi(t)) - At' fi(t)) on Z,
Since I1 (t,y-Q(t))j < g(t, on Do, Theorem 111-6-2 follows from Theorem
II1.6.1. 0
Let us apply Theorem III-6-2 to some initial-value problems.
Example 1 (The Osgood condition (cf. [Os])). Consider a real-valued function
g(t, u) = h(t)p(u) in the case when
(1) h(t) is continuous and h(t) > 0 on an interval 1 = {t : a < t < b},
64 III. NONUNIQUENESS
(}
(5) jo+ p(u) = +oo.
Assume that
(1) an R"-valued function f (t, y) is continuous on a domain V in the (t, yam)-space,
(ii) f satisfies the condition If (t,1%1) - f (t, il2)1 < 9(t, Iii - Y21) = -1721)
whenever t E Z, I171 - g2 l < K, (t, y"1) E D, and (t, y2) E D.
Let ¢1(t) and 42(t) be two solutions of an initial-value problem = f (t, M, y"(a) _
i such that
(a, T11 E D and (t, p1(t)) E D, (t, 2(t)} E D
is given by u(t) = c(t - a)', where c is an arbitrary constant. This implies that
Assumption 1 is satisfied by r(t), w(t), and g(t,u). In particular, choosing A = 1,
we derive the following result due to M. Nagumo (Nal and Na2J. Assume that
(i) f (t, y') is continuous on a domain V in the (t, y-)-space,
191 -921
(ii) f satisfies the condition j f(t, 91) - f (t, 92)1 < _ whenever a < t < b,
III, -Y2I<K,(t,fi)EDand (t,Y2)EV. a dY
Suppose that 1(t) and y+,2(t) are two solutions of the initial-value problem =
dt
f (t, y), g(a) = # such that
dy _ {tsin(f. if t 1 0,
dt 0 if t = 0, y(r) = n
EXERCISES III
0 for t = 0 ,
(b) solutions of (E) with the initial-values v(-1) = 0 and v(-1) = f 1 are not
n
unique;
(c) for any positive constant c,, the differential equation
do(t)
< I40 for t > 0 and o(0) = 0,
dt
dd(t)
dt
s ia(t)12 for 0 < t < 1 and lim
0+
1d(t)I = 0,
9
d9(t)
< 2tlm(t)I for 0 <- t < +oo,
dt i
l 10(t)i eXpft21 = 0,
dy(t) = sin
- y(t) lim y(t) = 0
It1 (-0
has only the trivial solution y(t) = 0.
Hint. Note that
/
yt
-sin(/
1M l = Y'- Y2cos1
/
sin( It1 `t1
!I
j,
t1) ltl
111-8. Let f (t, i, y) be an R"-valued function of (t, x, y-) E R x R" x R'". Assume
that
(1) the entries of f'(t, i, yl are continuous in the region A 0<t<
a, jxrj < a, jyj < b}, where a, a, and b are fixed positive numbers,
(2) there exists a positive number K such that j f (t, x"j, y- )- f (t, x"2, y -)l < Kjxt -x"2 j
if (t,i),y-) E A (j = 1, 2).
Let U denote the set of all R"'-valued functions u(t) such that ju(t)j < b for 0 <
t < a and that JiZ(t) - u(r)j < Lit - rj if 0 _< t < a and 0 < r < a, where L
is a positive constant independent of u E U. Also, let ¢(t; u+) denote the unique
solution of the initial-value problem 'jj u(t)), x(0) = 0', where u" E U. It is
known that there exists a positive number ao such that for all u E U, the solution
(t, u') exists and u)j < a for 0 < t < aa. Denote by R the subset of Rn+'
which is the union of solution curves {(t, 0(t, 65)) : 0 < t < ao) for all u E U, i.e.,
R = {(t, ¢(t, u)) : 0 < t < aa, u E U}. Show that R is a closed set in R"+t
Hint. See [LM1, Theorem 2, pp. 44-47) and [LM2, Problem 6, pp. 282-283).
111-9. Let f (t, 2, yy be an R"-valued function of (t, a, y) E R x R" x R'". Assume
that
(1) the entries of f (t, f, y-) are continuous in the region 0 = {(t, 2, yj : 0 < t <
a, Jx"j < a, jy"j < b}, where a, a, and b are fixed positive numbers,
(2) there exists a positive number K such that yl- f (t, x2, y-)J < Kjit-i2j
if (t,i1,y)EA (j=1,2).
Let U denote the set of all R'"-valued functions g(t) such that I i(t)J < b for 0 <
t < a and that the entries of u are piecewise continuous on the interval 0 <
t < a. Also, let (t; u) denote the unique solution of the initial-value problem
di = f (t, x, u"(t)), x(0) = 0, where u" E U. It is known that there exists a positive
dt
number ao such that for all u' E U, the solution (t, u) exists and Jd(t, u)j < a for
0 <- t < ao. Denote by 1Z the subset of R"' which is the union of solution curves
{(t, ¢(t, ul)) : 0 < t _< ao) for all u E U, i.e., R = {(t, (t, u)) : 0 < t < ao, u E U).
Assume that a point (r, ¢(r, uo)) is on the boundary of R, where 0 < r < 00 and
!!o E U. Show that the solution curve 0 <- t < r} is also on the
boundary of R.
Hint. jLM2, Theorem 3 of Chapter 4 and its remark on pp. 254-257, and Problem
2 on p. 258).
III-10. Let A(t, x) and f (t, x") be respectively an n x n matrix-valued and R"-
valued functions whose entries are continuous and bounded in (t,x-) E R"+' on
a domain 0 = { (t, x) : a < t < b, x" E R"), where a and b are real numbers.
Also, assume that (r, {) E A. Show that every solution of the initial-value problem
dx
= A(t, x'a + f (t, i), i(r) = t exists on the interval a < t < b.
dt
CHAPTER IV
The main topic of this chapter is the structure of solutions of a linear system
row and the k-th column; i.e., A = all a22 a . A matrix A E Mn(C)
and an2 "' ann
69
70 IV. GENERAL THEORY OF LINEAR SYSTEMS
h=1
PA(A) = An + Eph(A)An-h,
A° = In.
h=1
Now, let us prove the Cayley-Hamilton theorem (see, for example, [Be13, pp. 200-
201 and 220], (Cu, p. 220], and (Rab, p. 198]).
Theorem IV-1-5 (A. Cayley-W. R. Hamilton). If A E Mn(C), then its charac-
teristic polynomial satisfies PA(A) = O, where 0 is the zero matrix of appropriate
size.
Remark IV-1-6. The coefficients ph(A) of pA(A) are polynomials in entries ap,
of the matrix A with integer coefficients.
Proof of Theorem IV-1-5.
Since the entries of pA(A) are polynomials of entries a,k of the matrix A, they
are continuous in the entries of A. Therefore, if pA(A) = 0 for A E Sn, it is
also true for every A E Mn(C), since Sn is dense in Mn(C) (cf. Lemma IV-1-4).
Note also that if B = P'1 AP for some P E GL(n, C), then pB(A) = PA(A) and
pB(B) = P-'pA(A)P. Therefore, it suffices to prove Theorem IV-1-5 for diagonal
matrices. Set A = diag(A1, A2, ... , A.J. Then, pA(A) = (A - A1)(A - A2) ... (A - An)
and pA(A) = diag[PA(A1),PA(A2),... ,PA(A.)] = 0.
It is an important application of Theorem IV-1-5 that an n x it matrix N satisfies
the condition N" = 0 if its characteristic polynomial pN(A) is equal to A". If
N" = O, N is said to be nilpotent.
Lemma IV-1-7. A matrix N E Mn(C) is nilpotent if and only if all eigenvalues
of N are zero.
Proof.
If IV is an eigenvector of N associated with an eigenvalue A of N, then Nkp"= Akp'
for every positive integer k. In particular, N"p = A"p Hence, if N" = 0, then
A = 0. On the other hand, if all eigenvalues of N are 0, the characteristic polynomial
pN(A) is equal to A". Hence, N is nilpotent. 0
Applying Lemma IV-1-1 to a nilpotent matrix N, we obtain the following result.
72 IV. GENERAL THEORY OF LINEAR SYSTEMS
i= P2(A}.
J=1
we obtain
k
(IV.1.5) I. _ > Ph(A).
h=1
In the following two lemmas, we show that (IV.1.5) is a resolution of the identity in
terms of projections Ph (A) onto invariant subspaces of A associated with eigenvalues
Ah, respectively.
Lemma IV-1-9. The k matrices P, (A) (j = 1,2,... , k) given by (W-1.4) satisfy
the following conditions:
(i) A and P, (A) (y = 1, 2, ... , k) commute.
1. SOME BASIC RESULTS CONCERNING MATRICES 73
k
(IV.1.6) P,(A) _ >P,(A)Ph(A).
h=1
Then, (v) follows from (IV.1.6) and (iii). To prove (vi), let IT, be an eigenvector of
A associated with the eigenvalue Al. Note that (IV.1.2) implies Ph(A)) = 0 if h 0 j.
Therefore, we derive P,(A,) = 1 from (IV.1.3). Now, since P. (A)#, = P,(A3)p' #
we obtain (vi).
Lemma IV-1-10. Denote by V. the image of the mapping P,(A) : C" -. Cn.
Then,
(1) p'E Cn belongs to V. if and only if P,(A)p= p
(2) Pj(A)p"=0 for all fl E Vh if j 0 h.
(3) Cn = V1 Ei3 V2 e e Vk (a direct sum).
(4) for each j, V, is an invariant subspace of A.
(5) the restriction of A on V, has a coordinates-wise representation:
(A-)'1In)V1,1PJ,2...p'1.n,J = [Pi,1PJ,2...p3i.nsJNj
as the coordinates-wise representation relative to this basis. This implies that
(A - A,1.)'Pi(A)(P1,1Pi,2...p),nj) _ I
and
from Lemmas IV-1-9 and IV-I-10 and (IV.1.10). Hence, S is diagonalizable and N
is nilpotent. Furthermore, NS = SN since S and N are polynomials in A. This
shows the existence of S and N satisfying (a), (b), (c), and (d). Moreover, from
(IV.1.4), it follows that two matrices S and N are polynomials in A with coefficients
in the field Q(ajk, Ah).
Step 2. Uniqueness of S and N. Assume that there exists another pair (S, N)
of n x n matrices satisfying conditions (a), (b), (c), and (d). Then, (c) and (d)
imply that SA = AS and NA = AN. Hence, SS = SS, NS = SN, SN = NS,
and NN = NN since S and N are polynomials in A. This implies that S - S is
diagonalizable and N - N is nilpotent. Therefore, from S - S = N - N, it follows
that S-S=N-N=O.
Step 3. The case when S and N are real. In case when A is real, let 5 and N be
the complex conjugates of S and N, respectively. Then, A = S + N = 3° + N.
Hence, the uniqueness of S and N implies that S = 3 and N = N.
This completes the proof of Theorem IV-1-11.
Definition IV-1-12. The decomposition A = S + N of Theorem IV-1-11 is called
the S-N decomposition of A.
Remark IV-1-13. From (IV.1.11), it follows immediately that S and A have the
same eigenvalues, counting their multiplicities. Therefore, S is invertible if and only
if A is invertible.
Observation IV-1-14. Let A be an n x n matrix whose distinct eigenvalues are
A = S + N be the S-N decomposition of A. It can be shown
that n x n matrices P1, P2, ... , Pk are uniquely determined by the following three
conditions:
(i)
(ii) P,P1 = O if j 36 t,
(iii) S = A11P1 + A2P2 + ... + AkPk.
Proof.
Note that
In = P1(A) + P2(A) + ... + Pk(A),
{ Pj(A)Ph(A) = O if j h,
S = A1P1(A) + A2P2(A) + ... + AkPk(A).
k
First, derive that P, S = SP; = \j P,. Then, this implies that .X P1 = >ahPj P1. (A).
h=1
Hence, \jPjPA(A) = \h PiPh(A). Thus, PiPh(A) = 0 whenever j h. Therefore,
it follows that P1 = P1(A) = P2P}(A).
76 IV. GENERAL THEORY OF LINEAR SYSTEMS
Set
P1(A) = (A - 3)2
-2(A-4)(A-3)2'
Then,
-1 -2 134 198 2 2 -134 -198
1 2 -125 -186 -1 -1 125 186
P1(A) = 0 0 9 12
P2(A) =
0 0 -8 -12
0 0 -6 -8 0 0 6 9
Therefore,
2 -2 134 198
1 5 -125 -186
S = 4P1(A) + 3P2(A) =
0 0 12 12 I'
10 0 -6 -5
250 500 4000 500
-235 -470 -3760 -470
N = A - S =
15 30 240 30
-10 -20 -160 -20
3 4 3
Example IV-1-19. The matrix A = 2 7 4 has two distinct eigenvalues
-4 8 3
A1= 11, A2=1,and
1 ) 21 1 ) , 1 _ 1 _ (A+9)
PA(A)
PA(A) 100(A - 11) 100(A - 1)2*
Hence,
1 (A - 1)2 - (A + 9)(A - 11)
-
100 100
Set
P1(A) (A - 1)2 (A + 9)(A - 11).
_ P2(A) - -
100 100
Then,
1 0 56 28 1 1100 -56 -28
P1(A) =100- 0 76 38 , P2(A)
100
0 24 -38
0 48 24 0 -48 76
Therefore,
10 56 28
1
S = 11P1(A) + P2(A) = 10 0 86 38
0 48 34 1
20 -16 2
N=A - S= 10 20 -16 2
-40 32 -4
78 IV. GENERAL THEORY OF LINEAR SYSTEMS
(IV.2.1) dt = A(t)y',
where the entries of the n x n matrix A(t) are continuous on an interval I = It :
a < t < b). Let us prove the following basic theorem.
Theorem 1V-2-1. The solutions of (1V2. 1) forms an n-dimensional vector space
over C.
We break the entire proof into three observations.
Observation IV-2-2. Any linear combination of a finite number of solutions of
(IV.2.1) is also a solution of (IV.2.1). We can prove the existence of n linearly
independent solutions of (IV.2.1) on the interval Z by using Theorem I-3-5 with n
linearly independent initial conditions at t = to. Notice that each column vector of
a solution Y of the differential equation
(IV.2.2) dt = A(t)Y
on an n x n unknown matrix Y is a solution of system (IV.2. 1). Therefore, construct-
ing an invertible solution Y of (IV.2.2), we can construct n linearly independent
solutions of (IV.2.1) all at once. If an n x n matrix Y(t) is a solution of equation
(IV.2.2)on an interval I={t:a<t<b}and Y(t)EGL(n,C)for all tE1,then
Y(t) is called a fundamental matrix solution of system (1V.2.1) on.T. Furthermore,
n columns of a fundamental matrix solution Y(t) of (IV.2.2) are said to form a
fundamental set of n linearly independent solutions of (IV.2.1) on the interval T.
2. HOMOGENEOUS SYSTEMS OF LINEAR DIFF. EQUATIONS 79
-W(t)A(t)4s(t) + $(t)A(t)4;(t) = 0.
dt
This implies that the matrix %P(t)4i(t) is independent of t. Therefore, W(t)4i(t) =
%P(r)t(r) for any fixed point r E Z and for all t E Z. Note that the initial values
44(r) and %P(r) at t = r can be prescribed arbitrarily. In particular, in the case
when 4?(r) E GL(n,C), by choosing %P(r) = we obtain WY(t)4i(t) = In
for all t E Z. Thus, we proved the following lemma.
Lemma IV-2-4. Let an n x n matrix 4i(t) be a solution of (IV.2.2) on the interval
Z. Then, 45(t) is invertible for all t E I (i.e., a fundamental matrix solution of
(IV.2.1)) if 4i(r) is invertible for some r E Z. Furthermore, 4 (t)-1 is the unique
solution of (IV.2.8) on Z satisfying the initial condition Z(r) = 4i(r)-1.
Observation IV-2-5. Denote by 4?(t; r) the unique solution of the initial-value
problem
-
(IV.2.4) = A(t)Y, Y(r) = In,
dt
where r E Z. Then, 1(t; r) E GL(n, C) for all t E Z. The general structure of
solutions of (IV.2.1) and (IV.2.2) are given by the following theorem, which can be
easily verified.
Theorem IV-2-6. The C"-valued functwn y(t) = 44(t; r)it is the unique solution
of the initial-value problem
(t, r E Z)
80 W. GENERAL THEORY OF LINEAR SYSTEMS
t 1
In the general case, we define exp I J A(s)ds] by
V,r
+M
exp[B(t)] = In
T11.
B(t)"', where B(t) = Jt A(s)ds.
m=1
However, generally speaking, (IV.2.7) holds only in the case when B(t) and
B'(t) = A(t) commute. In particular, 4(t; r) = exp[(t - r )AJ if A = A(t) is
independent of t. In §IV-3, we shall explain how to calculate exp((t - r)AJ,
using the S-N decomposition of A. Also, (IV.2.7) holds in the case when
A(t) is diagonal on the interval I. A less trivial case is given in Exer-
cise IV-9. It is easy to see that B(t) and A(t) commute if A(t) is a 2 x 2
upper-triangular matrix with an eigenvalue of multiplicity 2. For exam-
ple, the matrix A(t) = [coStt I satisfies the requirement. In this
(jsin t o sin r
case, 4b(t; r) = exp I J t A(s)ds = exp sint - sin r, )
1
exp(sin t -sin r) 10
t 1 r
1 J1
if
(4) det Y(t) = det Y(r) exp trA(s)ds) if Y(t) satisfies (IV.2.2), where det A
and trA are the determinant and trace of the matrix A. This formula is
known as Abel's formula (cf. (CL, p. 28]).
Proof.
Regarding detY(t) as a function of n column vectors {y'1(t),... of
Y(t), set det Y(t) = 7(g, (t), ... , y (t)). Then,
(IV.2.8)
d det Y(t)
dt
- [:
L P(... , A(t)ym(t), ... ).
M=1
and
(IV.3.5)
h j
PJ(A)Ph(A) (3, h = 1, 2, .... k).
0 (A) if h =
36 j
The two matrices S and N commute.
Denote by V, the image of the mapping PJ(A) : C" Cn (cf. Lemma IV-I-10).
It is known that Sp = \,)5 for pin VJ . Hence, Sl p" = \1)5 and
+00
exp[tS]p' = 1 + E (A2t)n }ic earh=1
n-i h
On the other hand, exp(tN] = In + A Nh since N is nilpotent. Therefore,
h=1
(IV.3.9)
0 0 0 1 0 0
P1(A) = 0 0 0, P2(A) = 0 1 0
1 1 1 -1 -1 0
Set
-2 0 0 0 1 0
S = PI(A) - 2P2(A) = 0 -2 0 , N=A-S= 0 0 0.
3 3 1 0 -1 0
3. HOMOGENEOUS SYSTEMS WITH CONSTANT COEFFICIENTS 83
S = (if)P2(A) - (if)P3(A),
then S = A. This implies that N = 0. Thus, we obtain
Using
(e'' t(1 + if)) = cos(ft) - \/3-sin(ft),
2 (e'-1t(1 - if)) wa(ft) + f sin(ft),
84 IV. GENERAL THEORY OF LINEAR SYSTEMS
we find
a(t) b(t) c(t)
exp[tA] = 1 c(t) a(t) b(t)
3
b(t) c(t) a(t)
where
a(t) = I + 2cos(ft),
b(t) = 1 - {cos(ft) + f sin(ft)} ,
c(t) = 1 - {wa(ft) - f sin(ft))
Remark IV-3-5. Fhnctions of a matnx In this remark, we explain how to define
functions of a matrix A.
I. A particular case: Let A0, I,,, and N be a number, the n x n identity matrix,
and an n x n nilpotent matrix, respectively. Also, consider a function f (X) in a
neighborhood of A0. Assume that f (A) has the Taylor series expansion (i.e., f is
analytic at A0)
(h)
f h?Ao)(A
f(A) = f(A0) + 1 - Ao)h.
h=1
n-1 (h)
Since N is nilpotent, the matrix >2f is also nilpotent. Therefore, the
h=1
characteristic polynomial pf(A0,+N)(A) of f(AoI + N) is
Pf(aol-N)(A) = (A - 1(A0))".
II. The general case: Assume that the characteristic polynomial PA(A) of an n x n
matrix A is
PA(A) = (A - .l1)m1(A - A 2 ) ' - 2 ... (A - Ak)mr.
where A1,... , Ak are distinct eigenvalues of A. Construct P, (A) (j = 1, ... , k), S,
and N as above. Then,
Therefore,
Since P2(S) = P,(A) (cf. Observation IV-1-15), this definition applied to S yields
/
log(Ao + u) = 1000) + log t 1 + -o) = logl o) +
+O° m+1
(-I)m
\\ m=1
Io
we obtain
n-1 (-1)m+1
log(.1oIn + N) = log(Ao)ln +
m=1
m
P1(A) = 0 L9 L9 0
P2(A) = 2s 50
12
25 6
50 -12 19
0 0
25 25 5 25
Define S = 11P1(A) + P2(A) and N = A - S. Then N2 = 0. Also,
( sin(11 + x) = -0.99999 + 0.0044257x + 0.499995x2 + 0(x3 ),
t sin(1 + x) = 0.841471+0.540302x-0.420735x 2 + 0(x3).
Therefore,
sin(A) _ (-0.9999913 + 0.0044257N)P1(A) + (0.84147113 + 0.540302N)P2(A)
1.92207 -1.8957 -0.407549
1.0806 -1.42252 -0.591695
-2.1612 0.845065 0.1834
sin(A) -
-- + (2h(-1)h
+ 1)!
A2h+1
_1h
sin(11) = -0.99999 and 112h+1 = -117.147.
(2h + 1)!
h=O
dy
(IV.4.1) = A(t)y
dt
in the case when the n x n matrix A(t) satisfies the following conditions:
(1) entries of A(t) are continuous on the entire real line R,
(2) entries of A(t) are periodic in t of a (positive) period w, i.e.,
Look at the unique n x n fundamental matrix solution 4'(t) defined by the initial-
value problem
dY = A(t)Y,
(IV.4.3)
.it
Y(0) = In
Theorem IV-4-1 (G. Floquet [Fl]). Under assumptions (1) and (2), the funda-
mental matrix solution 4i(t) of (IV.4.1) defined by the initial-value problem (IV.4.3)
has the form
(IV.4.6) 4'(t) = P(t) exp[tB],
where P(t) and B are n x n matrices such that
(a) P(t) is invertible, continuous, and periodic of period w in t,
(Q) B is a constant matrix such that 4'(w) = exp[wB].
Observation IV-4-2. As was explained in Observation IV-3-8, letting 4'(w) _
S + N = S(I + M) be the S-N decomposition of 4(w), we define log($(w)) by
log(4?(w)) = log(S) + log(1 + M), where
log(S) = log(A1)P1(4'(w)) + log(a2)P2(4'(w)) + - + log(Ak)Pk('6(w))-
and
n- 1 l)m+l
log(!! + M) = E (- m Mm.
M=1
In the case when A(t) is a real matrix, the unique solution 4 (t) of problem (IV.4.3)
is also a real matrix. Therefore, the entries of 4'(w) are real. Since S and N are real
matrices, the matrix M = S-'N is real. Therefore, log(I + M) is also real. Let
us look at log(S) more closely. If \j is a complex eigenvalue of 4'(w), its complex
conjugate A3 is also an eigenvalue of 4'(w). In this case, set A,+1 = A3. It is easy
to see that the projection Pj+1(4?(w)) is also the complex conjugate of P,(4'(w)).
However, if some eigenvalues of 4'(w) are negative, log[S] is not real. To see this
more clearly, rewrite log[s] in the form
log[S] _ E log[A,]Pi(4?(w)) + E log1A3]P'(4'(w))-
A, <0 other j
uation, let us look at S2. By virtue of the relations given in Lemma IV-1-9, we
obtain
S2 =
(,\J)2P+ _y2P
A, <0 other j
Notice that
log(S2] _ E log[(Ai)2]Pj(4'(w)) + 2 E log[a3]P,(4(w))
af<0 other 7
is a real matrix. Therefore, we can find a real matrix
log[4'(w)2] = log[S2J + 2log(In + MI.
Now, observe that 4'(,)2 =' (2w) and 4'(t + 2w) = 4'(t). (2w). Thus, setting
(IV.4.7) B = _L log{4'(w)2] and P(t) = 4'(t) exp[-tB],
we obtain the following theorem.
4. SYSTEMS WITH PERIODIC COEFFICIENTS 89
Theorem IV-4-3. In the case when the matrix A(t) is real, the matrix 4s(t) has
the form
P(t) exp[tB],
where P(t) and B are n x n real matrices such that
(a) P(t) is invertible, continuous, and periodic of period 2w in t,
(p) B is a constant matrix such that 4?(w)2 = exp[2wB].
dP(t)
Remark IV-4-4. In the case when 4?(t) = P(t) exp[tB], we have dt
A(t)P(t) - P(t)B. Therefore, (IV.4.1) is changed to
-
dz- = Bz
(IV.4.8)
dt
It I = exp(sin t)I2
P(t) = i(t) exp [ 0 Ot, = 4i(t) l
and j are unknown vectors in R", y = [q1, ?{(t, y7 = 2y"TH(t)y with a real
on 8H
ON Bpi an Ni
(2n) x (2n) symmetric matrix H(t), and Here,
8p 8q.
5. LINEAR HAMILTONIAN SYSTEMS 91
(IV.5.1) = JH(t)y,
where y is a unknown vector in R2n, H(t) is a real (2n) x (2n) symmetric matrix,
and
O In
(IV.5.2) J= [ -In O
The matrix J has the following important properties:
(IV.5.3) JT = -J, J-1 = -J,
where JT is the transpose of J.
Let fi(t) be the unique real (2n) x (2n) matrix such that
do(t)
= JH(t))(t) , 4(0) = 12n.
Lemma IV-5-1. The matrix 4(t) satisfies the condition 4(t)T J4(t) = J, where
t(t)T is the transpose of t(t).
Proof
Differentiate O(t)TJ4(t) to derive
I(t)TH(t)(-J)J4(t) +
a
I(t)T H(t)41(t) - O.
Observing that the left-hand side of this relation is the transpose of the right-hand
side, we conclude that H(t) is symmetric. 0
Observation 1. If GT JG = J, then (G-' )T JG-' = J and GJGT = J. In fact,
GT JG = J implies that (G-' )T JG-' = J. Then, taking the inverse of both sides
of the first relation, we obtain the second relation (cf. (IV.5.3)).
Definition IV-5-3. A (2n) x (2n) real invertible matrix G such that GT JG = J is
called a real symplectic matrix of order 2n. The set of all real symplectic matrices
of order 2n is denoted by Sp(2n, R).
It is easy to show that Sp(2n, R) is a subgroup of GL(2n, R) (cf. Observation 1).
Hence, Sp(2n, R) is called the real symplectic group of order 2n.
Observation 2. If we change Hamiltonian system (IV.5.1) by a linear transfor-
mation
(rV.5.1) becomes
(IV.5.5)
d J [p(t)TH(t)P(t) - JdP( 'P(t) i.
Since P-(t)P(t) = 12 and P(t)T E Sp(2n,R), we obtain
I P(t)_1
dP(t)
dt +
dP(t)
di P(t) =dlzd = 0,
1 P(t)-'J = JP(t)T.
dP(t)
t
Observe that -J P(t) is symmetric since P(t) E Sp(2n, R) (cf. Lemma
IV-5-2). This implies that (IV.5.5) is a Hamiltonian system.
Definition IV-5-4. Transformation (IV.5.4) is called a canonical transformation
if P(t) E Sp(2n, R).
Observation 3. For a given constant matrix G in Sp(2n, R), let us construct the
S-N decomposition G = S + N, where S is diagonalizable, N is nilpotent, and
SN = NS. Since G is invertible, S is invertible (cf. Remark IV-1-13). If we set
M = S-' N, then M is also nilpotent, and we derive the unique decomposition
(IV.5.6) G = S(12 + M), SM = MS.
5. LINEAR HAMILTONIAN SYSTEMS 93
The identity GTJG = J implies JGJ-' = (GT)-l. Using (IV.5.6), we can write
this last relation in the form
(ST)-1 (I2, + MT)-'.
(JSJ-') (J(I2n + M)J-1) =
Set Si = JSJ-1, M1 = J(I2n + M)J-' - 12,,, S2 = (ST)-', and M2 =
(12n + MT) -1 - 12n. It is not difficult to show that S. = 1, 2) are diagonal-
izable and that Af. (j = 1,2) are nilpotent. Furthermore, S,llf, = MSS, (j = 1, 2).
Hence, the uniqueness of S-N decomposition implies that S1 = S2 and 12n + M1 =
12n + M2. Therefore,
(IV.5.7) JSJ-' = (ST)-', J(I2n + Af)J-' = (12n + AfT)-1.
Thus, we proved the following lemma.
Lemma IV-5-5. If we write a symplectic matrix G of order 2n in form (IV.5.6)
by using the unique S-N decomposition of G (where Af = S- IN), the two matrices
S and 12n + M are also symplectic matrices of order 2n.
Remark IV-5-6. Let J be the 2n x 2n matrix defined by (111.5.2) and let H be a
2n x 2n symmetric matrix. Then, it is important to know that if A is an eigenvalue
of JH with multiplicity m, then -A is also an eigenvalue of JH with multiplicity
m. In fact, (JH)T = -HJ = HJ-'. Hence, J-'(JH)TJ = -JH, where AT
denotes the transpose of A.
Remark IV-5-7. Let G be a 2n x 2n symplectic matrix. Then, it is important
to know also that if A is an eigenvalue of G with multiplicity m. then is also an
eigenvalue of G with multiplicity m. In fact, JGJ-' = (CT)-'.
Remark IV-5-8. Assuming that the set of 2n x 2n symplectic matrices with 2n
distinct eigenvalues is dense in the symplectic group Sp(2n, R), it can be shown
that if I (and/or -1) is an eigenvalue of a symplectic matrix G, its multiplicity is
even. Also, det G = 1. In fact, slim, 1. It is also known that det A is equal
to the product of all eigenvalues of A.
Remark IV-5-9. Assuming that Sp(2n, R) is a connected set, we can show that
det G = 1 for all G E Sp(2n, R). In fact, (det G)2 = 1 and the 2n x 2n identity
matrix is in Sp(2n, R).
Part 2
Now, assume that the real symmetric matrix H(t) of system (IV.5.1) is periodic
in t of a positive period w, i.e., H(t + w) = H(t). Then, applying Theorem IV-4-3
and Remark IV-4-4 to system (IV.5.1), we conclude that the unique (2n) x (2n)
matrix solution 0(t) of the initial-value problem d dtt) = JH(t)4y(t), 4i(O) _
12n can be written in the form 0(t) = P(t) exp[tB], where P(t) and B are real
(2n) x (2n) matrices such that P(t) is invertible, P(t+2w) = P(t) for all t E R, and
B is a constant matrix such that 4i(w)2 = exp[2wB]. Furthermore, system (IV.5.1)
is changed to
dzF
(IV.5.8)
dt
= Bi
94 W. GENERAL THEORY OF LINEAR SYSTEMS
by the transformation
(IV.5.9) y = P(t)f.
The main concern of Part 2 is to show that a matrix P(t) can be chosen so
that the transformation (IV.5.9) is canonical. Note that Y = exp[tB] is the unique
solution of the initial-value problem = BY, Y(O) = I2,,. Therefore, if JB is
symmetric, the matrix exp[tB) is symplectic. Hence, if JB is symmetric, the matrix
P(t) = 4i(t) exp[-tB] is symplectic, since 4i(t) is symplectic. This implies that in
order to show that the transformation (IV.5.9) is canonical, it suffices to show that
we can choose P(t) so that JB is symmetric.
Hereafter we use notations and results given in §IV-4. For examples,
k
(1V.5.10)
I2. = E P,('t(w)), 0 (j # t),
,=1
Pj(4'(w))2 = P
Taking the transpose of (IV.5.10), we obtain
k
l2n = yP,(.6(w))T, 0 U 36t),
(IV.5.11) ,=1
\P(.t(w))T )2 = Pj(.0(W))T.
and
B= log[4i(w)2].
Proof.
Upon applying (IV.5.6) to G = 0(w), write 4'(w) in the form -O(w) = S(I2n+M).
Then, S is symplectic, i.e., ST JS = J = 12,,JI2n (cf. Lemma IV-5-5). Using the
formulas
t=1=1
k k
12. = > Pt(4 (w)), 12n = P
(IV.5.13) k k
S= ST = EA,P,(,p(w))T,
r=1 J=1
Fixing a pair of indices j and t and multiplying both sides of (IV.5.14) by Pj (4ti(w))T
from the left and by Pt(4i(w)) from the right, we derive
log[(S2)T](-J)I2n = log[(A,)21
A,A =1
Then,
1og[(S2)Tj(-J) _ log[(A1)2]P,(.t(w))TJPt(+(w))
= J log[S2].
Note that for any square matrix A, we have [log(A)IT = log(AT). Therefore,
[J log(S2)]T = - log[(S2)T]J. Hence, J log[S2] is symmetric.
Observation 5. Let us prove that J log[I2 + M) is symmetric. To do this, we
write first [12n + M]TJ[I2n + M] = J in the form J[I2n + M] = (12n + MT)-'J.
96 IV. GENERAL THEORY OF LINEAR SYSTEMS
Then, write [12n + MT J -1 in the form [12,, + MT J-1 = 12. + M. Using (IV.5.7), it
is not difficult to show that M is nilpotent and JM = MJ. Hence,
J 00 (-lmm+l MM _ (-lmm+l
(IV.5.15) km J
=
m=1 m=1 /
and
as a power series in x. L
Thus, we arrived at the following conclusion.
Theorem IV-5-11. In the case when the matrix H(t) of system (IV.5.1) is real,
symmetric, and periodic in t of a positive period w, there exist (2n) x (2n) real
matrices P(t) and B such that
(a) P(t) E Sp(2n, R) for all t E R,
(b) P(t+2w)=P(t) for alit ER,
(c) B is a constant matrix such that JB is symmetric,
(d) the canonical transformation y" = P(t): changes system (IV.5. 1) to L = BE.
For more information of exponentials in algebraic matrix groups, see (Mar] and
(Si4].
where the entries of the n x n matrix A(t) and the C"-valued function b(t) are
continuous on an interval Z = {t : a < t < b} and r E Z, if E C". As it was
6. NONHOMOGENEOUS EQUATIONS 97
explained in §IV-2, let the n x n matrix 4>(t; r) be the unique fundamental matrix
solution of the homogeneous system
dy
(IV.6.2)
dt
= A(t)y
determined by the initial-value problem
dY = A(t)Y,
(IV.6.3) Y(r) = In,
dt
where r E I. System (IV.6.2) is called the associated homogeneous equation of
(IV.6.1). We treat the solution of (IV.6.1) by using the knowledge of the funda-
mental matrix solution 4)(t; r) of (IV.6.2) (cf. §§IV-2 and IV-3).
Consider the transformation
(IV.6.5)
dz _
dt
r)z(r) = 4.
In fact, differentiating both sides of (IV.6.4), we obtain
di*
A(t)4)(t; r)l + b(t) = r)z""+ 4 (t; r) .
z= i + jt(s;rY1(s)ds,
t
(TV.6.6) l
(t; r)r1 +
Jr
r_
y" _ (t;r)Iit + j4(s;T)_1(s)ds]
t $(t; s)b(s)ds.
Here, use was made of the fact that 0(t; s) = 4'(t; r)O(s; r)-1 (cf. (2) of Remark
IV-2-7; in particular (IV.2.6)).
In a similar way, using (IV.6.6), we can change a nonlinear initial-value problem
The function g(t, y-) is the nonlinear term which satisfies some suitable condition(s).
In the case when the matrix A is independent of t, formulas (N.6.6) and (IV.6.7)
become
respectively.
Example IV-6-1. Let us solve the initial-value problem
where
-2 1 0 [2] #p= 1
A= 0 -2 0 , b(t)= 0 , 1
3 2 1 t 0
r t 1 1 + to-2t
exp[tA] j i0 + exp[-sA]b(s)ds } e-2t
111 fo 111
-9-t+5et-(l+t)e
where the coefficients a°(t), a1(t), ... , and the nonhomogeneous term b(t) are
continuous on an interval I = {t : a < t < b}. In order to reduce (IV.7.1) to
7. HIGHER-ORDER SCALAR EQUATIONS 99
d11
(IV.7.2) = A(t)f + b(t),
dt
where
0 1 0 0 ... 0 1
0 0 1 0 ... 0
A(t) =
an0
0
0
b(t) =
0t)
b(t)
ao(
d
at = A(t)y
(IV.7.3)
of (IV.7.2). Using 4i(t), we can solve problem (IV.7.2) (cf. §IV-6). The first
component of the solution of (IV.7.2) is the solution of problem (IV.7.1). Also,
the first components of n column vectors of the matrix $(t) give the n linearly
independent solutions of the associated homogeneous equation
of (IV.7.1).
Example IV-7-1. Let us solve the initial-value problem
111
111=u, 112=u',
y2=u",
and y= 1122
IY31
100 IV. GENERAL THEORY OF LINEAR SYSTEMS
dt
where
10 1 0 0 1
Three eigenvalues of the matrix A are 1, -2, and 3. The corresponding projections
are
-6 -1 1 1 3 -4 1
P1(A) _ -6 -6 -1 1 I , P2(A) -6 8 -2
15
-6 -1 1 12 -16 4
1-2 1 1
P3 (A)
= 10
-6 3 3
-18 9 9
and N = 0. Therefore,
1
e -615
12
8
-16
1
-2 + 10e -6 3 3.
4
3t
-18
1
9 9
1
Thus,
t
4D(s;0)-1 6(s) ds
and, consequently,
- 1 - (3t + 1)e-3t
1 30t + 25 - 17e'2t + 50et + 2e3L
+ 3[1 - (3t + 1)e-3t]
= 2(15 + 17e-2t
+ 25et + 3e3t)
1)e-3t] 60
9(1 - (3t + 1 2(-34e-2t + 25et + 9e3t)
a fundamental matrix solution of (IV.7.3) has the form 4P(t) = 0' (t) ¢2(t) ,
.01(t) 02(t)
where 01 (t) and ¢2 (t) are two linearly independent solutions of the associated ho-
1 .02(t)
mogeneous equation (IV.7.4). Since 4i(t)-1 = -02(t) where
W(t) -011(t) 01(t)
W(t) = det(4i(t)), the first component of the formula
where ill and q2 are two arbitrary constants. This is known as the formula of
variation of parameters (see, for example, [Rab, pp. 241-2461). Moreover,
Remark IV-7-4. Let 01 (t) and 02(t) be linearly independent solutions of a third-
order linear homogeneous differential equation
Also, let 0(t) be any solution of (IV.7.12). Then, (4) of Remark IV-2-7 implies that
(IV.7.13)
0
10'
01
401
02
452 =c3exp
[ft ao(s)ds]
'P 4i oz
where c3 is a constant. Write (IV.7.13) in the form
where
461 02 01 02 0'2
As(t)
COI
- J ao(s)ds]dr}
I 02
Ao
(T3 eXp I -
J
T ao(s) I dT
+ 02 (t) I ` Ao(r)2 e L
where cl and c2 are constants. This implies that 01(t), 02(t), and
02(r)
Aa(T)2
exp
[f'
i)
ao(s)dsj
1(r)
dr - -02(t) I Ao(r)2 exp [ - f ao(s)
)
ds] dr
EXERCISES IV
IV-1. Let A bean n x n matrix, and let A1, A2, ... , Ak be all the distinct eigenvalues
of A. On the complex A-plane, consider k small closed disks Of = {A : IA - A2 I <
r)} (j = 1,... , k}. Assume that Aj n At = 0 for j # 1. Set
and N=A-S,
where the integrals are taken in the counterclockwise orientation. Show that
(i) Pl + ... + Pk = In,
(ii) Pi Pt = O if U'41),
(iii) A = S + N is the S-N decomposition of A.
EXERCISES IV 103
f (A) _ 1
J=12rri
for t E R.
Remark. This result means that exp[tA] is the inverse Laplace transform of (sln -
U1
A) 1. For example, if A = [ ], then (s12 - A) -1 =
s2 + 1
[_i 1 ]. Hence,
IV-5. Show that J(A2 +T 21" )-'dr= A-1 arctan(tA-1) if every eigenvalue of
an n x n matrix A is a nonzero real number.
m
IV-6. For an n x n matrix A, show that
In
lim (In + A ) = eA.
+00 M
104 IV. GENERAL THEORY OF LINEAR SYSTEMS
where A is a constant n x n matrix, {6, ,j,S are three constant vectors in C", and
y E C' is the unknown quantity.
Hint. If we set y = e`AU, the given problem is changed to
Al =E, u"(0) =r1', u(0)
dt2
find a nonzero constant vector ii E C4 in such a way that the solution l(t) of the
initial-value problem
d9
A9, 9(0)
dY
(E) = A(t)Y - YB(t) F(t),
_WT
is given by
where A(t) and B(t) are n x n matrices continuous in the interval a < t < b,
(1) show that, if Y(to) -I exists at some point to in the interval a < t < b, then
Y(t)-i exists for all points of the interval a < t < b,
(2) show that Z = Y-1 satisfies the differential equation
dZ
= B(t)Z - ZA(t).
dt
IV-15. Find the multipliers of the periodic system dt = A(t)f, where A(t) _
[cost simt
-sint cost
106 IV. GENERAL THEORY OF LINEAR SYSTEMS
Hint. See ILM2, Theorem 1 of Chapter 2 on pp. 69-72 and Lemma 3A of Appendix
of Chapter 2 on pp. 161-163j.
IV-17. Let u(t) be a real-valued, continuous, and periodic of period w > 0 in t on
R. Also, for every real r, let ¢(t, r) and t1'(t, r) be two solutions of the differential
equation
d2y
(Eq) - u(t) y = 0
dt2
such that
C(r) = 10'(0.'r)
0(0, r) V(0, r) 1
,
i,I (0,r) J
JJJ -
Then, ((t, r) is a fundamental
matrix solution of the system
and y = 4'(t, r)c is the solution satisfying the initial condition y(r) = c'. Note that
C(r) = 4'(0, r) and M(r) = 4'(r + w, r). Therefore, 4'(t, r) = 4'(t, 0)C(r) and
4'(t + r) = 4'(t, r)M(r). Now, it is not difficult to see 4'(w, r) = 4'(w, 0)C(r) _
4'(0, r)M(r) = C(r)M(r) and 4'(w, 0) = M(O).
Hint for (II). Since eigenvalues of M(r) and M(0) are the same, the eigenvalues of
M(r) is independent of r. Solutions 77+ (t) of (Eq) satisfying the condition Y7+ (t +
w) = A+n+(t) are linearly dependent on each other. Hence, W+(t) = K+(t) is
independent of any particular choice of such solutions q+(t). In particular K+(t +
W) = 17'+ (t + w) = K+(t). Problem (II) claims that the quantity K+(r) can be
17+(t + w)
found by calculating eigenvectors of M(r) corresponding to A+. Furthermore,
A+ = exp I o
o
K+(r)drJJ I I.
The same remark applies to A_. The solutions 0±(x, r)
of (Eq) are called the Block solutions.
IV-18. Show that
(a) A real 2 x 2 matrix A is symplectic (i.e., A E Sp(2, R)) if and only if det A = 1;
(b) the matrix 01 N is symmetric for any 2 x 2 real nilpotent matrix N.
l 0J
Hint for (b). Note that det(etNJ = I for any real 2 x 2 nilpotent matrix N.
IV-19. Let G, H, and J are real (21n ) x (2n) matrices such that G is symplectic,
H is symmetric, and J = I 0n . Show that G-1JHGJ is symmetric.
IV-20. Suppose that the (2n) x (2n) matrix 4'(t) is the unique solution of the
initial-value problem L'P = JH(t)4', 4'(0) = 12n, where J = I 0n n and H(t)
l T J
is symmetric. Set L(t) = 4'(t)-1JH(t)4'(t)J. Show that d4'dt) = JL(t)4'(t)T,
where 4'(t)T is the transpose of 4'(t).
CHAPTER V
assuming that the entries of the C"-valued function f are convergent power series
in complex variables (x, y-) E Cn+I with coefficients in C, where x is a complex inde-
pendent variable and y E C" is an unknown quantity. The main tool is calculation
with power series in x. In §I-4, using successive approximations, we constructed
power series solutions. However, generally speaking, in order to construct a power
00
series solution y"(x) = E xmd n, this expression is inserted into the given differ-
M=1
ential equation to find relationships among the coefficients d, , and the coefficients
d,,, are calculated by using these relationships. In this stage of the calculation, we
do not pay any attention to the convergence of the series. This process leads us
to the concept of formal power series solutions (cf. §V-1). Having found a formal
power series solution, we estimate Ia',,, i to test its convergence. As the function
x-I f (x, y) is not analytic at x = 0, Theorem 1-4-1 does not apply to system (E).
Furthermore, the existence of formal power series solutions of (E) is not always
guaranteed. Nevertheless, it is known that if a formal power series solution of (E)
exists, then the series is always convergent. This basic result is explained in §V-2
(cf. [CL, Theorem 3.1, pp. 117-119) and [Wasl, Theorem 5.3, pp. 22-251 for the
case of linear differential equations]). In §V-3, we define the S-N decomposition
for a lower block-triangular matrix of infinite order. Using such a matrix, we can
represent a linear differential operator
where f2(x) is an n x n matrix whose entries are formal power series in x with
coefficients in Cn. In this way, we derive the S-N decomposition of L in §V-4 and a
normal form of L in §V-5 (cf. [HKS]). The S-N decomposition of L was originally
defined in [GerL]. In §V-6, we calculate the normal form of a given operator C by
using a method due to M. Hukuhara (cf. [Sill, §3.9, pp. 85-891). We explain the
classification of singularities of homogeneous linear differential equations in §V-7.
Some basic results concerning linear differential equations given in this chapter are
also found in [CL, Chapter 4].
108
1. FORMAL SOLUTIONS OF AN ALGEBRAIC DE 109
over C with the identity element given by E bmxm, where bo = 1 and b,,, = 0
m=0
if m > 1. (For commutative algebra, see, for example, [AM].) Also, we define the
derivative of f with respect to r by dx _ E(m+ 1)a,,,+,x' and the integral
m=0
"0 00
rX r
( f (x)dx by f f (x)dx = airi x"' for f = E a,,,xm E C[[xj]. Then, C[[x]]
0 o m=1 m=0
is a commutative differential algebra over C with the identity element. There are
some subalgebras of C[[x]j that are useful in applications. For example, denote by
C{x) the set of all power series in C[]x]] that have nonzero radii of convergence.
Also, denote by C[x] the set of all polynomials in x with coefficients in C . Then,
C{x} is a subalgebra of C[[x]] and C]x] is a subalgebra of C{x}. Consequently, C[x]
is also a subalgebra of C[[x]].
Let F(x, yo, y, , ... , yn) be a polynomial in yo, y,..... yn with coefficients in
C[[x]]. Then, a differential equation
(V.1.1) F ,...dxn
dy
y,,d"y
=0
f
is called an algebrutc differential equation, where y is the unknown quantity and x
is the independent variable. If F ` x, f, ... ,
dxn
= 0 for some f in C[[x]], then
such an f is called a formal solution of equation (V.1.1). In this definition, it is not
necessary to assume that the coefficients of F are in C{x}.
Example V-1-1. To find a formal solution of
the algebra C((x]]. Similarly, let x°C{x} denote the set of convergent series x°f (x),
where f (x) E C{x}. The set x°C{x} is a commutative differential module over
the algebra C{x}. Furthermore, if a is a non-negative integer, then x°C([x]] C
C[[x]]. If F(x, yo,... , y,) is a formal power series in (x, yo,... , yn) and if fo(x) E
xC[[x]j, ... , fn(x) E xC([xj], then F(x, fo(x),... , fn(x)) E C[[xJj. Also, if
F(x, yo,. .. , yn) is a convergent power series in (x, yo,... , yn) and if fa(x) E xC{x},
... , fn(x) E xC{x}, then F(x, fo(x), ... , f,,(x)) E C{x}. Therefore, using the no-
tation x°C[[x]]n to denote the set of all vectors with n entries in x°C[(x]], we can de-
fine a formal solution fi(x) E xC[[x]]n of system (E) by the condition x = Ax, i)
in C[[x]j . (Similarly, we define x°C{x}n.) Also, in the case of a homogeneous sys-
tem of linear differential equations xfy A(x)g, a series d(x) E x°C[[xj]n is said
where the coefficients f,,, (s) are polynomials in s and no is a non-negative integer
such that 0. Then, we can prove the following theorem.
Theorem V-1-3.
(i) The degree of f,,,, in s is not greater than n.
(ii) If zeros of fno do not differ by integers, then, for each zero r of f,,, there
exists a formal series x' (x) E xr+IC[[x]] such that P[xr(1 +O(x))] = 0.
Proof
(i) Since 6h(x'] = shx', it is evident that the degree of f,,,, in s is not greater than
n.
+o0
(ii) For a formal power series d(x) = E ckxk, we have
k=1
+oo
P[x'(I + b(x))] = P(x'] +
L. Ck P[x'+k J
k=1
+oo +oo +oo
= x'
{mnof"
(s)xm + > Ckxk
k =1
(frn(s
m=no
+ k)xm/
}
= x' fno(s)x +
+oo
((,fm(s) +
m -no
k=1
Ckfm-k(s + k)) x"`m
J1
xa+no AK(
+ + E Ckfno+m-k(s + k)) x"' .
M=1 k=1 J
where y E C' is an unknown quantity and the entries of the C"-valued function f
are convergent power series in (x, y7) with coefficients in C. A formal power series
00
(V.2.2) E x' , ,,, E xC([xfl" (c,,, E C" )
rn-t
is a formal solution of system (V.2.1) if
(V.2.3) X = f(x,0)
dx
where
(1) p = (pl,... ,p") and the p, are non-negative integers,
(2) jpj = pi + + pn and yam' = yi' yn^, where yi,...are the entries
of g,
Setting
xd'o
dx
= Ao + f (x, ¢) - A0 .
Then,
where
(J(x))p
1: xm7m
M=1
and Im E C n. Note that ry'm is determined when c1 i ... , c,,,_ 1 are determined and
that the matrices mI" - A0 are invertible if positive integers m are sufficiently
large. This implies that there exists a positive integers mo such that if 61, ... , C,,,a
are determined, then 4. is uniquely determined for all integers m greater than mo.
Therefore, the system of a finite number of equations
(V.2.5) mEm = A0 + (m = 1,2,... ,mo)
decides whether a formal solution ¢(x) exists. If system (V.2.5) has a solution
{c1 i ... , c,,,a }, those mo constants vectors determine a formal solution (x) uniquely.
Observation V-2-2. Supposing that formal power series (V.2.2) is a formal solu-
N
tion of (V.2.1), set ¢N (x) _ x'6,. Since
m=1
where
(1) 9N,o E xN+1C{x}" and g""N,p E C{x}",
(2) BN(x) is an n x n matrix with the entries in C{x},
(3) the entries of the matrix BN(X) - Ao are contained in xC{x}.
Observation V-2-3. System (V.2.7) has a formal solution
00
=E m=N+1
xm-'1'm
Furthermore, ¢ E xN+1C{x}n.
116 V. SINGULARITIES OF THE FIRST KIND
Remark V-2-6. Under the assumptions of Theorem V-2-5, system (V.2.1) pos-
sibly has many formal solutions in xC[[x)J". However, Theorem V-2-5 states that
there is only one formal solution in xN+1C[[xJJn
Proof of Theorem V-2-5.
We prove this theorem in six steps.
Step 1. Using the argument of Observation V-2-1, we can prove the existence and
uniqueness of formal solution (V.2.9). In fact,
xmrym.
m=N+1
and
(III) If(x, y-)i <_ HIxl "+1 + (IAoI + 1)Iyl for Ixl < b and Iyl <_ p.
2. CONVERGENCE OF FORMAL SOLUTIONS 117
Therefore,
H + IAoI 1]K
II7k+1(x)I < IxIN+I < K 1xIN+1 for jxj < S.
Step 5. Set 1
++1k(x)I
I1'Rk+1 -- nkll = maxiInk+iIZI InI < b}.
Then, since
I ik+1(x) - ik(x)I = fo
we obtain
j
IInk+1 - llkII N ++1 171k - '1k-11I 2lI'7k - 1k-l II
This implies that
lim ilk(x) _ #e+1(x) - #((X)
k-.+00 xN+1 xN+1
e=0
exists uniformly for Ixl < J.
118 V. SINGULARITIES OF THE FIRST KIND
Step 6. Setting
it is easy to show that n'(x) satisfies integral equation (V.2.12). It is also evident
that i7(x) is analytic for IxI < 6. Thus, the proof of Theorem V-2-5 is completed. 0
Now, finally, by using the argument given in Observations V-2-2 and V-2-3, we
obtain the following theorem.
Theorem V-2-7. Every formal solution E xC[[zjJ" of system (V.2.1) is conver-
gent, i.e., E xC{x}".
Remark V-2-8. In general, (V.2.1) may not have any formal solutions. However,
Theorem V-2-7 states that if (V.2.1) has formal solutions, then every formal solution
is convergent.
Remark V-2-9. If yi(x) = xA f (x) E XAC[[xfj" is a formal solution of a linear
system X!LY = A(x)y, the formal power series f is a formal solution of xd _
[A(x) - A!, }i. Therefore, f E C{x}" by virtue of Theorem V-2-7. This proves
convergence of the formal solution constructed in Theorem V-1-2.
where for each (j, k), the quantity AJk is an n, x nk constant matrix. Set
Al = All,
All 0 O .. 0
A21 A22 O .. 0
Am = (m > 2).
S1 = s11, S"'
=
rI Sm_1 0l (m > 2),
J L Cm Smm
[Aim_i
N, = Ni1, Aim = (m > 2),
fm O 1
Nmm
where Cm and fm are am x N,_1 matrices.
Proof
Consider the case m > 2. Since the matrices Sm and Aim are polynomials in Am
with constant coefficients, it follows that
Sm = [B m1 and Aim = I
fm1 01,
ILM0
where 13m-1 and Dm-1 are Nm-1 x Nm_1 matrices, Cm and fm are nm x Nm_1
matrices, and µm and vm are nm x nm matrices. Furthermore, Dm_1 and Vm are
nilpotent. Also, Bm-1Dm-1 = Dm-1Bm-i and tlmvm = 1.m µm. Hence, it suffices
to show that Bm-1 and µm are diagonalizable.
Note that since Sm is diagonalizable, Sm has Nm linearly independent eigen
vectors. An eigenvector of Sm has one of two forms k and [?.] , where p is an
PJ
eigenvector of Bm_1i whereas q is an eigenvector of µm. Therefore, if we count
those independent eigenvectors, it can be shown that Bm-1 has N,, -I linearly in-
dependent eigenvectors, while Pm has nm linearly independent eigenvectors. Note
that Nm = Nm_ 1 + nm. This completes the proof of Lemma V-3-1. 0
Lemma V-3-1 implies that the matrices Sm and Nn have the following forms:
S1 = SI1,
S11 0 0 ... 0
C21 S22 0 ... 0
Sm = (m>2)
Cmi Cm2 ... ... Smm
and
N1 = N11,
Nil 0 D ... 0
Y21 JN22 0 0
Aim = (m>2),
Fm 1 Fm2 ... ... Nmm
120 V. SINGULARITIES OF THE FIRST KIND
N11 0 0
f f21 N22 0 "'
N=
1m 1 Fm2 ... .Wmm 0
Then,
w
where f2(x) _ > xt1 and 1 E Let us identify a formal power series
t=o
ao
a1
xt dt with the vector p = a2 E C. Then, the operator C is represented
t>o
by the matrix
flo 0 0 0 ...
f2l In + f2o 0 0 ...
f12 l1 21, + f2o 0 ...
A =
11m f4.-l fZn-2 ... f21 mI, + fZa 0 .
5. A NORMAL FORM OF A DIFFERENTIAL OPERATOR 121
of A. Since A =
L
j) 11
al I. + So 0 0 ... ...
(V.4.2) S=
am am-l am-2 at min + SO 0
and
No O O
V1 No 0
V2 V1 No
(V.4.3) N=
Vm Vm-1 Vm-2 ... VI No 0 ...
respectively, where the a. and v, are n x n matrices and S2o = So +No is the S-N
decomposition of the matrix S2o.
00 00
where
and
Co
am_ 1 am-2
Since So is diagonalizable, there exist an invertible n x n matrix P0 and a diagonal
matrix Ao such that
Equation (V.5.4) can be solved with respect to Pt, since the linear operator Pt -+
(tl + So)P1 - PtAo is invertible if m is sufficiently large. In this way, we can find
A
P 1
.- - _. _.. _, f _. , l du' \ _ . _. 1
(V.5.6)
=x du +Aou".
dvo P(x)-'Lo(P(x)vo(x)1
dxx) + Aovo(x) = = P(x)`Co[v(x)P(x)1
(V.5.7) x
= P(x)-'v(x)Lo(P(x)1 = vo(x)P(x)-'Co(P(x)1 = vo(x)Ao
This shows that the entry i,k(x) on the j-th row and k-th column of the matrix
vo(x) must have the form v,k(x) = 7jkxAk-a', where ryjk is a constant. Since Lt(x)
is a formal power series in x, it follows that
Observe further that the matrix &. (x) can be written in the form
'711 'Yin
vo(x) = x-^°rx^°, where r=
'7ni Inn
On the other hand, since No is nilpotent, vo(x)P can be written in a form i (x)P =
X'pQP(x), where mp is a non-negative integer such that lim mP = +oo and
P+oo
the entries of the matrix Qp(x) are power series in x with constant coefficients.
Therefore, L0(x)P = 0 if p is sufficiently large. This implies that the matrix r is
nilpotent. Since v(x) = P(x)vo(x)P(x)-1, we obtain v(x)^ = O.
Thus, we arrive at the following conclusion.
Theorem V-5-1. For a given differential operator (V.5.1), let 0o = So +No be
the S-N decomposition of the matrix N. Then, there exists an n x n matrix P(x)
such that
(1) the entries of P(x) are formal power series in x with constant coefficients,
(2) P(O) is invertible and SoP(0) = P(0)Ao, where A0 is a diagonal matrix whose
diagonal entries are eigenvalues of Q0r
124 V. SINGULARITIES OF THE FIRST KIND
r=
Ynl Inn
with constants "ljk such that
dv'
?{o(vl = x'L fXLTf + zLLv + Aoz' 7} = xdx + (Ao + L)v
Hence,
to
d1U
(V.5.15) VV! = x + (Ao + L + r)u = 0.
Since L - A0 - L = -A0, the matrix 44(x) can be also written in the form
n-1 h
4i(x) = P(x)x-ne
In + (-1)h (lohx) rh
h=1
I n-i h
However, x-Ao and I nr >(-1)h [lohx[ rh do not commute, whereas x-A°-L
h=1
n-I [log x[ h
and In + E(-1)h h!
rh commute.
h=1 I
Remark V-5-7. The methods used in §§V-3, V-4, and V-5 are based on the
original idea given in [GerL[.
Example V-5-8.
In order to illustrate the results of this section, consider the differential equation
of the Bessel functions
(V.5.17) x + Q(x)y = 0, y=
0 -1 0 -1 0 0
where 11(x) = x - a2 =10+x11i c o = a2 and 0i = 1
0 0
4 4 4 0
To begin with, let us remark that the S-N decomposition of the matrix 11 is
given by S2o = So + No, where
O if a = 0, N if a = 0,
and
80 =
{Q if a > 0, N0 - to if a>0.
a 0
Po InoP0 if a = 0,
vo(x) r0 a(a)xn
if a>0,
0 0 J
+00
Conclusion V-5-9. There exists a unique 2 x 2 matrix P(x) = >xtPe such that
1=o
(1) the matrices Pl for t = 0,1, ... , m are given in Step 8,
(2) the power series P(x) converges for every x,
(8) the transformation y = P(x)u" changes (V.5.17) to
dii
(V.5.18) xaj + (Ao + vo(x)) i = 0'.
We illustrate the scheme given above in the case when a = 0. First we fix m = 2.
Then,
0 -1 0 0 0 0 0 0 0 0 00
0 0 0 0 0 0 0 0 0 0 00
1 1
0 0 1 -1 0 0 1 0 0 0
4 2
A2 = S2 = 1 1
1 0 0 1 0 0 0 1 0 0
4 4 4
0 0 0 0 2 -1 3 3 1 1
2 0
32 32 4 2
0 0 0 0 2 - 1 3 1 1
0 2
4 16 32 4 4
and
0 -1 0 0 0 0
0 0 0 0 0 0
1 1
0 -1 0 0
4 2
N2= 1
0 0 0 0 0
3 3 1 1
0 -1
32 32 4 2
0 0 0
16 T2- 9
1192 -320
64 -128
Calculating eigenvectors of S2, we find a 6 x 2 matrix P2 = -16 16
such
-32 48
0 1
1 0
that S2P2 = 0. Note that, in this case, Ao = 0.
128 V. SINGULARITIES OF THE FIRST KIND
Set
1 1 3 3
4 2 32 32
0 1 1 3
4 16 32-
M2 (x) = 1 o + xvl + x2v2
and
]
Then, Qz-' 11f2(x)Q2(x)
-1 2
+ O(x3). Note that
1 5
1,° 12 634 [192 [--21 4
[0 64 -128] = 2].
01]
64 64
0 -1
dy
(V.5.20) xdx + X 0 Y = 0'
4
to
d6
(V.5.21) x3i + [_1 2, u" = 0.
-
Furthermore, the transformation y" = P(x)Po iv changes (V.5.20) to
(V.5.22) x + 10 0
" = 0.
5. A NORMAL FORM OF A DIFFERENTIAL OPERATOR 129
1
0 0 0 0 0 0 0 0 0 0
4
1
0 0 1 -1 0 0 1 -1 0 0
8 4
A2 = 132= 3
0 -4 1 0 0
16
1
8 4
1 1
0 0
0 0 0 0 2 -1 1 1 1 1
2 -1
16 16 4
0 0 0 -4 2 J
3 1 3 1 1 2
64 16 16 8 4
and 4
0 0 0 0 0 01
0 0 0 0 0 0
1 1
0 0 0 0
8 4
Ar2 = 1 1
8 0 0 0 0
16
- 1 1 1 1
00
16 16 8 4
-3 1 1 1
0 0
64 16 16 8
384 32
192 -16
Calculating eigenvectors of S2, we find a 6 x 2 matrix P2 = _72
-124 such
2 0
5 1
1
0
that S2P2 = P2Ao. Note that, in this case, A0 = 1 2 1 . Set
1
L0 2
1 1 1 1
8 4 16 16
vl = 1 1 v2 = 1
3
16 8 64 16
M2(x) = xl/1 + x2v2,
and
Po
84 62 -48 -12
1'2 = [2 1j
(V.5.23) [19 3 J' Pl - [-72 -14] , ,
-1
(V.5.24) xdx + V = 0
0
to
(V.5.25) x- +
2 48I u = U.
0 1
2
For a further discussion, see IHKS[. A computer might help the reader to calculate
S2i N2, and P2 in the cases when a = 0 and a = 1. Such a calculation is not difficult
in these cases since eigenvalues of A2 are found easily (cf. §IV-1).
dP(x)
= P(x)(Ao + vo(x)) - Q(x)P(x)
00
and B(x) = Sto + E xmBm by the equation
m=1
where XXh, YYh, and H,h are m, x mh matrices, equation (V.6.1) becomes
and the power series is convergent. If there is no other linearly independent solu-
tion of this form, a second solution can be constructed by using the idea explained
in Remark IV-7-3. This second solution contains a logarithmic term. Similarly,
a third-order linear homogeneous differential equation has a solution of the form
+00
01(x) _ (x - a)° c, (x - a)" at a regular singular point x = a. Using this solu-
n=0
tion, the given equation can be reduced to a second-order equation. In particular, if
there exists another solution ¢(x) of this kind such that ¢1 and ¢2 are linearly in-
dependent, then the idea given in Remark IV-7-4 can be used to find a fundamental
set of solutions.
In general, a fundamental matrix solution of system (V.5.14) can be constructed
if the transformation y" = P(x)xLV of Theorem V-5-4 is found. In fact, if the
definition of fo(x) of Observation V-5-5 is used, P(x)xL2-no-L-r is a fundamental
matrix solution of (V.5.14). The matrix P(x) can be calculated by using the method
of Hukuhara, which was explained earlier.
where k is a positive integer and the entries of the n x n matrix A(x) are convergent
power series in x with complex coefficients, then the singularity at x = 0 is said to
be of the second kind
In §V-5, we proved the following theorem (cf. Theorem V-5-4).
Theorem V-7-1. For system (V.7.1), there exist a constant n x n matrix A0 and
an n x n invertible matrix P(x) such that
(i) the entries of P(x) and P(x)-1 are analytic and single-valued in a domain
0 < jxi < r and have, at worst, a pole at x = 0, where r is a positive number,
(ii) the transformation
(V.7.3) y' = P(x)i7
changes (V.7.1) to a system
(V.7.4) du = Aou".
Theorem V-7-2. For system (V.7.2), there exist a constant n x n matrix A0 and
an n x n invertible matrix P(x) such that
(i) the entries of P(x) and P(x)-1 are analytic and single-valued in a domain
V = {x : 0 < lxi < r}, where r is a positive number,
(ii) the transformation
(V.7.5) y = P(x)i
changes (V.7.2) to (V.7.4).
Proof.
Let 4i(x) be a fundamental matrix of (V.7.2) in D. Since A(x) is analytic and
single-valued in D, fi(x) _ I (xe2"t) is also a fundamental matrix of (V.7.2). There-
fore, there exists an invertible constant matrix C such that 46(x) = 4i(x)C (cf. (1)
of Remark IV-2-7). Choose a constant matrix Ao so that C. = exp[2rriAo] (cf. Ex-
ample IV-3-6) and let P(x) = 4(x)exp[-(logx)Ao]. Then, P(x) and P(x)-1 are
analytic and single-valued in D. Furthermore,
dP(x) = dam)
exp(_(logx)Ao] - P(x)(x-'Ao)
= x-(k+1)A(x)P(x) - P(x)(x-'Ao).
dg
(V.7.6) = F(x)y,
dx
where every entry of the n x n matrix F(x) is analytic and single-valued on the
domain V even if such an entry of F(x) possibly has an essential singularity at
x = 0. More precisely speaking, for system (V.7.6), there exist a constant n x n
matrix Ao and an n x n invertible matrix P(x) satisfying conditions (i) and (ii) of
Theorem V-7-2 such that transformation (V.7.5) changes (V.7.6) to (V.7.4).
Now, we state a definition of regular singularity of (V.7.6) at x = 0 as follows.
Definition V-7-3. Let P(x) be a matrix satisfying conditions (i) and (ii) such
that transformation (V.7.5) changes (V.7.6) to (V.7.4). Then, the singularity of
(V.7.6) at x = 0 is said to be regular if every entry of P(x) has, at worst, a pole
atx=0.
Remark V-7-4. Theorem V-7-1 implies that a singularity of the first kind is a
regular singularity. The converse is not true. However, it can be proved easily that
a regular singularity is, at worst, a singularity of the second kind. Furthermore, if
(V.7.2) has a regular singularity at x = 0, then the matrix A(0) is nilpotent. This
is a consequence of the following theorem.
134 V. SINGULARITIES OF THE FIRST KIND
Theorem V-7-5. Let A(x) and B(x) be two n x n matrices whose entries are
formal power series in x with constant coefficients. Also, let r and s be two positive
integers. Suppose that there exists an n x n matrix P(x) such that
(a) the entries of P(x) are formal power series in x with constant coefficients,
(b) det(P(x)] ,E 0 as a formal power series in x,
16 -1
(c) the transformation y = P(x)ii changes the system x' = A(x)y to x' 2i _
X
B(x)il.
Suppose also that s > r. Then, the matrix B(O) must be nilpotent.
Proof
Step 1. Applying to the matrix P(x) suitable elementary row and column opera-
tions successively, we can prove the following lemma.
Lemma V-7-6. There exist two n x n matrices
+oo +oo
T(x) _ xmTm and S(x) _ x"'Sm,
m=0 m=o
(V.7.7) ah(x)bhl/ = 0,
h=0
assume that the coefficients ah are convergent power series in x and that an (x) # 0.
Set y; = 6'-1y (j = 0,... , n - 1). Then, (V.7.7) becomes the system
f 0 1 0 0
0 0 0 ... 0 yl
(V.7.8) 6y = 91
/i! =
0 0 0 . 1
yn
a0 an-1
an as
If we change (V.7.8) by the transformation y" = diag[l, x-O, '.T-(n-1)0
Iii,
then
(V.7.9)
x°6u
0 1 0 ... 0
0 0 0 0
+ x°diag[l, a, 2a,... , (n - 1)a) u'.
0 0 0 ... 1
-xna ao
a.
-x' an-1
an
ah(x)x(n-h)O
Set bh(x) = . Choose a non-negative rational number a so that bh(X)
an(x)
(h = 0,... , n - 1) are bounded in a neighborhood of x = 0. Also, if a must be
positive, choose a so that bh(0) 0 0 for some h. Then, the matrix on the right-hand
side of (V.7.9) is not nilpotent at x = 0 if or > 0. Hence, we define a as the order of
the singularity of (V.7.7) at x = 0. System (V.7.2) can be reduced to an equation
(V.7.7) (cf. §XIII-5).
The following theorem due to J. Moser [Mo) concerns the order of a given sin-
gularity-
Here, r = rank(Ao) and min is taken over all n x n invertible matrices T of the
form T(x) = x-9 E x°T,,, where q is an integer, T are constant n x n matrices,
and the power series is convergent. Note that det T(x) 0 0, but det To may be 0.
Assume that m(A) > 1. Then, we have m(A) > p(A) if and only if the polynomial
P(A) =
c K[xj' (x E V)
for some positive number K and a real number m. These two numbers may depend
on 0 and V. In fact, Theorem V-7-2 and its remark imply that a fundamental matrix
solution of (V.7.6) is P(x)xA0. The matrix P(x) has, at worst, a pole at x = 0 if
and only if !P(x)I < Kjxjr in a neighborhood of x = 0 for some positive number
K and a real number p (cf. [CL, §2 of Chapter 4, pp. 111-1411). Another criterion
which depends only on a finite number of coefficients of power series expansion of
the matrix A(x) of (V.7.2) was given in a very concrete form by W. Jurkat and D.
A. Lutz [JL] (see also 15117, Chapter V, pp. 115-1411).
Now, let us look into the problem of convergence of the formal solution which
was constructed in Theorem V-1-3. Let
n
P = Eah(x)Jh
h=0
be a differential operator with coefficients ah(x) in C{x). Assume that n > I and
a,(x) 0 0. Defining the indicial polynomial of the operator P as in §V-1,
we prove the following theorem.
Theorem V-7-10.
(i) In the case when the degree of f b in s is equal to n, if we change the equation
P[y1 = 0 to a system by setting yt = y and yj = 61-1 [y) (j = 2, ... , n), the
system has a singularity of the first kind at x = 0.
EXERCISES V 137
(ii) In the case when the degree of fn ins is less than n, if we change the equation
P[y] = 0 by setting y1 = y and y, = b'-1 [y] (j = 2,... , n), the system has an
irregular singularity at x = 0.
Proof
n 00
EXERCISES V
m=1
x(1-x)!f 2 + +1)J
Y
- c3y = 0,
where a, /3, and -y are complex constants.
138 V. SINGULARITIES OF THE FIRST KIND
Comment. The series F(a, Q, ry, x) is called the hypergeometric series (see, for
example, [CL; p. 1351, 101; p. 159], and [IKSYJ).
V-3. For each of the following differential equations, find all formal solutions of
c
the form x'' I 1 + > cmxm] . Examine also if they are convergent.
L m=1
where b = xa, the quantities a, 3, and 7 are nonzero complex constants and m
is a positive integer.
V-4. Given the system
show that
(i) (E) has two linearly independent solutions
J2(t) = 1612 (t
(t)
)1 , where 02(t) = t + +00 bmt-'n - ¢1(t)logt,
m-1
with the constants bm determined by b_1 = 1, bo = 0, and
2m + 1
m(m + 1)bm - b,,,-, = (m = 1,2,... ),
m!(m+ 1)!
(ii) 1 lim t-1(Y(t)-e IN) = O for the fundamental matrix solution Y(t) = [if1(t)12(t)J,
(iii) the limit of e-'NY(t) as t -+ +oo does not exists.
V-5. Let y be a column vector with n entries and let Ax, y-) be a vector with n
entries which are formal power series in n + 1 variables (x, yj with coefficients in
C. Also, let u' be a vector with n entries and let P(x, u) be a vector with n entries
which are formal power series in n + 1 variables (x, u") with coefficients in C. Find
the most general P(x, u") such that the transformation u + xP(x, u7 changes
dy" du"
the differential equation = f (X, y-) to = 0.
ds
Hint. Expand xP(x, u) and f (x, u+xP(x, u)) as power series in V. Identify coeffi-
d[xP(x, V-)]
cients of with those of Ax, u'+zP(x, iX)) to derive differential equations
which are satisfied by coefficients of xP(x, u).
EXERCISES V 139
V-6. Suppose that three n x n matrices A, B, and P(x) satisfy the following
conditions:
(a) the entries of A and B are constants,
(b) the entries of P(x) are analytic and single-valued in 0 < [xj < r for some
positive number r,
dii
(c) the transformation y' = P(x)u changes the system xfy = Ay" to xjj = Bu.
Show that there exists an integer p such that the entries of xPP(x) are polynomials
in x.
Hint. The three matrices P(x), A, and B satisfy the equation xd) = AP(x) -
+"0
P(x)B. Setting P(x) = >2 xmPm, we must have mPm = APm - PmB for all
m=-oo
integers m. Hence, there exists a large positive integer p such that Pm = 0 for
ImI ? P.
V-7. Let ff be a column vector with n entries {y,... , and let A(ye) be an n x n
matrix whose entries are convergent power series in {yj, ... , yn} with coefficients
in C. Assume that A(0) has an eigenvalue A such that mA is not an eigenvalue of
A(0 for any positive integer m. Show that there exists a nontrivial vector O(x)
with n entries in C{x} such that y" = b(exp[At]) satisfies the system L = A(y-)y.
N
V-8. Consider a nonzero differential operator P = >2 ak(x)Dk with coefficients
k=O
ak(x) E C([x]], where D = dx. Regarding C([x]] as a vector space over C, de-
42(x))...(6
P = a+,(x)(b - &1(x))(b - - 0n(x)),
140 V. SINGULARITIES OF THE FIRST KIND
where all the functions ¢, (x) (j = 1, 2, ... , n) are convergent power series in x and
4f(O) = Aj.
Hint. Without any loss of generality, we can assume that an(x) = 1. Then, An +
n-1
Eak(O)Ak = (A - A1)... (A - An). Define constants {yo... , In-2} by A' ' +
k=0
n-2
,YkAk
= (A - A2) ... (A - An). For v] (X) E xC[[xI] and ch(x) E xC([x]l (h =
k=O
0,. . . , n - 2), solve the equation
( n-2
(C) P = (6 - Al - 2(7h+Ch(x))6h
h=0
where h=I,...,n-2.
n
V-10. Consider a linear differential operator P = E atbt, where ao,... , an are
t=o
complex numbers and 6 = x. The differential equation
(P) P[y1= 0
is called the Cauchy-Euler differential equation. Find a fundamental set of solution
of equation (P).
(6-06-8)(6-a-8)[yl =xn'Y'
where 6 = xjj and m is a positive integer, whereas a and /3 are complex numbers
such that they are not integers and R[al < 0 < 8t[01.
V-12. Find the fundamental set of solutions of the differential equation
Remark. Differential equation (LGE) is called the Legendre equation (ef. [AS, pp.
331-338] or [01, pp. 161-189]).
V-13. Show that for a non-negative integer n, the polynomial Pn(x) = 2nn!
1
[(1-x2) ] +n(n+1)Pn=0.
d
Show also that these polynomials satisfy the following conditions:
(1) Pn(-x) = (-1)nPn(x), (2) Pn(1) = 1, (3) JPn(t)j ? 1 for {xj > 1,
+00
E Pn(x)tn, and (5) lPn(x)I < 1 for Jxj < 1.
(4) 1 - 2xt + t = n=o
Hint. Set g(x) = (1 - x2)n. Then, (1 - x2)g'(x) + 2nxg(x) = 0. Differentiate this
relation (n + 1) times with respect to x to obtain
x2)g(n+2)(x)
(1,- - 2(n + 1)xg(n+1)(x) - n(n + 1)g(n)(x)
+ 2nxg(n+1) (x) + 2n(n + 1)g(")(x) = 0
or
(1 -X 2)g(n+2)(x) - 2ng(n+1)(x) + n(n + 1)g(n)(x) = 0.
Statements (1), (2), (3), and (4) can be proved with straight forward calculations.
Statement (5) also can be proved similarly by using (4) (cf. [WhW, Chapter XV,
Example 2 on p. 303]). However, the following proof is shorter. To begin with, set
2
Then, F(±1) = Pn(±1)2 = 1, p(X) _ (P"(x))
n(n + 1)
,and F(x) = Pn(x)2 if (x)
0. Therefore, for 0 < x < 1, local maximal values of Pn(x)2 are less than F(1) = 1,
whereas, for -1 < x < 0, local maximal values of Pn(x)2 are less than F(-1) = 1.
Hence, Pn(x)I < 1 for JxI < 1 (cf. [Sz, §§7.2-7.3, pp. 161-172; in particular,
Theorems 7.2 and 7.3, pp. 161-162]. See also (NU, pp. 45-46].)
The polynomials Pn(x) are called the Legendne polynomials.
V-14. Find a fundamental set of solutions of (LGE) at x = oo. In particular, show
what happends when a is a non-negative integer.
V-15. Show that the differential equation b"y = xy has a fundamental set of
solutions consisting of n solutions of the following form:
(logx)k-1-A
V-16. Find the order of singularity at x = 0 of each of the following three equa-
tions.
(i) x5{66y - 362y + 4y} = y,
(ii) x5{56y - 3b2y + 4y} + x7{b3y - 55y} = y,
(iii) x5y"' + 5x2y" + dy + 20y = 0.
V-17. Find a fundamental matrix solution of the system
dbi
x2 = xyi + y2,
dx
dy2
x2 = 2xy2 + 2y3,
dx
x2 dx3 = x3y! + 3y3.
with coefficients ak(x) and bk(x) in C{x} such that P[0] = 0 and Q I = 0. Show
J
that 0 is convergent. l
EXERCISES V 143
(VI.1.1) + 9(x)y = 0.
dx2
144
1. ZEROS OF SOLUTIONS 145
Proof.
Assume without any loss of generality that Sl < 1;2 and 01(x) > 0 on 1;1 < x < £2.
Notice that 41(1::1) = 0, (1:2) < 0. A contradiction
1 (l:l) > 0, 01(S2) = 0, and
will be derived from the assumption that 02(x) > 0 on S1 < z < 1;2. In fact,
assumption (ii) implies
_ 01 ( x ) d202 (x) =
02(x) d 20, (7)
VI .1. 2 [ 92 (x) - 91 (x)10 (x)0 2 (x)
1
and, hence,
E1
(x) - 91 (x)j¢ (x)02 (x)dx .
1
The left-hand side of (VI.1.3) is nonpositive, but the right-hand side of (VI.1.3) is
positive. This is a contradiction. 0
A similar argument yields the following theorem.
Theorem VI-1-2. Suppose that
(i) g(x) is continuous on an interval a < x < b,
(ii) 0i(x) and 02(x) are two linearly independent solutions of (VI.1.1),
(iii) 1;1 and 6 are successive zeros of 01 (x) on a < x < b.
Then, 02(x) must vanish at some point 3 between t;1 and 1;2.
Proof.
Assume without any loss of generality that Sl < £2 and .01 (x) > 0 on S1 < x < 1;2.
Notice that 01(11) = 0, 1 (S1) > 0, 01(12) = 0, and X1(1;2) < 0. Note also that
if 02(1) = 0 or 02(6) = 0, then 01 and 02 are linearly dependent. Now, a
assumption that 02(x) > 0 on fi < x < 2
contradiction will be derived from the assumption
a- (x) - p1(z) d2 02 (x) = 0 and, hence,
In fact, assumption (u) implies 02(x)
dS2 dx2
(VI.1.4) 02(2) 1(6) - 4'201) 1(W =0.
Since the left-hand side of (VI.1.4) is positive, this is a contradiction. 0
The following result is a simple consequence of Theorem VI.1.2.
Corollary VI-1-3. Let g(x) be a real-valued and continuous function on the in-
terval Zo = {x : 0 < x < +oo}. Then,
(a) if a nontrivial solution of the differential equation (VI.1.1) has infinitely many
zeros on I , then every solution of (VI.1.1) has an infinitely many zeros on
Zo,
(b) if a solution of (VI.1.1) has m zeros on an open subinterval Z = {x : a < x <
3} of Zo, then every nontrivial solution of (VI.1.1) has at most m + 1 zeros
on Z.
Lemma VI-1-4. Let g(x) be a real-valued and continuous function on the interval
To = {x : 0 < x < +oo}, and let 771(x) and M(x) be two solutions of differential
equation (VI 1.1). Then,
(a) W (x; 772 , ri2 ), the Wronskian of {rli (x), 712(x)}, is independent of x,
(b) if we set t;(x) =
771(x),
then
4(x) = c , where c is a constant.
772(x) dx 772(x)2
Also, if 77(x) is a nontrivial solution of (VI.1.1) and if we set w(x) = then
we obtain
dw(x)
(VI.1.5) + w(x)2 + g(x) = 0 .
Proof.
(a) It can be easily shown that
d 772(x)
771(x)
= 771(x) 772(x) -g(x) ! 712(x) =0
dx I77i(x) 772(x) 77i'(x) 772(x) 771(x) 712(x)
The following theorem shows the structure of solutions in the case when every
nontrivial solution of differential equation (VI.1.1) has only a finite number of zeros
on To= {x:0<x<+oo}.
Theorem VI-1-5. Let g(x) be a real-valued and continuous function on the inter-
val I. Assume that every nontrivial solution of differential equation (VI.1.1) has
only a finite number of zeros on T. Then, (VI.1.1) has two linearly independent
solutions 777(x) and 712(x) such that
(a) lim 771(x) = 0,
x-+oo 712(x)
(b)
+00 dx +00 dx
(VI.1.6) x)2 = +oo and 1 2
< +00,
1.0 771( Jxo 7r (x)
(a) Let (1(x) and (2(x) be two linearly independent solutions of (VI.1.1) such that
(, (x) > 0 (j = 1, 2) for x > xo > 0. Using (b) of Lemma VI-1-4, we can derive the
1. ZEROS OF SOLUTIONS 147
(1(x)
following three possibilities (1) =limo(2(x) = 0, (ii) =U +oo, and (iii)
(VI.1.7)2 + 1
A
z2
= 0,
where A is a constant, have infinitely many zeros on the interval -oo < x < +oo if
and only if A > 4
Proof
Theorem VI-1-8. Let g(x) be a real-valued and continuous function on the inter-
f+C0
valA. If 1 jg(x)ldx < +oo, then there exist unbounded solutions of differential
0
equation (VI. 1. 1) on Ia.
Proof.
The assumption of this theorem and (VI.1.1) imply that lim dq(x) = -Y exists
:-+oo dx
for any bounded solution n(x) of (VI.1.1). If 7 0, then i(x) is unbounded. Hence,
lim odd) = 0. Therefore, calculating the Wronskian of two bounded solutions
z
of (VI.1.1), we find that those bounded solutions are linearly dependent on each
other. This implies that there must be unbounded solutions. 0
Remark VI-1-9. Theorems VI-1-1 and VI-1-2 are also explained in ]CL, §1 of
Chapter 8, pp. 208-211] and [Hart, Chapter XI, pp. 322-403]. For details concern-
ing other results in this section, see also [Cop2, Chapter 1, pp. 4-33] and [Be12,
Chapter 6, pp. 107-142].
Let 4(x) and O(z) be two solutions of the homogeneous linear differential equa-
tion
such that
(VI.2.2) m(a) = sins, p(a)d'(a) = cosa, ,p(b) = sin f3, p(b)t'(b) = cosfi,
respectively. Then, these two solutions satisfy the boundary conditions
The two solutions 4(x) and /;(x) are linearly independent if and only if
The first basic result of this section is the following theorem, which concerns the
existence and uniqueness of solution of (BP).
Theorem VI-2-1. If the two solutions 4(x) and ip(x) of (VI.2.1) are linearly
independent, then problem (BP) has one and only one solutwn on the interval
I(a, b).
Proof
Using the method of variation of parameters (cf. Remark IV-7-2), write the
general solution y(x) of the differential equation of (BP) and its derivative y'(x)
respectively in the following form:
(VI.2.5)
6
VW = CIOW + C20 W + 4(x) 'j'(_)f (_) 4 + i,&(x)1= o(f)f
J P(f)WW a
where cl and c2 are arbitrary constants and W (x) denotes the Wronskian of
Now, using (VI.2.3) and (VI.2.4), it can be shown that solution
(VI.2.5) satisfies the boundary conditions of (BP) if and only if cl = 0 and c2 =
0. 0
Observation VI-2-2. It can be shown easily that p(x)W(x) is independent of x.
Observation VI-2-3. Under the assumption that the two solutions 40(x) and tfi(x)
of (VI.2.1) are linearly independent, the unique solution of (BP) is given by
rb
(VI.2.6)
y(x) = d(x)
Jx P(A)W (O 10(x) I i (- ))- W(f )
rb owfwdC + V,'(x) r= 4(W W 4.
Y 'W = O(x)
1r W Ja P(OW (O
150 VI. BOUNDARY-VALUE PROBLEMS
Setting
The function G(x,l;) is called Green's function of problem (BP). The following
theorem gives the characterization of Green's function.
Theorem VI-2-4. The function G(x, C) given by (VI. 2.7) satisfies the following
conditions:
(i) G(x, l:) is continuous with respect to (x, e) on the region D = {(x,) a < x <
b, a< £<b},
(ii) 8 (x, l:) is continuous with respect to (x, C) on the region D - a<
C < b},
(iii) at ({, l;), we have
0 and
-PW
2. STRUM-LIOUVILLE PROBLEMS 151
where C2(a, b) denotes the set of all real-valued functions which are twice contin-
uously differentiable on the interval I(a, b) = {x : a < x < b). Define also an
inner product (f, g) for two real-valued continuous functions f and g on I(a, b) by
(f,9) = fa
Since
(f, 4191) =
(()) +
4
[
J6f()
(f,L[g) - (L(f1,9)
={p(b)f(b)9'(b) - p(b)g(b)f'(b)} - {p(a)f(a)9 (a) - p(a)9(a)f'(a)}.
Also, since
( f (a) cos a - p(a) f'(a) sin a = 0, f (b) cos Q - p(b) f'(b) sin O= 0,
t 9(a) cos a - p(a)9'(a) sin a = 0, g(b) cos, - p(6)9'(b) sin /3 = 0
Observation VI-2-6. In Theorem VI.2.1, it was assumed that the two solutions
¢(x) and '(x) are linearly independent. Consider the case when this assumption
is not satisfied. So, assume that ¢(x) and ?P(x) are linearly dependent. This means
that
0(a) cos a - p(a)0'(a) sin a = 0 and ¢(b) cos ji - p(b)5'(b) sin fi = 0.
In other words,
(VI.2.8) lr[o) = 0 and 0 E V(a, b).
The boundary-value problem (BP) can be written in the form
(VI.2.9) G[yl = f and y E V(a, b).
Using (VI.2.8) and (VI.2.9), we obtain
(f, 0) = (C [y], 0) = (y, x[01) = 0,
if there exists a solution y of problem (VI.2.9). The converse is also true, as shown
in the following theorem.
Theorem VI-2-7. Assume that O(x) satisfies condition (VI.2.8). Then, if a real-
valued continuous function f (x) on T (a, b) satisfies the condition
(VI.2.10) (f. 0) = 01
problem (BP) has solutions depending on an arbitrary constant.
Proof.
Using the method of variation of parameters, write the general solution y(x) of
the differential equation of (BP) and its derivative y(x) respectively in the following
form:
(VI.2.11)
b
?% (z) = J p_ {
r P(f)ww
+ F' (r)
u P(t)
1-v _),
where u(x) is a solution of the linear homogeneous differential equation (VI.2.1)
such that 0 and p are linearly independent, two quantities ca and c2 are arbitrary
constants, and W(x) denotes the Wronskian of {0(x),p(x)}. Note that p(x)W(z)
is independent of x and that
(VI.2.12) p(a)cosa-p(a)u'(a)sina,-40, p(b)cosQ-p(b)p'(b)sinfl,40.
From (VI.2.10) and (VI.2.11), we derive
b
y(a) = ctm(a) + c2p(a) + 0(a)
Jc P(4)W(E)
r° A,
y'(a) = ca0'(a) + c2 (a) + 0'(a)
a -N)
y(b) = ciO(b) + c2p(b), y'(b) = c,O'(b) + c2p'(b)
The condition that y E V(a,b) and (VI.2.12) imply that ca is arbitrary and c2 =
0. 0
3. EIGENVALUE PROBLEMS 153
Remark VI-2-8. The materials of this section are also found in (CL, Chapters 7
and 11] and [Har2, Chapter XI].
These two solutions 6(x, A) and tp(x, A) are analytic in A everywhere in C. Also,
they are linearly independent if and only if
or, equivalently,
+'(a, A) cos a - p(a)0'(a, A) sin a = 0.
154 VI. BOUNDARY-VALUE PROBLEMS
z(b) cos - P(b) z'(b) sin,3 = (,u(b) cos$ - P(b) (b) sin 0) ° 0(Td # 0.
Ia a P(A)W w
0
operator C[y] = ((x)) + u(x)y and a vector space U(a, b) over the complex
number field C by
where C2(a, b; C) denotes the set of all complex-valued functions which are twice
continuously differentiable on the interval I(a, b) = {x : a < x < b}. Also, define an
inner product (f, g) for two complex-valued continuous functions f and g on 2(a, b)
by
b
(f, 9) = f (09(041
Jn
(f,'CWi) = 1
a
f \p(e) <J + u(C)9(C)] dC
Also, since
in the form
p(x) = (1 - p(x)g(x))rsin(O)cos(9)
and
by setting
!
(3) 01(x) (j = 1,2) are solutions of the differential equations
respectively.
Then,
02(x) > 61(x) on I(a,b) if 02(a) > 01(a).
Proof.
Set
9(x) - 1 stn2(02) - sin2(0)
(1x,61,62()1p() l
/ l 02 01
1 WS2(02)
h(x,02) _ (92(x) - 91(x))Sin2(62) + 1
P2(x) P1 (x)
Then, the function w(x) = 02(x) - 01(x) satisfies the differential equation
dw
- A(x,01(x),02(x)) w = h(x,02(x)) > 0.
dx -
Therefore,
w(x)eXP
L-J a
L I
r
w(a) + x h(s,02(s))eXP [- f ds > 0
Ja
on T(a, b). 0
Remark VI-3-9. The proof given above also showed that if pi (x) > p2(x) > 0 on
Z(a, b), then we obtain
(a) 02(x) > 61(x) on I(a,b) if 02(a) > 01(a),
02(x) > 01(x) for a < x < b if 02(a) > 01(a) and g2(x) > g1(x) on a < x < b,
( y) in the case when g2(x) > g1(x) on a < x < b, if 01(a) = 0 and 61(b) = 7r and
if 0<02(a)<rr,then 02(b)>a.
Also, y=rsin(0)>0 if and only if 0<0<r(mod 2rr)and p(x) = 1 > 0 at
0 = 0 (mod jr). Theorem VI-1-1 follows from (y).
Setting 9(x,,\) = u(x) - A, apply Lemma VI-3-8 to problem (EP).
Lemma VI-3-10. Let 0(x, A) be the unique solution to the initial-value problem
Step 1. To prove (i), apply Lemma VI-3-8 to pi(x) = 12(x) = p(x), 91(x) =
g(x, A1), and 92(x) = g(x, A2). If A2 < A1, then g2(x) > 91(x) on Z(a, b). Hence,
9(x, A2) > 9(x, A1) on a < x < b, since 9(a, A2) = 9(a, Al) = a.
Step 2. To prove (ii), choosing a real number m and a positive number P so that
u(x) > m and p(x) < P on I(a, b), determine 0o(x, A) by the initial-value problem
Then, since g(x, A) > m -A and P > p(x) > 0 for x E I(a, b), -oo < A < +oo
and a > 0, it follows from Lemma VI-3-8 that
9(c, A) > 9o(c, A) for a<c<b and - oo < A </+oo.
m-A
tan(8o(x,A)) = 0 at (x - a) = na for n=1,2....
P
and that 9o(x, A) is strictly decreasing with respect to A if x > a (cf. Step 1).
Furthermore, since tan(9o(x, m)) = x a, we must have 0 < 9o(x, m) < 2 for
P
a < x. Therefore,
9 m-P
()2) = ntr for n = 1, 2, ... .
Hence,
r
9 I x, m -
(.)2) > na for n = 1, 2, ... .
x-a
acc
Thus, we conclude that lim 9(c, A) = +oa if c > a.
Step 3. To prove (iii, first notice that 0(c, A) > 0 for -oo < A < +oo, since
P(IX)
0(a, A) = a > 0 and = > 0 if 0 = 0. Choose a positive number M so that
2(e)
lu(x) sin2(9) + 1 I <M on I(a, b).
Ax)
si n
Then,
dO < M - A sin2(0) on Z(a, b).
3. EIGENVALUE PROBLEMS 159
For any positive number 6 such that 0 < a < r - 6, fix another positive number
A(6) so that
M - A sin2(6) < -
c-10a
for --
6<0< r - b and A > A(6).
-
This implies that dO < 0 if 9 = 6. Hence, 9(x, A) > 6 for a < x < c if 9(c, A) > 6,
whereas9(c,A) <a-10<r-10<0if 9(x, A)>6fora<x<c. Thisisa
contradiction. Therefore, 0 < 9(c, A) < 6 for A > A(6). Thus, we conclude that
lim 9(c, A) = 0. 0
+00
Now, using Lemma VI-3-10, we prove the following theorem concerning problem
(EP).
Theorem VI-3-11. Assume that u(x) is continuous on the interval I(a.b) = (x:
a < x < b} and that p(x) is continuously differentiable and positive on I(a,b).
Then,
(1) problem (EP) has infinitely many eigenvalues:
AI >A2>...>An >...,
(2) lira An = -oo,
n
(3) every eigenfunction Q,,(x) corresponding to the eigenvalue An has exactly n-1
zeros on the open interval a < x < b.
Proof.
Assume without any loss of generality that 0 < a < r and 0 < 0 < r. Define
9(x, A) by the initial-value problem
Then,
d
(VI.3.14) + u(x)y = Ay
ds (P(x)±y)
d.T
that satisfies the condition y(a)cos (a) - p(a)LY(a)sin (a) = 0. Note that from
(VI.3.11), (VI.3.12), and (VI.3.13), it follows that p(x) A) = r(x, A) cos(9)(x, A).
160 VI. BOUNDARY-VALUE PROBLEMS
The eigenvalues of (EP) are determined by the condition 0(b,,\) = f3 ( mod sr). Since
8(b, A) is strictly decreasing as A +oo and since 8(b, A) takes all positive values
(cf. Lemma VI-3-10), the eigenvalues A. are determined by
Observe that
(I) ¢(x, A) = 0 if and only if 8(x, A) = 0 (mod a),
8=0(mod
(II) 2 =p(x) >0if a).
This implies that 4(x, A) has exactly k zeros on the open interval a < x < b if and
only if
kar < 0(b,,\) < (k + 1)7r.
Observe also that
(n - 1)-,r < ,3 + (n - I)a < nn.
Hence, the eigenfunction 45(x, has n - I zeros on the open interval
a<x<b. 0
Observation VI-3-12. As shown earlier, if u(x) > m and p(x) < P on Z(a.b),
then
na
P > n'rr for n = 1, 2, ... .
b-a
This implies that
/
(VI.3.15) an>m - P( bn,)z
- a
n = 1, 2,....
Observation VI-3-13. If u(x) < p and p(x) > p > 0 on I(a.b), determine 81(x, A)
by the initial-value problem
Then, since u(x) - A < p - A and p(x) > p > 0 for x E T(a, b), -oo < A < +oo,
and a < s, it follows from Lemma VI-3-8 that
0(c,,\) < Oi(c,A) for a<c<b and -00 < A < +oo.
91 (X, (-na \2
p xa I = (n + 1)ir for n = 1, 2, ... .
Hence,
P ( nTr )2)
< (n + 1);r for n = 1, 2,...
x-a
then Al < Al and A2 < A2. From (VI.3.15) and (VI.3.16) we derive the following
result concerning distribution of eigenvalues of (EP).
Theorem VI-3-14. For n > 3, eigenvalues An satisfy the following estimates:
and
2 2
rr 1 1
P(b- a) < IµI + 4PCb-a) C1
n2 + n2 n n
_An>
if n2 -P Wa
where in, p, P and p are real numbers such that
P > p(x) > p > 0 and p > u(x) > m for x E Z(a, b).
162 VI. BOUNDARY-VALUE PROBLEMS
Remark VI-3-15. Lemma VI-3-8, Lemma VI-10, and Theorem VI-3-11 are also
found in (CL, §§1 and 2 of Chapter 8] and [Hart, Chapter XI].
space V(a, b) over the real number field R in the same way as in §VI-2. Also, as
in §VI-2, define the inner product (f, g) for two real-valued continuous functions f
b
and g on Z(a, b) _ {x : a < x < b} by (f, g) = It is known that the
J
operator L has the following property:
Let r)1(x),-.
*2(x), . , be the eigenfunctions corresponding to Al, A2,...
such that 11,11 = 1 (n = 1, 2, ... ). Theorem VI-4-1 implies that (' h,''m) = 0 if
h # k. From these properties of the eigenfunctions rjn, we obtain
k k
11f
- >(f,vh)T1hII2 = Ilfll2 - E(f.rlh)2 > 0
h=1 h=1
and hence
+oo
E(f,7,lh)2
< IIffl2 (the Bessel inequality)
h=1
for any continuous function f on Z(a, b).
In this section, we explain the generalized Fourier expansion of a function f (x)
in terms of the orthonormal sequence {rl (x) : n = 1,2,... }. As a preparation, we
prove the following theorem.
4. EIGENFUNCTION EXPANSIONS 163
Theorem VI-4-2. Let µo not be an eigenvalue of (EP) and let f (x) be a contin-
uous function on Z(a, b). Then, the series
(f, 1h)t1h(x)
h=1 Ah - Jb
for any continuous function f(x) on 1(a, b). Since Go[g[f]l = f, we obtain
for any continuous functions f (z) and g(x) on Z(a, b). Also, since Go[rlh1 = c[nh1
Ahr1hUO
A01% = (Ah - l'o)llh, we obtain 9['7h] = . Hence,
(f, rlh)ghI
lz lz
< K (f, '1017h =K (f, ih)2,
h=l, h=11
where K is a positive constant such that J!G(x, ) Il < K on 1(a, b). Finally,
I 12
Theorem VI-4-3. For any continuous function f (x) on Z(a, b), the sum
+00
(h,nh) is equal to 9[f].
Proof
for every continuous function f (x) on Z(a, b). We prove (VIA.1) in four steps.
Without any loss of generality, we assume that Ah - uo < 0 for h > t . Also, note
that
(VI.4.2) (f,9e[9]) = (94f1,9)
for any real-valued continuous functions f and 9 on Z(a, b).
Step 1. It can be shown easily that Gr[>jh] = chrjh, where
for h = 1,2,... ,e,
Ch =
for h > t.
Suppose that 9t[f] 0 and IIfIII = 1, and set g = Gt[f] Then, IIGe[f]II -
IIGt[f]II'
sup I (Gt[w], w)I. This shows that sup 119t[f]II sup I (Gt[f]+f )I Thus, the
IIwIl=1 11!11=1 I1I11=1
proof of (VI.4.3) is completed.
Step 3. In this step, we prove that - sup I (Gt If], f) I is an eigenvalue of Gt. To
IlfI1=1
do this, set c = sup I(Ge[f], f)I. Note first that c > I(Gt[ne+I],ne+i)I = Ice+II >
Ill 11=1
0. Also, note that c = sup (911f], f) or -c = inf (Gt[f], f). Suppose that
Iif11=1 [if U=1
c = sup (9t If], f ). Then, there exists a sequence f (j = 1, 2.... ) of continuous
11111=1
functions on T(a, b) such that lim (Gt[ fi], fi) = c and IIf, II = 1(j = 1, 2, . . . )
Since {Gt[ fj] : j = 1,2.... } is bounded and equicontinuous on T(a, b), assume
without any loss of generality that lim Gt[ f)] = g E G(a, b) uniformly on T(a, b).
-.+00
Let us look at
II2
(VI-4.4) IIGt[fjI - cf, = I1Gt[fj]II2 + c2 - 2c(Gt[f,], f,) <- 2c2 - 2c(Ge[fjl , fi).
+00 +C0
f= (9+17 h)17h = 1: (GO[J]+nh)nh
h=1 Ah - {b h=1 Ah - 110
+00 +00
(1 £ofrlh])r)h
_ ah _ _ >(f, rlh)rlh
h=1 h=1
166 VI. BOUNDARY-VALUE PROBLEMS
For every continuous function f on 1(a, b) and a positive number e, there exists
an f, E V(a, b) such that 11f - fe II <_ e. Since
t t t
f - (f, 77h)rh = f -
/
ff + fe - 1: (ff,17h)nh + 1: (ff - f, 17077h,
h=1 h=1 h=1
we obtain lim I f -
t-.+00
(f , 7nh )T7h = 0. Therefore, f = 0 if (f, Tjh) = 0 for h =
l
h=1
1, 2,.... This proves the following theorem.
Theorem VI-4-5. If a continuous function f on 1(a, b) satisfies the condition
U, 17h) = 0 f o r h = 1, 2, ... , then f is identically equal to zero on the interval
1(a, b).
Furthermore, this, in turn, implies that if f (x) and g(x) are continuous on 1(a, b),
then
h=1
tions. This yields the Fourier sine series >bn sin(nx) of a function f (x), where
n=1
bn =
ao f/ f (x) sin(nx)ds. By virtue of Theorem VI-4-4, this series converges
uniformly to f (x) on 1(0, ir) if f (x) is twice continuously differentiable on 1(0, ir)
and f (0) = 0 and f (a) = 0.
Example VI-4-9. If f (x) is twice continuously differentiable on Z(-ir, ir), f (-vr) _
f (ir), and f'(1r) = f'(-7r), set fe(x) = 2 if (x)+f (-x)] and fo(x) [f (x)-f (-x))
4. EIGENFUNCTION EXPANSIONS 167
Then, fe satisfies the conditions of Example VI-4-7, while fo(x) satisfies the condi-
tions of Example VI-4-8. Thus, we obtain the Fourier series
cc
ao + [an cos(nx) + bn sin(nx)],
n=1
r
where an = * f (x) oos(nx)ds and bn = ! J f (x) sin(nx)ds. The Fourier
7r J
series converges uniformly to f (x) on Z(-7r, 7r).
Remark VI-4-10. The Fourier series of Example VI-4-9 can be constructed also
from the eigenvalue problem L2 = Ay, y(-7r) = y(7r), y'(-7r) = y'(ir). Including
this eigenvalue problem, more general cases are systematically explained in [CL,
Chapters 7 and 11]. For the uniform convergence of the Fourier series, it is not
necessary to assume that f (x) be twice continuously differentiable. For example,
it suffices to assume that f (x) is continuous and f'(x) is piecewise continuous on
1(-7r, 7r), and f (7r) = f(-7r). For those informations, see [Z].
Remark VI-4-11. The results in §§V1-2, VI-3, and VI-4 can be extended to the
case when p(x) is continuous on 1(a, b), p(x) > 0 on a < x < b, and p(a)p(b) =
0 under some suitable assumptions and with some suitable boundary conditions.
Although we do not go into these cases in this book, the following example illustrates
such a case.
Example VI-4-12. Let us consider the boundary-value problem
d f_dy . _
(VI.4.5)
y(x) is bounded in a neighborhood of x = 0, and y(l) = 0,
where u(x) and f (x) are real-valued and continuous on the interval 1(0, 1).
Step 1. Consider the linear homogeneous equation
(X2idv)
(VI.4.6) + u(x)v = 0
d
on the interval 1(0,1). Since 1 and logs form a fundamental set of solutions of
the differential equation ± (xdx = 0, a general solution of (VI.4.6) can be
constructed by solving the intergral)equation
where W(x) is the Wronskian of 10,p) and xW(x) = 0(1). Then, G(x,t) is
Green's function of problem (VI.4.5). The unique solution of (VI.4.5) is given
i
by y(x) = 1
positive number M is sufficiently large, the solution y = e-'(= for x < -M becomes
a linear combination a(()e-'t=+b(()e't= of two solutions a-K= and eK' for x > M.
The main problems are (i) the properties of a(() and b(() as functions of ( and (ii)
construction of u(x) for given data {a((), b(()). Keeping this introduction in mind,
let us consider a differential operator
(VI.5.1) G = - D2 + u(x),
where D = and u(x) is real-valued and belongs to C°°(-oo, +oo) such that
aj +o°
(VI.5.2) (1 + jxj)lu(x)jdx < +oc.
J
We study the differential equation
(VI.5.3) Gy = (2y,
where (is a complex parameter. First, we construct two basic solutions which are
called the Jost solutions of (VI.5.3).
Theorem VI-5-1. There exist two solutions f+(x, () and f_ (x, () of (VI.5.S) such
that
(i) ft are continuous for
(VI.5.4) -oc < x < +oo, 0,
(ii) ft are analytic in ( for £ (() > 0,
(iii)
e:
°
jf+(x, () - e'`=I <_ C(x)P(x) 1 +- 1(I
e-K=I e"x
f- (x, () - < C(-x)P(x)1 +
I(1
for (VI.5.4),, where q = Q3'((). C(x) is non-negative and nonincmasing,
t00
P(x) = I (1 + IrI)Iu(r)Idr, and O(x) = (1 + Irj)ju(r)jdr.
J o0
Proof.
Step 1. Construction of f+: Solve the integral equation
sin(((s - r)) u(r)y(r,
y(x, () = e"z ()dr
by setting
yo(x,() = eK=,
+00 sin((((- r)) u(r)ym(r,
ym+1(x, () ()dr (m > 0),
x
and
00
f+(x, () = E Ym(x, ()
m-o
170 VI. BOUNDARY-VALUE PROBLEMS
Step 2. Proof of (i) and (ii): To prove (i) and (ii), it suffices to prove that if
then
for (V1.5.4).
Inequality (VI.5.6) is true for m = 0. Assume that (VI.5.6) is true for an m;
then,
r+ Isin(((x - r)) I
Iym+i(x,C)J < mi
Z
Note that
sin(((x - r)) a z-r
2i(
(1 - e-2i(z-7)() = eft(Z-r)
1 e-2it=dz.
Hence,
sln(((x - r)) I
< e-q(z-T)(r - x) for r1 > 0.
C
Therefore,
+00
I ym+1(x,C)I <
(C(x)me-nx
f (r - x)tu(r)IP(r)mdr.
z
Since
j +00
(T - x)Iu(T)IP(r)mdr < C(x)f.
+00 (1
C(x)P(x)m+
+ I rI )I u(r)I p(r)mdr
m+1
we obtain
(C(x)P(x))m+ie-nZ
+00
= ( +
J
< 2Ixl1 I u(r){P(r)mdr + 2fz TI u(r)I P(T)mdT
Jz z z z
for x < 0.
5. JOST SOLUTIONS 171
'o I
I f+(x, 0- I <- E n,, (O(x)v(x))me-''z
l
m=1
(VI.5.7) 00
< C(x)p(x) e-"z
I i (C(x)p(x))m-1 I
m=1
follows from (VI.5.6). However, this estimate is not enough to prove (iii). So, let
us derive another estimate for large I(I. To do this, we shall prove that
or(x) = fx
" iu(r)jdr.
Inequality (VI.5.8) is true for m = 0. Assume that (VI.5.8) is true for an m. Then,
+ao
Iym+1(x, ()1 - M! I(I 1 Jz
I sin(((x - r))IIu(r)I e-"'a(r)mdT.
Note that
Hence,
+00 1 m+1
1 e-'7= a-'
Iym+1(x,()I m! iu(r)io(r)mdr = (m+ 1)! (!!-(X) .
KIm+1I
J. 1(i}
This establishes (VI.5.8). Therefore,
v(x)
(VI.5.9) If+(x, () - ei(zl < = -))M-,l
(i
1
M! OWC-1 I e-qz.
The proof of Theorem VI-5-1 for f+ (x, () can be completed by using (VI.5.7) and
(VI.5.9).
For f-(x,(), change x by -x to derive
This implies that I f+, f+} is linearly independent if 36 0. This, in turn, implies
that the solution f_ is a linear combination of I f+, 1+}. Set
(VI.6.1) f-(x,() = a(()f+(x,() + b(f)f+(x,()
It is easy to see that
i i f+(x+ f-(x,()
(VI.6.2) a(() =
and
(VI.6.3) b(() = Wif+,f-j = I f+(x,() f .(x,()
I.
(VI.6.6) a(iry) = 0.
This implies that all zeros of a(() for 3(() > 0 are purely imaginary.
Under Assumption VI-6-1, ft(x, () are analytic for 9'(() > - 2. Hence, a(()
has only a finite number of zeros for 3(() > 0 (cf. (VI.6.4)). Let S = irlj 0 =
1, 2, ... , N) be the zeros of a(() for (() > 0. Then, f+(x, ir73) are real-valued and
f+(x, ins) E L2(-oo, +oo). Set
imply that
d d
d w(f+(, f-1 = 2-f+f- and W(f+, f-cl = -2f+f_
Since a(() = 0, there exists a constant d(() such that f_(x,() = d(()f+(x,C).
Therefore,
W(f+t, f-) =
-2S+00
Thus, we obtain
+oo +oo f+dr = -id(()
da(()
-i f+f-dr = -id(()f # 0.
d( 00 oo Ci
Observation VI-6-3. The quantities a(() and b({) satisfy the following relation:
Proof.
From f_(x, -{), it follows that
for %() > 0 shows that the quantities q, and r({) determine a(().
Proof
loBja(()f2 = log(Ta(f)22/
Definition V1-6-5. The set {r((), (n,,n2,... ,nN), (cl,c2,... ,CN)} is called the
scattering data associated with the potential u(x).
Observation VI-7-2. Set gJ (x) = e-" cj f+(x, iii ). Then, from the fact that
e_,<=
f+(x, () - I as I(I - +oo, it follows that
Observation VI-7-3. If we set F(x,() = e-'t= f+(x, (), then -F" - 2i(F'+uF =
'` 91 +(x) , it follows that
0. Since F(x, ) = 1 - i-1
N
(VI.7.3) u(x) = 2E gj,(x).
.7_1
Observation VI-7-4. Let us solve (VI.7.2) for the g,(x). First, set
hi = gie'''= (j = 1,2,... ,N).
Then, (VI.7.2) becomes
c
(VI.7.4) ht + h = cte-''i: (t
J=1 rJt + I?j
Write the coefficient matrix of (VI.7.4) in the form IN + C(x), where IN is the
N x N identity matrix. Since
N +op N 2
E 717t = E 7 e-n,,= dx
)t°1 'b + 171 L i p
and 77, (j = 1, 2,... , N) are distinct, the matrix IN + C(x) is invertible for -oo <
x < +oo. Set L(x) = det(IN + C(x)). Then, manipulating with Cramer's rule, we
dL
can write g, (j = 1, 2,... , N) and in the following forms:
da
and
N
dL(x)
(VI.7.6)
dx
J=1
This implies that u(x) is a rational function of exponential functions and satisfies
Assumption VI-6-1 of §VI-6. To see this, the following remarks are also useful:
(a) we have the identity
N N
gg(x) 2 ?1 e2n'x9s(x)2,
j=1 j=1 i
from (VI.7.2). Multiplying both sides by ge, adding them up over f, and interchang-
ing the orders of summation, we obtain (a). To show (b), calculate the inverse of
the coefficient matrix of (VI.7.2). Using Cramer's rule on (VI.7.4), it can be shown
that he(x) -, 0 exponentially as x -+ oo. Also, (b) implies that e2°,, xg,(x)2 is expo-
nentially small as x -oo. Therefore, (a) implies that u(x) satisfies Assumption
VI-6.1.
e2+
gz 1
Example VI-7-5. In the case N = 1, if we determine g(x) by g(x) _
g /
1, then u(x) = 2g'(x). Since g(x) = c' we obtain u(x) = -- e2'rzg(x)2.
C
Qx 2c e+7x +
Also, FF17e-nx g(x) = 1 implies that g(x) = 2 e-n=
(211
In
xsech(11(x + p)), where p = `c , and, hence,
211
by calculating (VI.7.7).
178 VI. BOUNDARY-VALUE PROBLEMS
and, hence,
N N
Eata(x)9j'(x) + 27t 11 - 1 g, (x) =0 (t = 1,2,... ,N).
1=1 i=1 7e+7)
8. CONSTRUCTION OF A POTENTIAL FOR A GIVEN DATA 179
N N
E ajj(x){g,(x) + 2gegj(x)} - 237tE g, (x) = 0
j=1 j=1 ge +T7j
(e = 1, 2, ... , N)
or
N N N
g,
Eatj(x){gf(x) + 2gtgj(x)} - 2>9,(x)+ 2E 711+17., 92(x) = 0
3=1 j=1 j=1
(e = 1, 2, ... , N).
N 2o[s
E ata (x){gj"(x) + 2r7eg, (x) } + 2ge (9e(x) + 2gtgt(x)) - u(x)
,=1
N
+ 2E gj 9'(x) = 0 (e = 1,2,... N)
3=1 ge+g,
or
N N
Eat,j(x){g,"(x) + 217eg,(x)) - Eaea(x)u(x)9J(x)
,=1 f=1
N e2nas
+ aea(x)2%g,'(x) + 2171 C27jt
ct ) = 0 (e = 1,2,N).
... ,9e(x)
,=1
Thus, we derive
N
F'ae,,(x){9,(x) + 217,g,(x) - u(x)g,(x)}
j=1
N
2171 E atj (x)9, (x) + 217t ct 91(x) =0 (t = 1, 2, ... , N),
,=1
N
F'at,1(x){9;'(x) + 2s7,g,,(x) - u(x)g,(x)} = 0 (e= 1,2,... ,N).
,=1
-iJ=1
-i N
g1'
g;,
( ,=1
+ 2i(g - ttgJ
+i
+ 2rligl' - ugJ = 0.
- 2
N
9J
E
j=1
(+ tnJ
Observation VI-8-3. Since
N x
g., (x) en,"
f+(x,it7t) = e-fez 1 -
711+17.;
_ -gt(x)
ct
E L2(-oo,+oo),
J=1
N numbers -1l,2 (j = 1, 2,... , N) are eigenvalues.
N
Observation VI-8-4. Set a (cf. (VI.6.9) with r 0) and
J=1
N
g3 (x)
f-(x,() = a(()f+(x, -() =
a(S)e-Cz
1 + iE J=1(-ig,
E gJ (-oo) = 1 (e = 1, 2, ... , N)
J=1171+17J
(cf. (VI.7.2)), we obtain
j=1(+nJ
In fact, this follows from the fact that both sides of (VI.8.4) are rational functions
in ( with the same zeros, the same poles, and the same limits as ( , oo.
Observation VI-8-5. The functions f±(.x, () are the Jost solutions of Ly = (2y.
Proof.
Note first that lim f+(x. ()e-'t= = 1 (cf. (VI.7.1)). Also, we have
s+oo
Observation VI-8-6. Note that f_ (x, ) = a(() f+(x, -{) = f+(x, ) for ( =
C real. This means that 0. Hence, r(C) = 0. Thus, we conclude that u(x) is
reflectionless.
Remark VI-8-7. For the general r((), the potential u(x) can be constructed by
solving the integral equation of Gel'fand-Levitan:
+00
K(x, r) + F(x + r) + + F(x + r + s)K(x, s)ds = 0,
0
where
F(x) = 1R J +oo
2cL` cle-2n,=
J=1
(x,0).
u(x) 8x
If we set r({) = 0 in this integral equation, we can derive (VI.7.2) and (VI.7.3).
Details are left to the reader as exercises (cf. (Ge1LJ ).
Furthermore,
N
f+(x, -() = e`= 1 - g1 (r)
-( + trj7
is also a solution of
dX2
- (u(x) - A)y = 0, where J = (2.
Therefore,
N
(VI.9.1) P(x, A) = 11 ! 1+11- f+(T,C)f+(x, -()
3=1
A
)
satisfies the differential equation
3
ddx
(E)
2 dz3
+ 2(u(x) - A) dP + )P = 0.
182 VI. BOUNDARY-VALUE PROBLEMS
Let
+oo
(S) P = E an (PO 3 0)
n=0
be a formal solution in powers of A-1 of differential equation (E). Then,
and, hence,
A(P2)' = I - PZ it + (4)2 + u(x)P2 }, .
Therefore,
(i) the coefficients pn can be determined successively by
and
G(u, A) = f+(x, C)f+(x, where A=C2.
Hence, ao = 1 and
N +
1 + EN =
2
3=1 1 =1
(VI.10.1) 2 + (A - u(x))y = 0
under the assumption that
(I) u(x) is continuous for -co < x < +oo,
(II) u(x) is periodic of period 1, i.e., u(x + e) = u(x) for -oo < x < +oo.
The period a is a positive number and A is a real parameter. Denote by 01(x, A)
and 02(x, A) the two linearly independent solutions of (VI.10.1) such that
1(o
(VI.10.2) 01(0,A) = 1, A) = 0, 02 (0, A) = 0, d2(0,A) = 1.
Set
01(e, A) 02(t, A)
.D(A)
= O1 (t"\) (e A)
The two eigenvalues Z1(A) and Z2 (A) of the matrix are the multipliers of the
periodic system
d
(VI.10.3)
dx [Y2J = [u(--O-A 0, [y21
184 VI. BOUNDARY-VALUE PROBLEMS
(cf. Definition IV-4-5). It is easy to show that det [1(A)] = 1 for -oo < A <
+oo. Hence, the two multipliers Z1(A) and Z2(A) are determined by the equation
Z2 - f (,\)Z + 1 = 0, where
2(t,A)
(VI.10.4) f(A) =
Note that f (A) is continuous for -oo < A < oo (cf. Theorem 11-1-2). We derive
first the following conclusion.
Lemma VI-10-1. The two multipliers Z1(A) and Z2(A) of system (VI.10.3) are
f(,\) + f (A)2 - 4, ,
2l
2 l f (A) - f (A)2 - 4, ,
I Zt (A) I
=1 and I Z2(A) 1 = 1,
(iii) if f (A) = 2, then Z1(A) = Z2(A) = 1.
(iv) if f(A) _ -2, then Z1(A) = Z2(A) = -1.
ICCl2
Observation VI-10-2. If f (A) = 2, let be an eigenvector of 4;(A) associated
J
with the eigenvalue 1. This means that cl and c2 are two real numbers not both
zero and that
>2 if n is even,
f (µ") { < -2 if n is odd
(cf. (VI.10.7)). Here, use was made of the fact that jaj+ 1a1,
ifa76 0.
Observation VI-10-5. If
(VI.10.8) A < min{u(x) : -oo < x < +oo},
then f (A) > 2. In fact, dX21 (x, A) = (u(x) - A)01 (x, A) > 0 as long as 01(x, A) > 0
dol
and, hence, (x, A) > 0. Thus, we obtain s1(£, A) > 1. Similarly, 2 (P, A) > 1
if (VI.10.8) is satisfied. In this way, a rough picture of the graph of the function
f (A) is obtained (cf. Figure 1).
At A2 44
3 A3= f=2
T
3
I l4
-F
W lA
Ul U2 U3=N3U4 f= -2
u5 U6
FIGURE 1.
Actually, we can prove the following theorem.
186 VI. BOUNDARY-VALUE PROBLEMS
Theorem VI-10-6. Assuming that u(x) is continuous and periodic of period t > 0
on the entire real line R, consider three boundary-value problems:
i
w(0, A) = (0, )L),
d2w
+ (A - u(x)) w + O(x, A) = 0,
dx2
i (0,,\) =
TA_ ( dx
(cf. §lI-2). Therefore, using the variation of parameters method, we obtain
(VI.10.11) _
-1(x, A) _ (x, A) fT 01(t, A)02(t, A)dt ¢2(x, A)
f01(t"\)2dt'
d z (x,,\)
= 01(x"\) 102(t, A)2dt - 02(X,,\) f -01(t, A) 02(t, A)dt
10. PERIODIC POTENTIALS 187
and, hence,
df
Therefore,
dA
= rr Q(01 (t, A), ¢2(t, A))dt follows from (VI. 10.4), where
0
Note that
Thus, Q(01 (X, A), ¢2(x, A)) does not change sign for 0 < x < e if If (A)f < 2. It
follows that
df(A)
<0 if A < pl and f(,\)2 = 4,
d,\
(-1)1 d('\) < 0 if pl < A < pi+1 and f(A)2 = 4.
To prove this, notice that Q is a perfect square if f(,\)2 = 4 and that 02(1,.\) # 0
if A 0 pj for every j. Hence, dA\) ¢2(e, A) < 0. Also, using Lemma VI-3-11 (with
-A instead of A), we obtain
dzf
(a)
J< 0 if f (a) = 2 and d(\) = 0,
dal df(A)
>0 if f (A) _ -2 and
d,\
= 0.
71K (A)
=± f 0
t
[c101(t, A) + c202(t, \)12 dt = 0,
where cl and c2 are some real numbers. Therefore, cl = 0 and c2 = 0, since 01 and
02 are linearly independent. This means that
d (e, a) =
I
z
{e, ) = j42(s,A)2ds,
d t
0i (s, A)2 ds,
0
d2f
2 (A) = ft f t [Ol (t, A)02(s, A) - 41(s, A)4 2(t, A)J2dtds > 0
if f (A) = -2 and df (A) = 0. Note that, if A2,,-1 = A2,,, it follows from (VI. 10.2),
(IV), and (V) that 01(x, A) and 02(x, A) are two linearly independent solutions for
problem (B). Similarly, if w2r_1 = v2,,, it follows from (VI.10.2), (IV), and (V) that
01(x, A) and 62(x, A) are two linearly independent solutions for problem (C). Thus,
(3) and (4) are verified.
Step 5. The functions (respectively -yn(x)) have an even (respectively odd)
number of zeros on the interval 0 < x < 1, since /3, (0) = Qn(1) and 7n(0) Yn(1)
As can be seen in Figure 1,
Therefore, by virtue of Theorems VI-3-11 and VI-1-1, we conclude that fit, 1 and
per have more than 2n - 1, and less than 2n + 2, zeros on the interval 0 < x < 1.
Hence, they have exactly 2n zeros there. Similarly, since
and -t2n have exactly 2n - I zeros on the interval 0 < x < e. The function
,8a does not have any zero on the interval 0 < x < t since Ao < 111. Thus, (5), (6),
and (7) are verified. Finally, (2) follows from (VI.10.9), (VI), and (VII).
Definition VI-10-7.
(I) The set {A f(A)2 < 4} is called the stability region of the differential
:
equation (VI.10.1).
(11) The set {A : f(A)2 > 4} is called the instability region of the differential
equation (6210.1).
and
( sin(v1_T_-ax)
if A>a,
A-a
02(x, A) = x if A =a,
sinh(x) if A < a.
Therefore,
Thus, we conclude that in this case, f (A)2 -4 has only one simple zero a (cf. Figure
2).
FicURE 2.
The materials in this section are also found in [CL, Chapter 8j.
EXERCISES VI
day
(E)
&2
- u(x)y = 0
has at most a finite number of zeros on Z0,
(2) the differential equation (E) has a nontrivial solution 1)(x) such that
hm q(x) = 0.
Hint.
(1) Note that if y(xo) > 0, then y"(xo) > 0. Hence, y(x) > 0 for x > x0 if
1/(xo) > 0.
(2) It is sufficient to find a solution 0(x) such that 0(x) > 0 and 0'(x) < 0 for
x>xo.
EXERCISES VI 191
d2y
+ u(x)y = Ay, y(a) = 0, y '(b) = 0,
where u(x) is real-valued and continuous on the interval a < x < b. Show that
there exists a positive number K such that
2
n + ir
K for n = 1, 2, 3, ... .
n2 b-a n
VI-4. Assuming that u(x) is real-valued and continuous on the interval 0 < x <
+oo and that lim u(x) = +oo, consider the eigenvalue-problem
r+oo
+oo
(f,C(f)) = j>n(f,17n)2
n=1
b
(iv) nx)dx = 0 if n 96 m.
J0
VI-9. Show that the Legendre polynomials
2"n1 d"
Pn(x) !dx"((x2 - 1)'J (n = 0,1,2,...)
=
(ii) 1 Pn(x)Pm(x)dx = 0 if n 0 m,
J
(iii) j P n (x)2dx = 1 (n = 0,1, 2, ... ),
n z
(vi) if f (x) is real-valued and continuous on the interval jxj < 1, then
2
1 N
lim f
N-+oo 1
f (x) - E (n+ Z) (f, PP)Pn(x)) dx = 0,
n=O
f (x)PP(x)dx,
jxj < 1 if f, f', and f" are continuous on the interval jxj < 1.
Hint. See Exercise V-13. Also, note that if f (x) is continuous on the interval
jxj < 1, then f (x) can be approximated on this interval uniformly by a polynomial
in x. To prove (vii), construct the Green function G(x,t) for the boundary-value
problem
((I -x2)d/ +aoy=f(W),
y(x) is bounded in the neighborhood of x = ±1,
1 1
where ao is not a non-negative integer. Show that f f G(x, )2dx< < +oo.
J! 1J 1
Then, we can use a method similar to that of §VI-4.
VI-10. Assume that (1) p(x) and p'(x) are continuous on an interval Zo(a,b) _
{x : a < x < b}, (2) p(x) > 0 on Zo, and (3) u(x, A) is a real-valued and continuous
function of (x, A) on the region Zo x ll = {(x, A) : x E Zo, A E It} such that
lim u(x, A) = boo uniformly for x E Zo. Assume also that u(x, A) is strictly
A-too
decreasing in A E R for each fixed x on I. Denote by O(x, A) the unique solution
of the intial-value problem
Show that there exists a sequence {pn : n = 0, 1, 2, ... } of real numbers such that
(i) pn < 1An-1 (n = 1, 2, ... ), and lim pn = -oo,
n .+oo
(ii) 4(b, pn) = 0 (n = 0,1, 2, ... ),
(iii) ¢(x, .\) 0 on a < x < b for A > po, and q(x, A) (n > 1) has n simple zeros
,-
on a<x<bforpn<\<,un-1,
(iv) strictly decreases from +oo to -oo as A decreases from pn_1
to An-
VI-11. Assume that p(x) and u(x) are real-valued and continuous on the interval
1(0,1) = {x : 0 <- x < 1} and that p(x) is also continuous on 7(0,1), p(x) > 0
for 0 < x < 1, p(0) = 0, and p'(0) < 0. Show that the differential equation
194 VI. BOUNDARY-VALUE PROBLEMS
dx
(px ) d
dx)
+u(x)y = 0 has a fundamental set {0, ii} of solutions on the interval
0 < x < 1 such that
(i) slim 4(x) = 1 and Zli o p(x)O'(x) = 0,
for 0<x<
for 2<x 1.
Denote by 01(x, A) and 42(x, A) the two unique solutions of the differential equation
y
+ (A - u(x))y = 0 satisfying the initial conditions 4i(0, A) = 1, X1(0, A) = 0,
VI-13. For the scattering data {r({) = 0, (1,2,3), (1,1,1)}, find the potential u(x)
and the Jost solutions ft(x,().
VI-14. Calculate the scattering data for each of u(x + 1) and u(-x), assuming
that {r({), (271, ... ,T)N), (cl, ... , cN)} are the scattering data for u(x) and that
u(x) satisfies a condition Eu(x)I < Ae-k1=I for some positive numbers A and k.
Hint. Let f f(x, () be the Jost solutions for u(x). Two quantities a(() and
are given by
a(S) = and 1
The Jost solutions for u(-x) are ff(-x, (). Hence, the scattering data for u(-x)
are
b(l)'
a(C)
(171,...,17N)I
(
1
cia<(iv71)2
,...,- 1 l}
CNa<(LT,N)21 J
Note that
f-(x,iil3) = i a<(iul,)f+(x,iu,)
EXERCISES VI 195
VI-15. Let u(x) be real-valued, continuous, and periodic of period t > 0. Also,
for every real , let O(x, {, A) be the solution of the differential equation
(Eq) 2 + (A - u(x))y = 0
Am(t) - A
( + [', , A) = Q +00 2
m=1
e
Hint.
Step 1. Let us construct O(x, , A) for negative A. To do this, change (Eq) to the
integral equation
sinh(p(x - )) + 1 r sinh(µ(x -
p p f
e-v(x-E)+G(x, such{µ x- f )) + O( A 1
, A) = e-al=-E)
lim t'(4+f,.,A) = 1.
(sinh(e))
R(A) = t
-_,
m=1 \1
196 VI. BOUNDARY-VALUE PROBLEMS
mr
where Cm =
t
Step 3. If the function V)((+1, 1, A) is entire in .\ and ''(+t, 1,,\) = O(exp(1 JaJ))
as A -+ oo, we can write 1'(( + 1, t,,\) in the following form:
lp(f + t, (, a) = c
[i(A]
where c is independent of A. Here, we used the fact that cn,J is bounded
as m +oc (cf. Theorem VI-3-14).
Step 4. Note that
_((+t,t,X) - 'c' + [0-W
_- a
(1)
R(A) ( )H cam, - A ]
c -a cm -A
This implies the uniform convergence of the infinite product on the right-hand side
Z'((+ 1, t, _ = 1 = c
of (1) for -oo < A < 0. Thus, lim
a-.-OO R(a) l
Remark. For expressions of analytic functions in infinite products, see, for exam-
ple, (Pa, pp. 490-5041.
VI-16. The functions are defined in §VI-9. Calculate G4(u) and G5(u).
I
VI-17. Use the same notations as in Theorem VI-10-6. Show that if J u(t)dt = 0,
0
ft
then Ao < 0. Also show that if / u(t)dt = 0 and = 0, then u(x) = 0 identically
Jo
on-oc<x<+00.
Hint. Since j3o(z) # 0 on -oc < x < +oo, set w(z) = !Lx). Then, w'(x) +
w(x)Z + A - u(x) = 0 (cf. 1A'IW, Theorem 4.4, p. 621).
V1-18. Set F(() = log f (() for `a( > 0, ( # 0, where
(i) f (() is continuous for QJ( > 0, ( 0,
(ii) f (() is analytic for `£( > 0,
(iii) ((f (() - 1) is bounded for !a( > 0,
(iv) f (() # 0 for !'( > 0, (0 0.
Denote by SR the semicircle {(: 1(I = R,0 < arg(< r) which is oriented counter-
clockwise, where R is a positive number. Show that
1 1 F(z) dz = f F(() (`1( > 0,1(1 < R),
(a) 2ri IR 2Ri 0 (`'(<0,1(1<R),
ASYMPTOTIC BEHAVIOR OF
SOLUTIONS OF LINEAR SYSTEMS
Definition VII-1-1. Let f (t) be a C"-valued function whose entries are continu-
ous on an interval Z = {t : to < t < +oo}. Let us denote by A the set of all real
197
198 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
Example VII-1-2.
J+ (0) = -00, A (exp[-t2j) _ -00, A (tm) = 0 (for all constants m),
{ A (exp[ort]) = a, A (exp[t2j) _ +00.
Proof.
Since there exists a positive constant K such that I f (s)I < Keia(f)+`i' for e > 0
and large values of s, it follows that
Hence,
log If(3)1
< a(f)
lim sup
t-+m s
It<5<+m
Also, for a fixed positive number a and any positive integer in, there exists a large
log m +
value of sm such that If (sm)1 and lim sm = +oo. Hence,
sm
log If(s)
lim sup
t-+oo<a<+ao s
I
> (f )
dy
(VII.2.1) = A(t)9
dt
under the assumption that the entries of the n x n matrix A(t) are continuous and
bounded on an interval T = It : to < t < +oc}. Let us start with the following
fundamental result.
Theorem VII-2-1. If y" = fi(t) is a nontrivial solution of system (VII. 2. 1) and if
f A(t)e < K on the interval I = It : to < t < +oo} for some non-negative number
K, then
(VII.2.2)
Proof.
Let the n x n invertible matrix 4P(t) be the unique solution of the initial-value
dY
problem 7 = A(t)Y, Y(to) = In, where In is the n x n identity matrix. Then,
200 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
dZ
4i(t)-1 is the unique solution of the initial-value problem = -ZA(t), Z(to) _
dt
I (cf. Lemma IV-2-4). Since 1, we obtain
1 + KJ I4(t)Idt,
to
1 +K/ I t
,I dt
for t r- T.
Therefore,
)4i(t)I < exp[K(t - to)) and I < exp[K(t - to)] for t E T
(cf. Lemma 1-1-5). Observe that fi(t) = 4D(t)j(to) and b(to) From
the definition of the norm of a matrix, it follows that
exp[-K(t - to)] < 1 (t)I < 1 (to)j exp[K(t - to)] for t E Z.
The structure of solutions of (VI1.2.1) according with their type numbers is given
in the following result.
Theorem VII-2-5. Let {(1, ¢2r ... , iin} be a fundamental set of n linearly in-
dependent solutions of system (VII.2.1). Then, the following four conditions are
mutually equivalent:
(1) for every j, the total number of those 4t such that A (y5t) = A, is h, (cf.
(VII. 2.5)),
(2) for every j, the subset {¢t : A (Qt) < A, } is a basis for V,,
Proof.
Assume that (1) is satisfied. Then, the total number of those ¢t such that
m
A (3t) < A, is Fht = ryr = dims Vj. Hence, (2) is also satisfied. Conversely,
t=j
assume that (2) is satisfied. Then, the total number of those t such that A (fit) <
Aj is equal to dime V) = -,. Hence, (1) is also satisfied (cf. (VII.2.5)).
clot = tt
202 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
with a matrix A(t) whose entries are continuous and periodic of a positive period c..'
on the entire real line R, Liapounoff's type numbers )k1, A2, ... , Am at t = +oo and
their respective multiplicities h1, h2, ... , hm are determined in the following way:
There exists an n x n matrix P(t) such that
(i) the entries of P(t) are continuous and periodic of period w,
(ii) P(t) is invertible for all t E R.
3. CALCULATION OF LIAPOUNOFF'S TYPE NUMBERS 203
(VII.2.7)
where µl, ... , Pk are distinct eigenvalues of A whose respective multiplicities are
m1, ... , mk, is the mj x mj identity matrix, and M1 is an mj x m, nilpotent
matrix (cf. (IV.1.10)).
Consider a system of the form
Assumption 1.
(i) For each j, the matrix A. has the form
Proof.
We prove this theorem in eight steps.
Step 1. Differentiating both sides of (VII.3.3), we obtain
m
(VII.3.5) = A) % + E Tj tat
t0r t=1#t
Bjt t + Tt V
Aj 4' Bjj +
h*
BjhThj i, +
t AjTjt + Bjt +
hit
BjhTht zt,
3. CALCULATION OF LIAPOUNOFF'S TYPE NUMBERS 205
2, = F BjnTnv zv
e=1 Vol v=1 h#v
t#, h#t
F t = B» + B)hTh) M)
h#j
Define the Tit by the following system of differential equations:
(VII.3.6) T (.l r e)
dtt h#t
Then,
where j t and
Wjt(r,s,T)
(VII.3.8)
= exp [(A, - At)r] exp[rEj j exp[rNj[UJt(s, T) exp[-rEtj exp[-rN,].
On the right-hand side of (VII.3.7'), choose the initial points rat in the following
way
(VII.3.9) ra t
Step 3. Let us prove the following lemma.
( +00
j`
to
if Aj > At
if A < A 1t..,. (i.e., j < t),
(ie j > t)
Lemma VII-3-2. Let a be a positive number and let f (t, s) be continuous in (t, s)
on the region Z x Z, where Z = {t : to < t < +oo}. Then,
(VII.3.10)
III ft"
t exp[-a(t - s)] f (t, s)dsl < - max l f (t, s)to
-s<t y
J
for each fixed t E T. Also, if I f (t, s)' is bounded for to < t < s < +oo, then
(VII.3.11) rt
J+00
exp[a(t - s)j f (t, s)dsl <_ 1 sup { I f (t, $)I : to < t < s < +oo}
a
I
exp[a(t - s)]ds = - 1a . O
+oo
Step 4. Let us estimate s, T) fort < s < +oo if A j > at, and for to < s < t
if A_, < at. If A, > At, r < 0, and s > t, it follows that
Irlm'+mi-2)
i1',t(r,s,T)I < Ko(l + exp 12(A, - A,) 7-1 IVit(s,T)I
for some positive number 1Co. It can be shown `easily that there exist a positive
number X and a non-negative valued function p(t) such that
I + Ir1m'*m`_2 exp [(A1 - At)r < X, ] for r < 0,
B, (s) Q(t) for t < s < +oo and all pairs (p, q),
t-+00
lim p(t) = 0.
Hence, if Aj > At, r < 0, and s > t, we obtain IW,t(r,a,T)I < /C2,3(t) {1 + ITI2}
for some positive number 1C2, where ITI = maxIT,t[. Similarly, the estimate
t#j
I Wjt(r, s, T) I < 1C219(to) { 1 + ITI2) is obtained by choosing a positive number 1C2
sufficiently large, if A., < at, r > 0, and to < s < t.
3. CALCULATION OF LIAPOUNOFF'S TYPE NUMBERS 207
IWj1(T,s,T)
- Wj1(r,s,T)I
K3$(t) {1 +ITI + ITI} IT - TI if A, > A1, T < 0, s > t,
TP1t;)t(t) =
f [ exp
i..
[(A, - A,)(t - s)j 9,1(t - s, s, Tp(s))ds,
where j 54 a and p = 1, 2, .... Suppose that
Step 8. In this final step, we prove that the bounded solution Tjt of (VII.3.7)
satisfies condition (2) of Theorem VII-3-1. It easily follows from Steps 6 and 4 that
1
if > X1.
For (j, f) such that ), < at, write the right-hand side of (VII.3.7') in the following
form:
t s)] IV1t(t - s, s,T(s))ds
Jto exp 12 (A
Q 11
< exp L2 (.1j - '\t)(t - a)] 12 (a) - at)(a - s)J 14 t(t -- s, s, T(s))ds
I 1'. a exp `
2KO(t°)
< {l + C2} exp 1(aI - AE)(t - a) J ,
At - A) 1:5
jexp
t 1(A3 - a)(t - s)] - s, s, T(s))dso
2K1\1 211-
< C2} {(to)exP {(A, - At)(t - )1 +
J
Lemma VII-3-3. If N is a nilpotent matrix, then for any positive number e such
that 0 < e < 1, there exists an invertible matrix P(e) such that
P(e)-INP(e)I < Xe
for some positive number IC independent of e.
Proof.
Assume without any loss of generality that N is in an upper-triangular form with
zeros on the main diagonal (cf. Lemma IV-1-8). Set A(e) = diag[1, e, ... , e"-1].
Then, A(e)-'NA(e) = [ ek-Jz,k ] (a shearing transformation). Hence, as 0 < e <
1 and j < k, we obtain IA(e)-1NA(e)I < e(N(. 0
Let us find Liapounoff's type numbers of system (VII.3.4), i.e.,
dzj
a. = IA + B,,(t) + z, (7 = 1,2,... ,m).
h*j
dy log 1-0(t)I
then, for every nontrivial solution ¢(t) of the system = B(t)y, lim
dt t--+oo t
p exists and p is the real part of an eigenvalue of the matrix A.
Remark VII-3-8. The conclusion of Theorem VII-3-1 still holds even if condition
(iv) of Assumption 1 of this section is replaced by IB)k(t)j < f (t) (j, k = 1, 2,... , m)
on the interval Zo, where f (t) satisfies
rP
(VII.3.14) sup(1 +p- t)-1 J f(s)ds - 0 as t +oo.
p>t t
To see this, let us assume that a positive-valued function f (t) satisfies condition
tP
(VII.3.14). Set h(t) = sup ((1 + p - t)-1 ( f (s)ds) for a fixed positive number
p>t \\ ft
t. Also, set E(t) = suph(r). Then, lim E(t) = 0, and E(t2) < E(t1) if t1 < t2.
r> t t--+oo
Now, it is sufficient to prove the following lemma.
Lemma VII-3-9. For any positive numbers t, to, and c, it holds that
lim
t-.+oo
f t e-`(t-`) f (s)ds = 0.
o
3. CALCULATION OF LIAPOUNOFF'S TYPE NUMBERS 211
Proof of (a).
for to < a < r. Suppose that there exists a positive number 6 such that
and
0(s) > (!± b E(to) for a < s < r.
Then,
E(to) + c l! + 6 E(to)(r - a) < E(a)(1 + (r - o)).
This is a contradiction. \
Proof of (b).
Set ,L(t) = f et-f (s)ds for 0 t T for a fixed T > 0. Then, "(t) _
r
cO(t) - f (t) and, hence, w(r) - ?P(t) = cJ ;(s)ds J f (s)ds for t < r < T.
r
Suppose that there exists a positive number 6 such that t
and
1l'(s) > (!+o)E(t) for t < s < r.
Then,
E(t) + c (1+ 6) E(t)(r - t) < E(t)(1 + (r - t)).
This is a contradiction. This, in turn, proves that
T
f (s)ds < C1 + c j E(t) for T > t.
J
+430
Since lim E(t) = 0, the integral + e'('-")f (s)ds exists and
t-+o ft
j+00
ec(t-e)f(s)ds < 1+ 1 f E(t).
e \ c
212 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
b(t) = q(to) cost - to) + 0'(to) sin(t - to) - J t R(s)¢(s) sin(t - s)ds.
to
Choose a positive number K so that 4(to)j + kd'(to)j < K and choose another
KQ
positive number At so that M > 1 > K. Then,
if jb(t)j < At for to < t < t1. Hence, 14(t)I < M for to < t < +oc. 0
In this section, we explain the behavior of solutions of a system of linear dif-
ferential equations under a condition similar to the Hukuhara-Nagumo condition.
Precisely speaking, we consider a system of the form
(VII.4.5) A,k(t) = al(t) -)tk(t) and D-,k(t) = R(AJk(t)) (j,k = 1,2,... ,n).
Concerning the functions A. (t) (j = 1, 2,... , n), the following is the main assump-
tion.
Assumption 3. The functions Al (t), A2(t), ... , A,, (t) are continuous on the inter-
val 4. Furthermore, for each fixed j, the set of all positive integers not greater
than n is the union of two disjoint subsets P.t and P32, where
(i) kEP) I if
lim I Djk(r)dr = -oo and Dik(r)dr < K for 0 < s < t
t-+oo 0 Li
for some positive number K,
(ii) k E Pj2 if
`t
f/a
Dik(r)dr < K for s > t > 0
(2) t1 1 Q(t)
(3) the transformation
(VII.4.6) y' = V. + Q(t)] E
changes system (VII.4.2) to
d"
(VII.4.7) dt= A(t)i
on the interval Z0, where I is the n x n identity matrix.
Prool
We prove this theorem in six steps.
214 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
where
4?(t, s) = diag(Fl (t, s), F2(t,1s), ... , Fn(t, s)],
Step 3. Letting q,k(t) and rik(t) be the entries on the j-th row and the k-th
column of Q(t) and R(t), respectively, write the integral equation (VII.4.10) in the
form
[ft
(VII.4.10') ggk(t) =
J
exp Aik(r)drr)k(s) + F ds,
11
A=1
4. A DIAGONALIZATION THEOREM 215
where j, k = 1, 2,... , n and the .1jk(t) are defined by (VII.4.5). Note that
t
( J(Ajk(r)dr}
t
lexp I
I = exp I J Djk(T)dr] < eK
ll a
if
0< s< t for k E Pj1 and t< s< +oo for k E Pj2,
where j, k = 1, 2,... , n and the Djk(t) are defined by (VII.4.5). The initial points
rjk are chosen as follows:
to if k E P1I,
rjk = { +00 if k E Pj 2
gojk(t) = 0,
where r(t) = max(j,k) Irjk(t)I. Using assumption (VII.4.4), choose to so large that
r+oo
e'11 + nCJ r(s)ds < C.
Jto
Then, from (VII.4.11), it follows that Igpjk(t)I < C on the interval Z.
Step 5. Similarly, if (VII.4.11) holds for p = 1, 2, ... , we obtain
P
Igp+Iuk(t) - gpjk(t)l < (2} C
Step 6. Let us prove that the bounded solution qjk(t) satisfies condition (2) of
Theorem VII-4-3. If k E Pj2, then Tjk = +oo. Hence,
lim q)k(t)
o [ft ll
=f
.\)k(T)dr] [r)k(S) +> rJh(S)hk(s)J ds
exp h=1
= eXp Lf t Dk(r)dTJ
f
s
\.k(T)dr] {r)k(S) +
exP
rJh(s)hk(s)
[fC.Jk(r)dr] [rJk(s) +
h-1
n
1
dsl
rJh(S)ghk(S)ds
h=1
r)h(s)hk(s)l1 ds
JI
( t +oo
< exp J f D)k(r)dr] eK (1 + nCj ft) r(s)ds
l o
if (q)k(t)(< C on Z. Observe also that
I
f t
exp I f t
AJk(T)dr] [r)k(s) +
h=t
rJh(s)hk(s) ds
t-»+0
lim f t D)k(r)dT = -oo if k E P)
for a fixed a, we obtain
t
exp f DJk(r)dr] eK [1 + nC]
L o
1
f+ to
r(s)ds <
e
for large t. Therfore, for any positive number e, there exists such that Iq,k(t)( <
e for t > t(e). This complete the proof of Theorem IV-4-3.
4. A DIAGONALIZATION THEOREM 217
(VII.4.12) d 22 + p(t)rt = 0.
If we set
A(t) = 0 1l
1= do ,
0J
-P(t)
dt
'\t(t)
= i p(t)112,
911 _ [i p(t)1/2]
1
A2(t) = -t p(t)'/ pct) _ [-i pt)1/2J
dPo(t) - ip'(t) [0 01
POW-1 =
-i [ip(t)'/2 11
1 ,
dt 2p(t)'/2 1 1 '
2p(t) '/2 jP(t)1/2
218 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
and
P(to pt)ti21
0
Po(t)-'A(t)Po(t) _ 1/2 _i
[i
The transformation y = Po (t) i changes system (VII.4.13) to
- dP t)j z.
- P0(t)-1 {A(t)Po(t)
Using the computations given above, we can write this system in the form
(VII.4.14) ]
.ii = {t P(t) 1/ 2 [0 01] - 49t) [ 11 11
Suppose that
(1) a function p(t) is continuous on the interval Zo = {t : 0 < t < +oo},
(2) there exists a positive number c such that p(t) c > 0 on the interval Zo,
(3) the derivative p'(t) of p(t) is absolutely integrable on Io,
Then, Theorem VII-4-3 applies to system (VII.4.14) and yields the following theo-
rem (cf. IHN2]).
Theorem VII-4-9. If a function p(t) satisfies the conditions (1), (2), and (3)
given above, every solution of equation (IV.4.12) and its derivative are bounded on
the interval Zo.
The proof of this theorem is left to the reader as an exercise. Note that condition
(3) implies the boundedness of p(t) on the interval 1 .
If p(t) is not absolutely integrable, set
where
Ao = Al =
Anticipating that
(i) q(t)I and Ip'(t)I are of the same size,
(ii) Jq'(t) I and Ip"(t)I are of the same size,
choose q(t) and E so that two off-diagonal entries of the matrix on the right-hand
side of (VII.4.16) become as small as Ip'(t)I2 + [p"(t)I. In fact, choosing
q(t) = 8-ip(t)312
p'(t) , E = [01 -11
0J
S. SYSTEMS WITH ALMOST CONSTANT COEFFICIENTS 219
we obtain
(12 + q(t)E]' = 1 + I9(t)2 (12 - q(t)E(
and
q(t)E]-1 { [1(t)1/2
(12 + Ao - 4p(t) A11 (12 + q(t)E1 - dt) E}
1 jip(t)1/2A0 _ p'(t) EAo - p'(t) Al
1 + q(t) 2 8p(t) 4p(t)
+g(t}AoE+E(p,p',p") - q(t)
dtt)12}
I lp(t)1/2Ao- p,(t)12+E(p,p,p')-q(t)dq(t)12 ,
1 -r q(t) 2 I 4p(t) dt
where
E2 = -12, EAo = -AoE = 0 Il
11 0
and E(p, p', p") is the sum of a finite number of terms of the form
12
a (P (t) + $ p"(t)
p(t)h/2
with some rational numbers or and /3 and some positive integers h. Applying The-
orem VII-4-3 to system (VII.4.16), we can prove the following theorem.
Theorem VII-410. Suppose that
(1) a function p(t) is continuous on the interval Zo = {t : 0 < t < +oo},
(2) there exists a positive number c such that p(t) > c > 0 on the interval ID,
(3)
+00
(4) limop'(t) = 0.
Then, every solution of equation (VII.4.12) is bounded on the interval 7o.
The proof of this theorem is left to the reader. Condition (3) of Theorem VII-4-10
does not imply the boundedness of p(t) on the interval Io. For example, Theorem
VII-4-10 applies to equation (VII.4.12) with p(t) = log(2 + t).
Assumption 4. The matrix A has n mutually distinct eigenvalues p 1, u2, ... , An,
and the matrix V(t) satisfies the conditions
+00
t
lim V(t) = 0
+ Cc
and + [V'(t)[dt < +oo.
o
Let A1(t), 1\2 (t) .... , an(t) be the eigenvalues of the matrix A+V(t). Then, these
are continuous on the interval I. Furthermore, it can be assumed that
lim ),(t) = µj
t-+, (j = 1,2,... ,n).
Choose to > 0 so that A) (t),.\2(t), ... , an(t) are mutually distinct on the interval
I={t:to<t<+oo}. Set
F(t, A) = det[AII - A - V(t)].
Also,
j (t, A, (t)) + as (t, \,(t)) A (t) = 0 (j =1,2,... n)
OF
(t, \j (t)) # 0 (j = 1,2,... ,n),
on I.
8A
8F (t,
A, (t ))
since III, µ2i ... , p, are mutually distinct. Observe that .X (t) is
aF
8a (t, af(t))
linear homogeneous in the entries of the matrix V'(t). In this way, we obtain the
following lemma.
Lemma VII-5-1. Under Assumption 4, the derivatives of the eigenvalues of the
matrix A + V (t) are absolutely integrable over the interval I = {t : to < t < +oo},
i. e.,
JA(t)dtl
w
< +oo (j = 1,2,... ,n).
An eigenvector p, (t) of the matrix A + V(t) associated with the eigenvalue as(t)
can be constructed in the following manner. Observe that the characteristic poly-
nomial of A + V(t) can be factored as
(VII.5.2) (A+V(t)-.11(t)In)(A+V(t)-.\2(t)In)...(A+V(t)-an(t)IJ) = 0
5. SYSTEMS WITH ALMOST CONSTANT COEFFICIENTS 221
J1 (A+V(t) - Ah(t)In) # 0,
h 0j
lim fl(A + V(t) - Ah(t)In) = H(A - phln) 36 0,
t-+oo
h#l h#j
are eigenvectors of the matrix A associated with the eigenvalues p. (j = 1, 2,... , n),
respectively. Observe that the entries of the vectors pF,(t) (j = 1,2,... , n) are
polynomials in the entries of V(t) and A1(t),... ,An(t) with constant coefficients.
Hence,
+00
Ip"j'(t)I dt < +oc (j = 1,2,... ,n).
Jta
Thus, we proved the following lemma.
Lemma VII-5-2. Under Assumption 4, them exists a non-negative number to
such that
(1) the matrix A + V (t) has n mutually distinct eigenvalues A1(t)...... n (t) on
the interval I = It : to < t < +oo},
(2) the eigenvalues Al(t), ... , An(t) are continuously differentiable on I and
lim Aj(t) = p, (j = 1,2,... ,n),
(3) the matrix A + V(t) has n eigenvectors p1(t),... ,#n(t) associated with the
eigenvalues \I(t) .... ,An(t), respectively, such that lim 73 (t) = 4j (j =
t-+oo
1, 2, ... , n) are eigenvectors of the matrix A associated with the eigenvalues
p) (j = 1,2,... ,n),
(4) the derivatives of the eigenvalues A,(t) (j = 1,2,... , n) and the derivatives
o f the eigenvectors p ' , (t) (j = 1, 2, ... , n) with respect to t are absolutely inte-
grable on the interval Z.
Set
Then,
lim Po(t) = Q, lim A(t) = M,
t-+oo
Po(t)-'[A + V(t)]Po(t) = A(t), Q-'AQ = M,
f +oo I Po(t)_ 1
dP°(t) I
dt < +oo.
to dt
Observe that the transformation y"= P°(t)z changes system (V11.5. 1) to
dl t)1 i= [A(t) - Po(t)-'d t)]
dt = POW-1 [[A + V(t)]P0(t) - Y.
Observation VII-5-5. Let us look into the case when the matrix A has multi-
ple eigenvalues. To do this, consider system (VII.3.1) under Assumption 1 given
in §VII-3. By virtue of Theorem VII-3-1, system (VII.3.1) is changed to system
(VII.3.4) by transformation (VII.3.3). Furthermore, (VII.3.4) is changed to
(j = 1, 2, ... m),
(VII.5.3)
dt [ N 3 + R 3 (t)) u 3
where
Let us look at the quantity sk(t - s)h. Note that 0 5 k 5 r - 1 and 0 5 h < r - 1.
224 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
Case 1. k + h < r - 1. In this case, since sk(t - s)h = sk+h C t _ 11, define
- s
Sk(th'
8)h N°R(s)ds = / t 8k(th! S)h NhR(s)ds.
Jt
(hp) Sk+pth-p
Case 2. k + h > r. In this case, look at sk(t - 8)h = D-1)p
p_Q
j Sk+pth-PNhR(S)dS
= J+ao p) sk+Pth-PNhR(s)ds.
; k+Pt h- PNhR{s)d s =
-3
k
h
)
r
t
o -
kh $ k+h-
Pt P Nh R(s)d$, where t > to.
Setting
r
U(t) = f t ske(`-e)NR(s)ds,
6. AN APPLICATION OF THE FLOQUET THEOREM 225
we obtain
dU(t)
= NU(t) + tkR(t).
dt
Step 3. Let us construct n x n matrices Uk (t) (k = 0, 1, 2, ... , r - 1) by the integral
equations
ft
Uk(t) = Nk + 1 ske(t-")NR(s)Ui(s)ds
Then,
lira
t-++00
Uk(t) = Nk
Observe that
tkUk(t) tkNk
rt e(t_s)NR(s)skUk(s)ds.
k! k! n k!
Hence, setting
tkUk(t)
k!
k=0
we obtain r-1
k r
1(t) = eIN +
t
e(t-,)NR(s) s
ds.
Y' r k!!
k=o n
This implies that
do(t)
_ (N + R(t))W(t).
dt
It can be easily shown that
L(t)c = etNc' - Nk"
lim Urn
t-+oo tk t-+00 tic
Assumption VII-6-1.
(1) The entries of an n x n matrix A(t, ej are continuous in (t, e) E R x A(r)
and analytic in f E A(r) for each fixed t E LR, where FE C'" unth the entries
et, ... , A(r) = IF: Ie < r}, and r is a positive number.
(2) The entries of A(t, e) are periodic in t of a positive period w.
(3) The entrees hl,... , h,,, of a C"t -valued function h(t) are continuous on the
interval T(to) = {t : to < t < +oo} for some non-negative number to and
lim h(t) = 6.
t+00
Let us consider the following two cases.
Case 1. The function h(t) is supposed to be continuously differentiable on T(to)
and satisfy the condition
+00
(VII.6.2) J Ih(t)Idt = +oo and f + Ih'(t)Idt < +oo.
I.
lim
t -+00
+00 +00
(VII.6.3) r= +00, / = +oo,
0
L00I;(t)ldt
Ih"(t)Idt < +oo, J + Ih'(t)I2dt < +oo
J0 0
sin(f)
For example, the function h(t) _
satisfies (VII.6.3).
I satisfies (VI1.6.2), whereas h(t) =
f
Observation in Case 1. In Case 1, we use the following lemma.
Lemma VII-6-2. If a matrix A(x, ej satisfies conditions (1) and (2) of Assump-
tion VII-6-1, them exist n x n matrices P(t,e) and H(t") such that
(i) the entries of P(t, ) are continuous in (t, e) E R x A(f) and analytic in
FE A(f) for each fixed t E R, where f is a suitable positive number.
(ii) P(t + w, ej = P(t, t-) for (t, F) E R x ©(f),
(iii) P(t, fj is invertible for every (t, t) E R x A(f),
(iv) the entries of H(t) are analytic in FE A(f),
2ai
(v) any two distinct eigenvalues of H(0) do not differ by integral multiples of -,
w
(vi) a P(t, a exists for (t, e) E R x L(f) and given by
Proof
In order to prove this lemma, construct a fundamental matrix solution $(t, e)
of the differential equation d = A(t, e)y by solving the initial-value problem
dX = A(t, e-) X, X (O) = I, where In is the n x n identity matrix. The en-
dt
, bog((w
tries of 4!(w, ej are analytic in A(r). Define H(e) by H(ej = and
P(t,') = 4i(t, cl exp(-tH(E)]. Then, (VII.6.4) follows.
The most delicate part of this proof is the definition of H(). Details are left for
the reader as an exercise (cf. [Sill).
Changing system (VI I.6.1) by the transformation
Proof
In fact,
dil
= P(t, h(t))-1 {A(t(t))P(t&(t)) - dt [P(t, h(t))] } u"
iiT
1 < <m
under assumption (VII-6.2). Also, observe that H(i(t)) does not contain any pe-
dH(i(t))
riodic quantities and dt is absolutely integrable. Therefore, if eigenvalues
of H(6) satisfy suitable conditions, the argument given in §VII-5 applies to system
(VII.6.6).
228 VII. ASYMPTOTIC BEHAVIOR OF LINEAR SYSTEMS
(VII.6.10) dt = {Bo(h(t)) + B,
Notice that the entries of the matrix B, (t, E, µ) are periodic in t of period tv and
that B1 (t, E, 6) = 0. Furthermore, any two distinct eigenvalues of Bo (6) do not
27n
differ by integral multiples of -. Set D(r) = IE + lµI < r}.
In Case 2, the following lemma is used.
Lemma VII-6-4. There exist n x n matrices P, (t, e, {'t) and HI (E, µ) such that
(:) the entrees of P1 (t, E, ui) are continuous in (t, E, u') E R x D(r) and analytic in
(E, )i) E D(r) for each fixed t E Ii£, where r is a suitable positive number,
(i:) P, (t + w, F,)!) = P, (t, e, it) for (t, F, tt+) E 1 x D{r
(iii) P1(t, E, 6) = O for (t, i) E R x 0(r),
(iv) the entries of H1 (E, u') are analytic in (E, Et) E D(r) and H1 (E, 6) = 0 for
E E O(r),
(v) at P, (t, e, µ') exists for (t, E, p) E R x D(r) and is given by
19
Pi(t,e,u) ={Bo(e) + B1(t,e,µ)) (In + Pi(t,E,ls)}
(VII.6.11)
+ P1 (t, i,#)) { Bo(E) + Hi (E, µ) }
(VII.6.12)
a P1(t,Ej) = Bo(E)P1(t,E,i) - Pi(t,E., )Bo(Ej
+ {B1(t,E,N)P:(t,E,Ji) - +B1(t,,rA-)} -Hl(E,u-)-
Proof of Lemma VII-6-4.
M
Given p = (pi,... ,p,,,), where the p, are non-negative integers, denote EIpjI
=1
and 4' um by Ipl and 91, respectively. Set
8
(VII.6.13) p = B0(E)P1,p - P1,pBo(E) + Q1,p(t, E) -
where
Hence,
(VII.6.15)
x 1 Q1,p(s, ) - H1,$,(e)) exP[sBo(E))ds} exp[-t B0(Z)1,
where C(e) and H1,p(e) are n x n matrices to be determined by the condition that
P1,p(t,e) is periodic in t of period w, i.e.,
It is not difficult to see that condition (VII.6.16) determines the matrices C(ep)
and Hl,p(ej. Then, the matrix P1,p(t, E) is determined by (VII.6.15). The conver-
gence of power series P1 and H1 can be shown by using suitable majorant series. 0
In the same way as the proof of Theorem VII-6-3, the following theorem can be
proven.
Theorem VII-6-6. The transformation
dv = f
H(i(t)) + H1(h(t), h'(t))
dt l
where P1(t, E, µ) and H1(E, µ) are those two matrices given in Lemma VII-6-4.
In + P1(t,h(t),h'(t))j_1 8Ej
f0+00
(VII.6.19)
Observe also that the matrix H(h(t))+H1(h(t), h'(t)) does not contain any periodic
terms. Furthermore, H(6) + H1(6, 6) = H(0). It is clear that the derivative of
H(h(t)) is not necessarily absolutely integrable over to < t < +oo. Therefore, in
order to apply the argument of MI-5, the matrix H(h(t)) must be examined more
closely in each application. Details are left to the reader for further observation.
EXERCISES VII
VII-l. Find the Liapounoff's type number of each of the following four functions
f (t) at t = +00:
(1) exp [t2sin (1L1) sint(dt](3) inin(exp[3t],exp[5t]sin67rt]),
]' (2) exp Lfo
(4) the solution of the initial-value problem y"-y'-6y = eat, y(O) = 1, y'(0) = 4.
VII-2. Find a normal fundamental set of four linearly independent solutions of
the system dy = Ay" on the interval 0 < t < +oo, where
log(l+x) e= expx2
= x3 1+ 1+ x sin x
exp(e') cosx arctanx
on the interval 0 < x < +oo. Find Liapunoff's type number of det 4i(x) at x = +oo.
EXERCISES VII 231
VII-4. Assuming that the entries of an n x n matrix A(t) are convergent power
series in t-1, calculate lim p log I (t) for a nontrivial solution ¢(t) of the differential
0
i
L
0 0 1+t2
Hint. See [GH].
VII-7. Show that if a function p(t) is continuous on the interval 0 < t < +oo and
lim t-Pp(t) = I for some positive integer p, the differential equation d2
t +p(t)y =
t +oo
0 has two linearly independent solutions rlf(t) such that
77, (t)
= P(t)114(+1 +o(1))exp [f iJ t
to
d4] ,
parameters. Set A(x, f) = {(0) 11. Also, let Ro, po, and ao be arbitrary but
fixed positive numbers. Suppose that 0 < po < 2. Show that the system
d:V
= A(x,f)y, y = 17121
Show that there exist a non-negative number to and an n x n matrix T(t) such that
(1) the derivative (t) exists and the entries of T(x) and . (x) are continuous
E
in t on the interval 2 = It : to < t < +oo},
(2) t limo T(t) = 0,
(3) transformation (I + T(x)) z changes the system
(diag[A1(t),A2(t),... +T(t))y
dt =
to
d
= diag(A1(t) + bl (t), A2(t) + b2(t), ... , An(t) +
dt
where y E C, i E C, n functions b1(t),... ,b,,(t) are complex-valued and
continuous in t on T, and lim b,(t) = 0 (j = 1, ... , n).
t +00
EXERCISES VII 233
VII-15. Using Theorem VII-6-3, find the asymptotic behavior of solutions of the
differential equation
STABILITY
dy
(VIII.1.1) t=
d
A(ro) = Zo x D (ro) = {(t, y1 E J"+I : 0 < t < +oo, 1171 < ro}.
Also, assume that a solution mo(t) of (VIII.1.1) is defined on the entire interval 10
and that (t, mo(t)) E A(ro) on Yo. The main topic in this chapter is the behavior
of solutions of the initial-value problem
(VIII.1.2)
dt = f (t, o'
as t +oo. To start, we introduce the concept of stability.
235
236 VIII. STABILITY
Definition VIII-1-1. The solution do(t) is said to be stable as t - +oo if, for
any given positive number c, there exists another positive number b(c) such that
whenever I&(0) - 7 < 5(c), every solution 4(t) of initial-value problem (VIII.1.2)
exists on the entire interval Zo and satisfies the condition
has the unique solution y = (t, r, i) if (r, rl) E 0(ro). Then, qs(t, r, 771 is continuous
with respect to (t, r, '). Therefore, for any r E Zo and any given positive numbers
T and f, there exists a positive number p(r,T,e) such that whenever 100(r) - )1 <
p(r, T, c), the solution 0(t, r, 71) exists on the interval 0 < t < T and I0o(t) -
¢(t, r, Y-7)1 < E on the interval 0 < t < T (cf. §11-1). This implies that if the solution
0o(t) is stable as t +oo, then for any r E Zo and any positive number c, there
exists another positive number b(r, c) such that whenever loo (,r) - n7 < b(r, E), the
solution (t, r, rt) exists on the entire interval 10 and (t, r, J) satisfies the condition
IV'0(t) - (¢(t, r, T-D I < E on I.
We also introduce the concept of asymptotic stability.
Definition VIII-1-3. The solution do(t) is said to be asymptotically stable as
t-4 +00 if
(i) the solution do(t) is stable as t - +oo,
(ii) there exists a positive number r such that whenever I o(0) - i7 < r, every
solution fi(t) of initial-value problem (VIII. 1.2) satisfies the condition
Hence, the study of the solution o(t) of (VIII.1.1) is reduced to that of the trivial
solution z1(t) = 0 of (VIII.1.5). Thus, the solution .o(t) of (VIII.1.1) is stable
(respectively asymptotically stable) as t -+ +oo if and only if the trivial solution of
(VIII.1.5) is stable (respectively asymptotically stable) as t -- +co. In the following
sections, we shall study stability and asymptotic stability of the trivial solution.
The following three examples illustrate stability and asymptotic stability.
Example VIII-1-5. The first example is the system given by
FIGURE 1.
Example VIII-1-6. The second example is a second-order differential equation
(VIII.1.9)
d-t2
+ g(t7) = 0,
238 VIII. STABILITY
(VIII.1.11)
di = Az + #(z-),
iii
where
Set
17
Then,
dH(yi(t),y2(t)) dyl(t) (t) dy2(t)
dt
= 9 ( Y1 (t))
dt
+ Y2
dt
Therefore,
dH(yi(t),y2(t))
= 9(YI(t))y2(t) - y2(t)9(yI(t)) = 0
dt
we obtain ('(t) = 2(G(a) - G((1(t) + a)) + co >- Co as long as 0 < (1(t) < po.
Hence, there must be a positive number to depending on (o such that (I (to) = Pa
no matter how small to may be. This implies that the trivial solution of (VIII.1.11)
is not stable.
Case when g'(a) > 0: For a given positive number p, let V (p) be the connected
component containing 0 of the set {[; : G(( +a) < G(a)+p}. Then, V (pl) C V (P2)
if p1 < p2. Furthermore, if g(a) = 0, g(a) > 0, and if a positive number p is
sufficiently small, there exist two positive numbers e1(p) and e2(p) such that
(1) e2(P) < e1(P),
(2) Plmoe1(P) = 0,
(3) {( : ISO < C2 W} C 7) (P) C < el(p)} (cf. Figure 2).
S=o
G= G(S+a)
G=G(a)+p
G=G(a)
1 P)
FIGURE 2.
Observe also that G(a) < G(( + a) < G(a) + p for [; E V (p) if p is a sufficiently
small positive number.
For a given positive number e, choose another positive number p so that ei(p) < e
and p:5 e. Choose also the initial value z(0) so that Jz1(0)j < e2 (2) and z2(0)2 < p.
Then, G(z1(0)+a) < G(a)+ since z1(0) EP (2). Hence, (VIII.1.14) implies that
G(zi(t) + a) < G(a) + p for all
2 t. Observe that the set {z1(t) : all t} is connected
and zi(0) E D (2) C V (p). Therefore, zi(t) E D(p) for all t. Hence, Iz1(t) <
ei (p) < e. On the other hand, G(a) < G(z1(t) + a) < G(a) + p since z1(t) E D (p)
for all t. Hence, (VIII.1.14) implies that2(tz2< G(z1(0) +a) - G(a) + < p < e.
This proves that the trivial solution of system (VIII.1.11) is stable as t -+
2 +oo.
The analysis given in Example VIII-1-6 is an example of an application of the
Liapounoff functions to which we will return in Chapters IX and X.
The third example is the following result.
Theorem VIII-1-7. If f (x, y) and g(x, y) are real-valued continuously differen-
tiable functions such that
(i) f (0, 0) = 0 and g(0, 0) = 0,
(ii) (f (x, y), g(x, y)) 0 (0, 0) if (x, Y) # (0, 0),
(iii) 8x (x, y) + 49Y (x, y) = 0,
then, the trivial solution (x, y) = (0, 0) of the system
aj(ro,c1,c2)
It is known that the matrix 1L(t,cl,c2) = is the
L ac, 0Ce
unique solution of the initial-value problem
1of of
dX ax 8y X. X (0) = I2.
dt a9 09
ax Dy tr,y)=Oit.c,,cz)
(cf. (4) of Remark IV-2-7). Now, it follows from (VIII.1.15) that the area of
D(ro, ro) is equal to the area of 0(ro). This is a contradiction. 0
2. A SUFFICIENT CONDITION FOR ASYMPTOTIC STABILITY 241
Step 3. Fix a non-negative number T and two positive numbers 6 and 61 in such
a way that
p-K(p(T)+co6'o) > 0, 0<61 <6, K61 <6.
Assume that 19(T)J < 61. Then, inequality (VIII.2.3) holds for T < t < T1 as long
as Iy(t)I < bon the interval T < t < T1. This, in turn, implies that
Jg(t)J < K61 < 6 for T < t < T1.
This is true for all T1 not less than T. Hence, inequality (VIII.2.3) holds for t > T.
Step 4. If Jy(t)J < b < 1 for 0 < t < T, there exists a positive number is such
Id t)
that I < n19(t)j, for 0 < t < T. This implies that Jy-(t)J < Jy-(0)JeK' as long
as Jy(t)I < 6 < I for 0 < t < T. Therefore, Jy(T)J is small if Jy(0)j is small. Thus,
it was proven that (VIII.2.3) holds for t > T if Iy(0)J is small. This completes the
proof of Theorem VIII-2-1. O
Example VIII-2-2. For the following system of differential equations d = A#+
where
y= [y
Y
j A= r -0.4
-
-2
.2
1 (Y2 (
+ y2) I 11
1
the trivial solution y = 0 is asymptotically stable as t -' +oc. In f act, the char-
acteristic polynomial of the matrix A is PA(a) _ (A + 0.3)2 + 1.99. Therefore, the
real part of two eigenvalues are negative.
Remark VIII-2-3. The same conclusion as Theorem VIII-2-1 can be proven, even
if (VIII.2.2) is replaced by
I9(t,y)l 5 e(t,yyly7 for (t,y-) E o(ro) _ {(t,y-) : 0 5 t < +oo, Iy-I <_ ro},
where ro > 0, e(t, y-) > 0 for (t, y) E i(ro), lim e(t, y) = 0,
and
t-1+19{o
(iii) the real part of an eigenvalue of the matrix A is positive.
dy
Then, the trivial solution of the system = Ay + §(t, y-) is not stable as t -. +oo.
dt
A proof of this theorem is given in (CL, Theorem 1.2, pp. 317-3181. We shall prove
this theorem for a particular case in the next section (cf. (vi) of Remark VIII-3-2).
An example of instability covered by this theorem is the case when g'(a) < 0 of
Example VIII-1-6. The converse of Theorem VIII-2-7 is not true. In fact, Figure 3
shows that the trivial solution of the system
dY2
I
= -yI, = (yi + yi)112
is not stable as t - +oo. Note that, in this case, eigenvalues of the matrix A are
-1 and 0.
Y2
Yt
FIGURE 3.
for max(jil, Iy1) < po and max(11 i, Ii1) < po, where po is a positive number and
u oL(p) = 0. Furthermore, assume that gf(0, 0) = 0 (j = 1, 2). Two matrices AI
and A2 are respectively constant n x n and m x m matrices satisfying the following
condition:
le(t-9)A' I < Kie-o1(t-a)
for t > s,
(VIII.3.2)
t Je(t-s)A21 < K2e-02(t-8) for t < s,
where K, and of (j = 1,2) are positive constants. Condition (VIII.3.2) implies
that the real parts of eigenvalues of AI are not greater than -01, whereas the real
parts of eigenvalues of A2 are not less than -02. Assume that
(VIII.3.3) a1 > 02.
Let us change (VIII.3.1) to the following system of integral equations:
I
x(t, c) = e1AI C + J e(1-a)A' gi (Y(s, c-), y(s, c))ds,
0
(VIII.3.4)
At, c ) = Jt to0 e(t 8)A2 2(x(s, c), y(s, c))ds.
and
t +00
e-02(t-8)e-(0 $ -')8ds
IJ e(t-a)A2 2(j(s, c), y(s, c))dsl < K2L(p)p f
+ 00 t
K2L(P)P e
al -o2-E
3. STABLE MANIFOLDS 245
K2L(p) < 1
This implies that if a positive number p is chosen so small that
01-o2-e - 2
and K1L(p) < and if the arbitrary vector c' in R" satisfies the condition
Kl I i < 2
, then, using successive approximations, a solution (i(t, cl , y(t, c)) of
2
(VIII.3.4) can be constructed so that
x(O, c-)= c and max(Ii(t, c1I, Iy(t, c)I) < pe-(c for 0 < t < +oo.
Details are left to the reader as an exercise.
Remark VIII-3-2.
(i) The positive number a is given to start with and the choice of p depends on
e. However, since this solution approaches the trivial solution, the constant e
may be eventually replaced by any smaller number, since the right-hand side
of (VIII.3.1) is independent of t. This implies that the curve (i(t, c), y'(t, c-))
is independent of a as t -. +oo. More precisely, if a solution (x"(t), y"(t)) of
(VIII.3.1) satisfies a condition
(c) 9(0,6) = 6.
Suppose further that A has an eigenvalue with positive real part. Then, applying
Theorem VIII-3-1 to the system = -Ay' - g(y-), we can construct the stable
manifold U of the trivial solution. This means that if a solution fi(t) of (VIII.3.6)
starts from a point on U, then urn Q(t) = 6. This shows that the trivial solution of
t - +oo, and Theorem VIII-2-7 is proved for (VIII.3.6).
The set U is called the unstable manifold of the trivial solution of (VIII.3.6).
The materials in this section are also found in [CL, §§4 and 5 of Chapter 81 and
[Hart, Chapter IX; in particular Theorem 6.1 on p. 2421.
is a formal power series with coefficients fp E C', where p = (pl,... , p") with
n
non-negative integers pl, ... , p", > Ph, and yam' = yl' yP,,-.
h=1
The following theorem is a basic result concerning formal simplifications of sys-
tem (VIII.4.1).
Theorem VIII-4-2. Under Assumption VIII-4-1, there exists a formal power se-
ries
with a constant n x n matrix Bo and the formal power series g(ui) _ E upgp
Ipi>2
with coeficients gp in Cn such that
(iia) the matrix B0 is lower triangular with the diagonal entries Al.... , An,
and the entry bo(j, k) on the j-th row and k-th column of B0 is zero
whenever A, j4 Ak,
(iib) for p with IpI > 2, the j-th entry gp, of the vector gp is zero whenever
A1 n
(VIII.4.5) 1: PhAh
h=1
Proof
Observe that if y" = P(d), then
d9 = p p Pp
dt
Po + 1: [ L Pp P22 Pp ... un Pp Boi + F 4Lpgp
W>2 Ipl>2
and
Furthermore,
poop P_p
Po + [Plup,5, P2upP-p ... Bo u'
ul u2 Un J
Ipl>2
n
= PoBou" + phAh u"PPp + uh
plp),A-
ul
lpl>2 h_1 1<h<35n
where 03,k is the entry of B0 on the 7-th row and k-th column.
Let us introduce a linear order pi -< p2 for p. = (p,',... , pin) (j = 1, 2) by the
relation
Plh = p2h for (h < he) and P1ho < P2ho
Now, calculating the coefficients of iip 1) on both sides of (VIII.4.1), we
obtain
for Jpj > 2. From (VIII.4.6), it is concluded that the diagonal entries A1, ... , A,, of
Bo are eigenvalues of A and that this allows us to set A = B0 and P0 = In, where
I,, is the n x n identity matrix. Then, (VIII.4.7) becomes
(VIII.4.8)
(PhAh)PP + 9p - Bo,6,
h=1
= ff(PP, p) + 9$'05P" 9P, : 10 < 1p1)
successively, where PP,, and gp,, are the j-th entries of the vector P. and
respectively, and FP,, are known quantities. If >phAh - Aj 36 0, set 9,j = 0 and
h=1
solve (VIII.4.9). If >phAh - A? = 0, then set 9P,, = Fr,, and choose PPJ in any
h=1
way. This completes the proof of Theorem VIII.4.2.
Observation VIII-4-3. Assume that R(AE) < 0 for j = 1, ... , r and R(A3) > 0
for j 1,... , r. In this case, if uh = 0 for h I, ... , r, the a-th entry of the
vector Bou + g"(u) is zero for t ¢ I, ... , r. In fact, look at g"(u"). Then, the f-
n
th entry of the coefficient g"p is zero if Al 96 >phAh. Note that, if t > r and
h=1
n
At = > Ph Ah, then (Pr+ 1, . Pn) 0 (0, ... , 0). Hence, in such a case, uP = 0 if
h=1
(ur+1, ... , u,) = (0, ... , 0). Therefore, the 1-th entry of the vector Boti + 9"(uis
zero forI>rif(ur+1,...,u,,)=(0,....0).
Observation VIII-4-4. Under the same assumption on the Aj as in Observation
VIII-4-3, set (ur+1, ... , un) = (0,... , 0). Then, the system of differential equations
on (u1,... has the form
dud A3u3
= + QJ,huh
dt
(VIII.4.10)
... u p , ( i=1 , -...- - , r)-
+ 9(P,.
.P04ul1
A,=p,a,+ +p.a.,lai>2
r r
Observe that since A. - ph Ah # 0 if ph is sufficiently large, the right-hand
h=1 h=l
members of (VIII.4.10) are polynomials in (ul,... ,ur).
4. ANALYTIC STRUCTURE OF STABLE MANIFOLDS 249
Observation VIII-4-5. Assume that R[Ah+1J <_ R[Ah]. Then, (V111.4.10) can be
written in the form
du
dul
= Aiu1 and = A)u) + u_,-i) for j=2,... ,r.
dt
Hence, system (VIII.4.10) can be solved with an elementary method. To see the
structure of solutions of (VIII.4.10) more clearly, change (ul, ... , ur) by u2 _
ea'ivi (j = 1,... , r). This transformation changes (VIII.4.10) to
dv)
Q),hvh
dt
an
g(P,..
...vf (.i
A, =P, a,+...+prAr,Ipl>2
This, in turn, shows that the general solution of (VIII.4.10) has the form u) _
ea'ii) (t, cl, ... , cr) (j = 1.... , r), where (cl,... , c,.) are arbitrary constants and
0,(t,c1,....c,-) are polynomials in (t.cl,... ,G)
An analytic justification of the formal series 15(ii) is given by the following the-
orem.
Theorem VIII-4-6. In the case when the entries of the Z"-valued function f'(y-)
on the right-hand side of (VI11.4.1) are analytic in a neighborhood of 0, under the
same assumption on the A) as in Observation VIII-4-3, the power series P(d) is
convergent if (ur+i, ... , u") = (0,... , 0).
The proof of this theorem is straight forward but lengthy (cf. [Si2J). The key
fact is the inequality
A) - >PhAh
h=1 h=1
r
for some positive number a if E ph is large. In this case, a iajorant series for P
h=1
can be constructed.
The construction of such a majorant series is illustrated for a simple case of a
system
dy
dt
= Ay + f (y),
ui"A,. Then,
lal>2
Ap
a= AOPI-1) for all p (IpJ > 2).
Set
1
Then, F(y-) is a majorant series for ft y-). Determine a power series V (u-) = uupii
IpI>2
by the equation v = ICI 0(u" + v'). Set F(u + V(U)) _ iBa. Then,
Ipl>2
B
vp = Ii for all p (JpJ > 2).
It can be shown easily that Y(ul) is a convergent majorant series of Q(ur). This
proves the convergence of Q(ii).
Putting P(u") and the general solution of (VIII.4.10) together, we obtain a par-
ticular solution y" P(#(t, cl) of (VIII.4.1), where
=
(t,cj = (e'\It7Pi(t,cl,... ,cr),... ,e*\1 'Wr(t,cl,... ,cr),0,... ,0).
This particular solution is depending on r arbitrary constants c = (cl,... , c,.).
Furthermore, this solution represents the stable manifold of the trivial solution of
(VIII.4.1)if W(Aj)>0for j#1,...,r.
Remark VIII-4.7. In the case when y, A, and AM are real, but A has some
eigenvalues which are not real, then P(yl must be constructed carefully so that the
a b
particular solution P(i(t, c)) is also real-valued. For example, if A = ,
b a
the eigenvalues of A are a ± ib. If IV = ['] is changed by ul = 1/i + iy2i and
= - iy, system (VIII.4.1) becomes
dul
= (a + ib)ul + 9P,,p2u 'uZ
OFF
Pl+P2>2
(VIII.4.11)
due
_ (a - ib)u2 + ,
dt
by the transformation
PP,.P2'UP11t?221
VI +
p, +p2>_2
(VIII.4.13)
7P,.p2V2P1VP13.
V2 +
P,+p2>2
V 2 V2
Now, system (VIII.4.12), in turn, is changed back to - = At by wI =
VI - V2
and w2 = ,
where (w1, w2) are the entries of the vector tu. Observe that
2i
U l + u2
2
w1 + E
p,+p2>2
uI - u2
U'2 + g2.P,,pzu'P1 u"
2i
p, +p3 22
where ql,p,,p2 and g2,p,,p2 are real numbers. Similar arguments can be used in
general cases to construct real-valued solutions. (For complexification, see, for
example, [HirS, pp. 64-65].)
For classical works related with the materials in this section as well as more
general problems, see, for example, [Du).
AI =
2
+
4 q and A2 = 2 - 4
- q.
It is known that
P2
p = AI +A2, q = AIA2, and Al - A2 =2 - q.
4
252 VIII. STABILITY
Also, let t and ij be two eigenvectors of A associated with the eigenvalues Al and
A2, respectively, i.e., At = All;, l 0 0, and Au = A2ij, ij 0 0. Observe that, if
y(t) is a solution of (VIII.5.1), then cy(t +r) is also a solution of (VIII.5.1) for any
constants c and T. This fact is useful in order to find orbits of equation (VIII.5.1)
in the phase plane.
Case 1. Assume that two eigenvalues Al and A2 are real and distinct. In this
case, two eigenvectors { and ij are linearly independent and the general solution of
differential equation (VIII.5.1) is given by
FIGURE 4. FIGURE 5.
1c: In the case when Al > 0 > A2 (i.e., q < 0), the phase portrait of orbits of
(VIII.5.1) is shown by Figure 6. The trivial solution 0 is unstable as Itl +oo.
Note that as t -+ -oo (or +oo), only two orbits of (VIII.5.1) tend to 0. The point
(0, 0) is called a saddle point.
2
Case 2. Assume that two eigenvalues Al and A2 are equal. Then, q = 4 and
Al=A2=29& 0.
for every direction n, there exists an orbit which tends to 6 in the direction n" as t
tends to -oo (respectively +oo). The point (0, 0) is called an unstable (respectively
stable) proper node if p > 0 (respectively p < 0).
p>0 p<0
FIGURE 6. FIGURE 7-1. FIGURE 7-2.
2b: Assume that A is not diagonalizable; i.e., A = p 212 + N, where 12 is the 2 x 2
identity matrix and Nris a 2 x 2 nilpotent matrix. Note that N 0 and N2 = O.
Hence, exp[tA] = exp 12t] {I2 + tN}. Observe also that a nonzero vector c' is an
eigenvector of A if and only if NcE = 0. Since N(Nc) = 0, the vector NcE is
either 6 or an eigenvector of A. Hence, NE = ct(-) , where is the eigenvector
of A which was given at the beginning of this section and a(c) is a real-valued
linear homogeneous function of F. Observe also that at(c) = 0 if and only if c' is a
constant multiple of the eigenvector . The general solution of (VIII.5.1) is given
by y(t) = exp [t] {c + ta(c7}, where c" is an arbitrary constant vector. In this
case, the phase portrait of orbits of (VIII.5.1) is shown by Figures 8-1 and 8-2. The
trivial solution 6 is unstable (respectively stable) as t - +oo if p > 0 (respectively
p < 0). The point (0, 0) is called an unstable (respectively stable) improper node if
p > 0 (respectively p < 0).
p>0 p<0
Case 3. If two eigenvalues \I and .12 are not real, then q > and
4
Al =a+ib, A2=a-ib, a= 2,
Note that b > 0. Set A = a12 + B. Then, B2 = -b212, since two eigenvalues of B
254 VIII. STABILITY
Smbbt)
are ib and -ib. Therefore, exp[tB] = (cos(bt))I2 + B.
3a: Assume that a = 0 (i.e., p = 0). Then, the general solution of (VIII.5.1) is
slnbbt)
given by '(t) = exp[tB]c = (cos(bt))c"+ Bc, which is periodic of period 2b
in t. The phase portrait of orbits of (VIII.5.1) is shown by Figures 9-1 and 9-2.
The trivial solution 0 is stable as It 4 +oo. The point (0, 0) is called a center. It
is important to notice that every orbit y(t) is invariant by the operator . In fact,
b
B j(t)
T
= cos(bt)
b
Bc-(sin(bt))c = (cos (bt + ))e+
sin bt +
b
r,
2 Be= 17 (t +
j
2b
)
(0.0)
p>0 p<o
FIGURE 10-1. FIGURE 10-2.
Let us summarize the results given above by using Figure 11:
(1) (0, 0) is an improper node.
(2) (0, 0) is a saddle point.
6. ANALYTIC SYSTEMS IN R2 255
(4) (IV)
(3) (3) (111) (Ill
-------- -------
FIGURE 11. FIGURE 12.
(VIII.6.1)
dt Ay" + E ypfp,
tp!>2
where y' E LR2 with the entries yj and y2, A is a real, invertible, and constant
2 x 2 matrix. p = (pi, p2) with two non-negative integers pi and p2i Ipl = pi + p2,
yam' = y'' y2 , the entries of vectors fp E R2 are real constants independent of t, and
the series on the right-hand side of (VIII.6.1) is uniformly convergent in a domain
A(po) = (y E R2 : fyj < po) for some positive number po. Let us look into the
structure of solutions of (VIII.6.1) in the following five cases.
Case 1. If the point (0, 0) is a stable proper node of the linear system = Ay, then
A = )J2i where A is a negative number and 12 is the 2 x 2 identity matrix. To apply
Theorem VIII-4-2 to this case, look at the equation A = pi.1 + p2A on non-negative
integers pi and p2 such that pl + p2 > 2. Since no such (pi, p2) exists, there exists
an R2-valued function P(u) whose entries are convergent power series in a vector
it E R2 with real coefficients such that 5 () = 12 and that the transformation
= P( u reduces system (VIII.6.1) to dt` = Ail. This, in turn, implies that the
256 VIII. STABILITY
point (0, 0) is also a stable proper node of (VIII.6.1) and that the general solution
of (VIII.6.1) is y = P(eJ1°cl, where c is an arbitrary constant vector in R2.
Case 2. If the point (0, 0) is a stable improper node of the linear system dt = Ay,
then we may assume that either (1) A = AI2+N, where 1A is a negative number and
N 4 0 is a 2 x 2 nilpotent matrix, or (2) A = I 1 a2 J , where Al and A2 are real
negative numbers such that Al > A. In case (1), the same conclusion is obtained
concerning the existence of an 12-valued function P(i7). Therefore, the point (0,0)
is a stable improper node of (VIII.6.1), and the general solution of (VIII.6.I) is
y" = P(eAt(12 + tN)c), where c is an arbitrary constant vector in 1R2. In case (2),
looking at the equations Al = p1 A 1 + p2A2 and A2 = p1 At + p2A2 on non-negative
integers pi and p2 such that pl + p2 ? 2, it is concluded that there exists an 1R2-
valued function P(ui) whose entries are convergent power series in a vector ii E R2
dP
with real coefficients such that (d) = 12 and that the transformation y" = P(u)
['Ju.&Ajul
reduces system (VIII.6.1) to dt = 2u21, where (u1, u2) are the entries of
1
i , M is a positive integer such that A2 = MA1, and -y is a real constant which must
be 0 if A2 54 MA1 for any positive integer M. Therefore, in case (2), the general
A,:
([(''
solution of (VIII.1.6) is Cty"7t=+ C2)e
PAlt ) , where cl and c2 are arbitrary
constants. This, in turn, implies that the point (0, 0) is a stable improper node of
(VIII.6.1).
dy
Case 3. If the point (0, 0) is a stable spiral point of the linear system = Ay",
at
then we may assume that A = I b ab 1, where a and b are real numbers such that
a < 0 and b # 0. The eigenvalues of A are At = a ± ib. This implies that there are
no non-negative integers pi and p2 satisfying the condition A = p1 A+ +P2A_ and
pl + > 2. Therefore, there exists an R 2-valued function P(g) whose entries are
convergent power series in a vector u" E R2 with real coefficients such that LP (0) _
8u
12, and the transformation y = P(u) reduces system (VIII.6.1) to dt = Au (cf.
Remark VIII.4.7). This, in turn, implies that the point (0, 0) is also a stable spiral
point of (VIII.6.1) and that the general solution of (VIII.6.1) is y = P(eAC1, where
c" is an arbitrary constant vector in 1R2.
Case 4. If the point (0, 0) is a saddle point of the linear system = Ay, the
eigenvalues At and A2 of A are real and At < 0 < A2. Construct two nontrivial
and real-valued convergent power series fi(x) and rG(x) in a variable x so that
¢(eA'tcl) (t > 0) and ,G(eA2tc2) (t < 0) are solutions of (VIII.6.1), where cl and c2
are arbitrary constants (cf. Exercise V-7). The solution y' = *(eAt tct) represents the
stable manifold of the trivial solution of (VIII.6.1), while the solution y" =1 (eA2tc2)
represents the unstable manifold of the trivial solution of (VIII.6.1). The point (0, 0)
6. ANALYTIC SYSTEMS IN R2 257
is a saddle point of (VIII.6.1). In the next section, we shall explain the behavior of
solutions in a neighborhood of a saddle point in a more general case.
dy
Case 5. If the point (0, 0) is a center of the linear system - = Ay, both eigenvalues
at
01
of A are purely imaginary. Assume that they are ±i and A = [0
J.
Set
y" = 2
1i
1 v, where v =
[t1]. Then, the given system (VIII.6.1) is changed
L J 2
to
du Ig(vl,v2)l
(VIII.6.2)
dt - 9(27)
012, v1)
where
+00
g(tvl, v2) = ivl + 9P'9111 2, 012,V0 = -iv2 + p.9v2v1
p+9=2 p+9=2
(VIII . 6 . 3)
du1
dt = i W ( ulu2 ) u1,
d = -iC
due
0 ( u1U2 ) U2,
+oo
where w(z) = 1 + E Wmxm, by a formal transformation
m=1
ul + h(ul,u2)
(T) v = f(17) = u2 + h(u2,u1)
Here,
h(ul,u2) _
+00
hp.qupU2, h(u2,u1) _
+0 _
hP,9u'lU2
p+q=2 p+9=2
In particular, h(ul, u2) can be construced so that
(VIII.6.4) the quantities hp+1,p are real for all positive integers p.
Observation VIII-6-2. We can show that if one of the Wm is not real, then y = 0
is a spiral point (cf. Exercises VIII-14). Hence, let us look into the case when the
+00
Wm are all real. Furthermore, if a formal power series a(t) = 1 + with
m=1
real coefficients am is chosen in a suitable way, the transformation
(VIII.6.5) u1 = a(B11j2)131, u2 = W10002
258 VIII. STABILITY
dul due
(VIII.6.3.1) = iu1, -iu2.
dt dt =
Note that transformation (VIII.6.5) does not change (VIII.6.3.1). Using a trans-
formation of form (VIII.6.5), change h(ul, u2) so that
is a formal solution of system (VIII.6.2) which depends on two real arbitrary con-
stants.
Let
Vt + H(t,
(S) u(t, , ) _
fe-'t
be the solution of system (VIII.6.2) satisfying the initial condition
Note that
+«0
H(t, , ) _ HP,q(t)ef°, H(tto = E
D+9=2 P+v=2
are power series in { and which are convergent uniformly on any fixed bounded
interval on the real t line.
6. ANALYTIC SYSTEMS IN R2 259
0
h(ceic, ee ")e-'tdt = 0,
0
= 0.
Then,
= c + -(c, c) and t; = e + ?(c, c),
where =(c, e) and c(e, c) are convergent power series in (c, e).
Now, we can prove that two formal solutions
+h(ce",Ce-'t)
ce`t
v'(t,c+F(c,e),e+ =2(c' i!)) and
ee-tt + h(&-'t, cent ) j
of system (VIII.6.2) are identical. The first of these two is convergent; hence, the
second is also convergent. This finishes Case I.
Observation VIII-6-4. Let us look into Case II. Assume that system (VIII.6.3)
has the form
dt2 = -iu2(1 +
co(uiu2)"'°)
ee-'t(1+c0(ce)'"0) +
is a formal solution of system (VIII.6.2) which depends on two real arbitrary con-
stants.
Again, as in Observation VIII-6-3, let (S) be the solution of system (VIII.6.2)
satisfying the initial condition (C). Set
+
ee-tt(1+c0(ct)'"0) + h(ee-it(t+co(ct)'"O),ceut(l+co(et)"O))
260 VIII. STABILITY
as formal power series in (c, e). This series has a formal period
27r
T(cc) = with respect to t, i.e.,
1 + co(ce)m0
E(c, c)
(VIII.6.8) v(T(cc), =[c, e], E[c, c])
E(e, C)
Solving (VIII.6.7) with respect to (c, e), we obtain two power series in
c= and e = e( , £).
(VIII.6.9)
Using (VIII.6.9) as equations for P({,£), it can be shown that the formal power
series is convergent. This implies that
1/7+o
C(S,S)C(CS) [(CO) 1/ J
and
`T(cC) +oo
J/ {e + H(s, , ) } e'8(1+co(cc)"° )ds
1 + t he+1,P(cz)p
0
P=1
since the quantity hP+1,P are real (cf. (VIII.6.4)). This proves that C(Cr/ ) is conver-
C(l, 7)
gent. Hence, c(l;, ) and 4) are convergent. Thus, the proof of the convergence
of A g) in Case II is completed.
Thus, it was proved that if all of the coefficients w,,, on the right-hand side of
system (VIII.6.3) are real, the point (0,0) is a center of system (VIII.6.1). The
general solution of (VIII.6.1) can be constructed by using various transformations
of (VII.6.1) which bring the system to either (VIII.6.3.1) or (VIII.6.3.2). Periods
of solutions in t are independent of each solution in Case I, but depend on each
solution in Case H.
7. PERTURBATIONS OF AN IMPROPER NODE AND A SADDLE PT. 261
Remark VIII-6-5. The argument given above does not apply to the case when
the right-hand side of (VIII.6.1) is of C(O°) but not analytic. A counterexample is
given by
2 _
r2 = Y1 + y2+
dt [h(1) h(r) y'
where h(r2) = e-1jr2 sin I I. Using the polar coordinates (r,O), this system is
r
changed to
dt = rh(r2) and
d
and periodic solutions are given by r2 = 1
MR
with any integers nt. This is an
example of centers in the sense of Bendixson (cf. [Ben2, p. 26] and §VIII-10).
For more information concerning analytic systems in JR2, see [Huk5].
Using thepolar coordinates (r, 0) in the g-plane, write the vector g in the form
y=r [c?}. Then, system (VIII.7.1) is written in terms of (r, B) as follows:
where A =
all a12 J and g"(y) = [iW)l Here, use was made of the formula
a21 22 92 M
dr = r de
dy1
cos (9) + dye sin(6), = -dy1 sin(g) + dye cos(9).
dt dt dt dt dt dt
In this section, we consider the case when the two eigenvalues Ai and A2 of the
matrix A are real, distinct, and at least one of them is negative, i.e., A2 > Al
and Ai < 0. Throughout this section, assume that all = Al, a22 = A2, and
a12 = a21 = 0. Then, in terms of the polar coordinates (r, 0), system (VIII.7.1) can
be writen in the form
(VIII.7.5)
1
r [Ai(cos(8))2 + A2(sin (e))2 + (e) + g2(g)sin (e))]
dt =
d9
= (A2 - Ai )sin (0) cos (0) + r (e) + 92(y-)cos (e))
2
if 0 < r < p(e) (cf. Figure 13).
Observation VIII-7-4. If 0 > A2 > Al, there exists a positive number ro > 0
such that dt < Air for 0 < r < ro.
7. PERTURBATIONS OF AN IMPROPER NODE AND A SADDLE PT. 263
Observation VIII-7-5. In the case when 1\2 > 0 > AI, find a real number wo
such that 0 < wo < 2 and tanwo = -1. Then, for any given positive number
2
E > 0, there exists another positive number p(E) > 0 such that
Observation VIII-7-7. In the case when 0 > A2 > A1, the point (0, 0) is a stable
improper node of (VIII.7.1) as t -+ +oo (cf. Figure 17).
Observation VIII-7-8. In the case when A2 > 0 > A1, the point (0,0) is a saddle
point of (VIII.7.1) (cf. Figure 18).
Choose three real numbers A1, A2, and A3 so that Al < 0, A2 > Al, and
A3
(VIII.7.6) 51 >1+ v.
Let µ(y) be a real-valued function such that (a) u is bounded on tit, (b) p is continu-
ouslv differentiable in y" on e - {(0.0)}. (c)U(y) = A3 on Do. and
(d) µ(y') = A2 on DI. Set 9(y) = 0 Then, 9(y) = 0 if y" E Dl.
(l4(y) - A2) Y21-
J
Furthermore, 0 < yl. 11121 5 (1 + ()y1+" if y"" D1, and
(VIII.7.7)
ddll
= -\01- dt = AM + (µ(y) - A21y2
7. PERTURBATIONS OF AN IMPROPER NODE AND A SADDLE PT. 265
(I +e)Yi
FIGURE 19. FIGURE 20.
Observation VIII-7-10. If we assume some condition on smoothness of §(y-),
there is only one orbit of (VIII.7. 1) that tends to the point (0, 0) in the direction
9 = 0 as t - +oc.
Theorem VIII-7-11. Assume that all = A1, a22 = A2, and a12 = a2I = 0, and
that
(1) A2 > \1 and Al < 0,
(2) g(j) is continuous in a neighborhood of
(3)y-.o
lim " = 0,
lye
b2(t)
urn 01(t) = 0,
i t-.+oo
lim 02(t) = 0, lim
t--+co 01(t)
= 0.
Proof.
The existence of such orbit is seen in previous observations. To show the unique-
ness of such orbit, rewrite (VIII.7.1) in the form
I
[:1].
where u = Observe that (a) 161 = 1 if Jul < 1, (b) -L (d) = U g(0, y2)
-0 Y2
= d, and (c) - 1 < 0.
i
Write system (VIII.7.9) in the form
l
- 1 I u + G(yi.u).
Q 1 A2u
92(yiu) - a 91(yl U
G(yi, u) = -
1
1 + -9i(y1iZ)
aiyi
ac
It is easy to show that lim 5:u (y1, u) = 0 uniformly on Jul < 1. Hence, there exists
v --o
a non-negative valued function K(yj) such that
!G(yi, u) - G(y1, v)! < K(yi )lu - vl
whenever Jul < 1, !v! < 1 and !y1! is sufficiently small. Furthermore, slim K(yt)
=0.
Let u = 01(yi) and u = ip2(yi) be two solutions of (VIII.7.9) such that
(1) ti i (yi) and .'2(y1) are defined for 0 < yi < r) for some sufficiently small
positive number 'I,
(2) lim i'1(yi) = 0 and lim i2(yl) = 0.
vl o+ v,
Set '(yi) = 01 (y1) - s (yi ). By virtue of uniqueness, assume without any loss
of generality that y(yi) > 0 for 0 < yi < ri. Then, for a sufficiently small positive r?,
d
y1 -)tt'(yi) for 0 < y1 < i , where 7 = 2 1 I < 0. This implies that
J
d
[y1 "V,(yi)j < 0 for 0 < yi < rj. Hence, 0 < y1 lim yi 'W(yi) = 0 for
dyi Y:- 0
0 < y1 < 17. Therefore, ;1(yi) = 0 for 0 < y1 < r).
The materials of this section are also found in (CL, §§5 and 6 of Chapter 151.
Theorem VIII-8-1. Under the assumption given above, the point (0, 0) is a stable
proper node of system (VIII.8.1) as t --, +oo.
(VIII.8.3)
(VIII.8.4)
ds
-g(z)
Observe that
171
(VIII.8.5) 9(o) = n = ?12 '
the corresponding solution z(s) = et y'(t) of system (VIII.8.4) is the solution of the
initial-value problem ds = r(7)2 f zz2
l
= rj, where
z" C 2 /
(VIII.8.6) r(i) = (n,)2 + (m)2.
Hence,
the following two conditions: (a) r(t) is strictly decreasing as t -+ +oo and (b)
lim r(t) = 0. Therefore, we obtain (c) 0 < r(t) <_ p for t >_ 0.
t-. +00
Since r(t) is strictly decreasing for t _> 0, the variable t can be regarded as a
function of r, i.e., t = t(r), where t(p) = 0 and lim t(r) = +oo. Use r as the
independent variable. Then, (VIII.8.7) becomes
Let 0 = 4D(r) = 0(t(r)) (0 < r < ro) be the solution of (VIII.8.10) satisfying the
initial condition (b(p) = w. Then,
for 0 < r < ro. Now, by virtue of condition (iii), the improper integral
to
r-.0 JP F(s, 4(s))ds exists. Observe that 0(t) = 4(r(t)) = w +
F(s, $(s))ds = lim r
J P
r(t)
F(s, (s))ds and that lim r(t)
t--+oo
= 0. Therefore, lim 0(t)
t-'+oo
=w+
P
0
Thus, we conclude that the point (0,0) is a stable node of
JP
(VIII.8.1) as t +oc.
Step 2. In this step, we prove that the point (0,0) is a proper node. To do
this, for a given real number c, find a solution (r(t), 0(t)) of (VIII.8.7) such that
rim 0(t) = c and t -+oo
t-+oo
lim r(t) = 0. Selecting a sequence {pm : m = 1, 2, ... } of
positive numbers such that 0 < p,,, < ro and lim pm = 0, define a sequence of
functions {cm(r) : in = 1, 2.... } by the initial-value problems
d0
= F(r, 0), 0(Pm) = c (m = 1, 2.... ),
WT
respectively. Those functions are defined for 0 < r < ro, and
Set
for 0 < r < ro. Since the sequence {4D m(r) : m = 1, 2.... } is bounded and
equicontinuous on the interval 0 < r < ro, assume without any loss of generality
that lim fi(r) exists uniformly on the interval 0 < r < ro (otherwise
M +00
choose a subsequence). It can be shown easily that
dr
dt
= )r + 92(17(r))sm(4'(r)), r(0) = ro,
where g(r) = r [c?1]. The function r(t) is defined for t > 0, is decreasing,
and tends to 0 as t -4 +oo. Set 0(t) = $(r(t)). Then, it is easily shown that
(r(t), B(t)) is a solution of (VIII.8.7) such that limo 0(t) = 1i mt(r) = c.
t +0
0
The materials of this section are also found in (CL, §3 of Chapter 151.
1at
Therefore, a positive number ro can be fouind so that r(t) < r(0) exp J for t > 0
if 0 < r(0) < ro. This, in turn, implies that lim r(t) = 0 and that dt < - if
t -+oo
2
b > 0, while de > - 2 if b < 0, whenever 0 < r(0) < ro. Therefore, lim00(t) = -oo
if b > 0 and lim 0(t) = +oo if b < 0, whenever 0 < r(0) < ro. Thus, we conclude
t too -
that (0, 0) is a stable spiral point of system (VIII.7.1) as t -+ +oo. 0
The materials of this section are also found in fCL, §3 of Chapter 151-
Case 2. Assume that b < 0. If there exists a positive number po such that
whenever 0 < p < po, then the point (0, 0) is an unstable (respectively stable)
spiral point as t - +oo (cf. Figures 24-1 and 24-2).
d
dt [y2J [ yuj + yi +y2 I yy1J
dr
since = 0.
d8
Example VIII-10-2. The point (0, 0) is an unstable spiral point of the system
d = [ Y2 + ` y 2I + Y22 yI
dt l Y2 l -Yi J Y2
1
as t - -oc, since dr
dt
= r2, which has the solution r = 1
ro - t
r(0) = and
ro
r +oo when t ro. (Note: t < ro.)
Example VIII-10-3. For the system
dt = - y, d y = x + x2 - xy + cry2,
the point (0, 0) is (a) a stable spiral point if a < -1, (b) a center if a = -1, and
(c) an unstable spiral point if a > -1.
Proof.
Use the polar coordinates (r, 0) for (x, y) to write the given system in the form
(1+rcos8(cos20-cos0sin8+asin28)) = r2sin8(cos28-cos0sin8+asin20).
10. PERTURBATIONS OF A CENTER 273
Setting
too
r(9, c) _ E r,,, (9)c"', where r(0, c) = c,
m=1
and comparing r(2r, c) and c for sufficiently small positive c, it can be shown that
(1 + Or
r1(t) = 1, r2(2r) = 0, r3(2r) = 4
Thus, r(27r, c) < c if a < -1. Therefore, (a) follows. Similarly, r(2r, c) > c if
a > -1. Therefore, (c) follows.
In the case when a x2 - xy - y2 = (x - +2 ) (x - (1-2 ) . Set
W = - (1 Then. a = (1 Therefore, changing x and y by u =
X - (1 + ,/)y
2
and v = x - (1 -
2
)y, the given system is changed to
du dv
Wv(1 + u), = ---u(1 + v).
dt
is independent of t. This implies that in the neighborhood of (0, 0), orbits are closed
curves. This shows that (0, 0) is a center. 0
Example VIII-10-4. The point (0,0) is a center of the system
if (1) b is a nonzero real number, (2) A = 101 1J, (3) the entries of the R2-
"
valued function g"(yj is continuous and continuously differentiable in a neighborhood
of 0, (4) lim = 6, and (5) there exists a function M(y) such that M is positive
g-6 1Y1
valued, continuous, and continuously differentiable in a neighborhood of 0 and that
8(Mf1)(y) + 8(M 2)(y..) = 0 in a neighborhood of 0, where f, (y-) and f2 (y-) are
the entries of the vector MY + §(y).
Proof
The point (0, 0) is either a center or a spiral point. Look at the system
EXERCISES VIII
VIII-I. For each of the following five matrices A, find a phase portrait of orbits
d:i
of the system = AY.
5 1
VIII-2. Find all nontrivial solutions (x, y) = (d(t), P (t)). if any, of the system
-xy2, _ -x4y(1 + y)
dt =
satisfying the condition lim (th(t),tb(t)) = (0,0).
c-.+00
VIII-3. Let f (x, y) and g(x, y) be continuously differentiable functions of (x, y)
such that
(a) f (0, 0) = 0 and g(0, 0) = 0,
(b) (f(x,y),g(x,y)) # (0,0) if (x,y) 0 (0,0),
(c) f and g are homogeneous of degree m in (x, y), i.e., f (rx, ry) = rm f (x, y) and
g(rx, ry) = rmg(x, y). Let (r, 8) be polar coordinates of (x, y), i.e., x = r cos 8
and y = r sin 0. Set F(8) = f (cos 0, sin 8) and G(O) = g(cos 8, sin 8).
(I) Show that
dr
= rm(F(8) cos 8 + G(8) sin 8),
(S) dt
d8
dt
= rm-1(-F(8)sin8 + G(O)cosO).
(II) Using system (S), discuss the stability property of the trivial solution of each
of the following three systems:
(i) = x2 - y2
dt
= 2xy;
4ii
(iii)
d = x4 - 6x2y2 + y4, dy = 4x3y - 4xy3.
EXERCISES VIII 275
VIII-4. Let J be the 2n x 2n matrix defined by (IV.5.2) and let H be a real constant
2n x 2n symmetric matrix. Show that the trivial solution of the Hamiltonian system
a = JHy is not asymptotically stable as t +oo.
V11I-5. Show that the trivial solution of the system
dy'
d = Ay" + 9(t, y-)
0(ro) = To x D(ro) = {(t, yl : 0 < t < +oo, Iyj < ro) for some ro,
(iv) g(t, yj satisfies the estimate
dy = Ay + Ng + 9(y), y = Iyil,
where
(1) A is a negative number,
(2) N is a real constant nilpotent 2 x 2 matrix and N # 0,
(3) the 1R2-valued function y'(yj is continuous in a neighborhood of
(4) 19(y-) 11-- cly-1I+" for some positive numbers c and v in a neighborhood of
CA
Hint. Suppose N = 0 2 . If we set yj = r cos 0 and y2 = r sin 6, we obtain
0 0
where 9"(y'i = {9i]. Hence, if r(0) is small, r(t) is bounded by r(0) and
92(V
lim r(t) = 0 for t >- 0.
t+0 This implies that if 1#(0)I is small, g(t) is bounded
and lim y(t) = 0'. Look at
t-+o
+ +00
e-aae_a'
6(+ = 9(0) +
0
exists.
Now,
(1) if d(+oo) = 0, then d(t) = 0 identically, since, in this case,
d(t) = f+a
e-X$e-ajV9(y(s))ds;
Hence, the point (0, 0) is a stable improper node of the given system.
VIII-7. Determine whether the point (0,0) is a center or a spiral point of the
system
dt = y
+ xy3 - y7, _ -x + xy2 - yg.
Hint. This system does not change even if (t, y) is replaced by (-t, -y).
VIII-9. For the system
d [y,] [yj + 1 [ y2
dt y2 --
- yIn yi + y2 yi
as t-++oc.
EXERCISES VIII 277
dt -r' dt = In r'
VIII-11. Suppose that a solution ¢(t) of a system
dy _
AJ + §(y-)
dt
satisfies the condition
lim
t-r 4(t) = 0
for a positive number r. Show that if
(1) A is an n x n constant matrix,
(2) the n-dimensional vector W(if) is continuous in a neighborhood of 0,
(3) lien 9(Y) = 0,
V-6 !y1
then ¢(t) = 6 for all values oft.
(Note that the uniqueness of solutions of initial-value problems is not assumed.)
VIII-12. Assume that
(i) AI.... .. 1n are complex numbers that are in the interior of a half-plane in the
complex A-plane whose boundary contains A = 0,
(ii) there are no relations \, = plat + P2 1\2 + + pnan for j = 1, ... , n and
non-negative integers p,,... , pn such that pi + - - + pn >_ 2,
(iii) f(y' = yam' ft, is a convergent power series in ff E Un with coefficients
jp1>2
fpEC'.
Show that there exists a convergent power series ((u) _ ul'Qp with coefficients
Inl>2 _
dg
Qp E Cn such that the transformation if = u" + Q(ur) changes the system - _
dt
A#+ f (y-) to
j = Au, where A = diag[.1t, A2,4, ... , A ].
VIII-13. Show that the trivial solution of the system
dx dr
- -st(x +X22) + x2ezl+ 2, dt
-22(22 + 22) _ Xlez'+ 2
is aymptotically stable as t +oo. Sketch a phase portrait of the orbits of this
system.
VIII-14. In Observation VIII-6-2, it is stated that if one of the is not real,
then y" is a spiral point. Verify this statement.
VIII-15. Discuss the stability of the trivial solution of the system
explicitly.
CHAPTER IX
AUTONOMOUS SYSTEMS
has a unique solution y = p(t, y7). System (IX.1.1) is called an autonomous sys-
tem since the right-hand side f (y) does not depend on the independent vari-
able t.
Observe that p(t + r, r) is also a solution of (IX.1.1) for every real number T.
Furthermore, At + r, rl) = p(r, il) at t = 0. Hence, uniqueness of the solution of
initial-value problem (IX.1.2) implies that p(t +r, rt) = p(t, p(-r, rt)) whenever both
sides are defined. For each il, let T(rl) be the maximal t-interval on which the
solution p(t, r) is defined. Set C(rt) = {y" = p(t, rl) : t E Z(rl)}. The curve C(rl) is
called the orbit passing through the point il. Two orbits C(iji) and C(rj2) do not
intersect unless they are identical as a curve. In fact,
279
280 IX. AUTONOMOUS SYSTEMS
(1) S1 n S2 = 0,
(2) S1 U S2 L+(rf)
Set d = distance(S{{l, S2) = min{lyl - y21 yl E S1, y2 E S2}. Note that d >
1. :
0. Set also So = g: distance(y, SI) = Then, So is not empty, bounded,
and closed. Furthermore, So n C+(i) =20. Choose two points y"I E SI and
y ' 2 E S 2 and t w o sequences {tk : k = 1, 2, ... } and {sk : k = 1,2,...) of
lim tk = +oo, lim sk = +oo,
real numbers so that tk < Sk (k = 1, 2, ... ), k-+oo k-+oo
limp P(tk, y"1, and k lim G p1sk, r7) = 92. Assume that r), SO <
k
2 and distance(p(sk, r), SI) > Then, there exists a Tk for each k such that
2.
tk < Tk < sk and p-(Tk, rl) E So. Choose a subsequence (at : e = 1,2.... ) of
{rk : k = 1, 2, ... } so that lim at = +oo and lim p(at, r) = # exist. Then,
Y E So n C+ (Y-)) = 0. This is a contradiction.
(5) C+ (1-7) is invariant: It must be shown that if ff E C+ (r), then p1t, y) E L+(1-7)
f o r all t E 11(y). In fact, there exists a sequence {tk : k = 1, 2, ... } of real numbers
such that Iii o tk = +oo and k lim o P-14, r) = 17. From (IX.1.1) and the continuity
k
of P(t,y) of y", it follows that for each fixed t.
li tk, n) = UM P(t, Pptk, r1)) = Pit, y) E L+(7). El
k +co k
The materials of this section are also found in [CL, Chapter 16, §1, pp. 389-3911
and [Har2, Chapter VII, §1, pp. 144-1461.
that there exist a mat number t and a positive number r such that 0 < r < ro,
Dt C {y : Iy1 < r}, and Vg(y-) f (y) < 0 on De. Then, L+ (1-7) C Me for all n E Vt.
Proof.
Set u(t) = V(p(t,rl")). Then,
dot)
= V9(r(t,r7))' dtp(t,i) =
du(t)
Therefore, < 0 as long as p(t, t) E Vt. This implies that u(t) < V(1-7) < e
dt
for r) E Vt as long as p(t, n-) E Vt. Hence, p(t, rl E Dt for t > 0 if ij E Vt.
Consequently, Ip(t, rlI < r for t > 0. It is known that C+ (171 is nonempty, bounded,
closed, connected, and invariant (cf. Theorem IX-1-4). Furthermore, C Vt.
d t)
Let po be the minimum of V(rl') for Iy7 < r. Then, po < u(t) and < 0 if
it E 1)t. Hence, lim u(t) = uo exists and no > po. This implies that V(y) = uo
for all y E C+(t)). Since C+(71 is invariant, V(p(t, y)) = uo for all y E L+(171 and
t > 0. Therefore, VV(p(t, y)) &I t, y)) = 0 for all y E C+ (q) and t > 0. Setting
t = 0, we obtain Vc(y) f (y) = 0 if y E C+(rl, i.e., C St if i E Vt. Hence,
C+() C Mt for if E Vt.
The following theorem is useful in many situations and it can be proved in a way
similar to the proof of Theorem IX-2-1.
Theorem IX-2-2. Let V (y) be a real-valued and continuously differentiable func-
tion for ally E lR' such that VV(y) f (y) < 0 for all y" E R". Assume also that the
orbit p(t, rte) is bounded fort > 0. Set S = 1g: Vi(y) f(y) = 0) and let M be the
largest invariant set in S. Then, L+ (7-1) C M.
The proof of this theorem is left to the reader as an exercise.
In order to use Theorem IX-2-2, the boundedness of p(t, must be shown in
advance. To do this, the following theorem is useful.
Theorem IX-2-3. If V (y) is a real-valued and continuously differentiable function
for all y" E lR^ such that VV(y) f(y) < 0 for Iyi > ro, where ro is a positive number,
and that lim V (y) = +oo, then all solutions p(t, r)7) are bounded for t > 0.
1Q1+00
Proof.
It suffices to consider the case when p(to,rl > ro for some to > 0. There are two
possibilities:
(1) IP-(t,r1)I > ro fort > to,
(2) IP(t,n')I > ro for to < t < t, and Ip1ti,17)1 = ro for some tl > to.
Case (1). In this case, dt V (p( t, r)1) = Vj(p(t, rl) f (p(t, 4-7)) < 0 fort > 0. Therefore,
V (r(t, V (r(to, t))) for t > to. Hence, p(t, tt") is bounded for t > 0.
Case (2). In this case, it can be shown that
V(p(t, 71)) <_ max IV(P(to,' )), max(V(y) : 1yi <_ ro)1
3. ORBITAL STABILITY 283
in a way similar to Case (1). The details are left to the reader as an exercise. 0
[y2] = yl +y2
d. [ - 2]
passing through the point ij = I ] is given by
(IX.3.1) = f (y,
284 IX. AUTONOMOUS SYSTEMS
di7 Of
(IX.3.2) =
is called the first variation of system (IX.3.1) with respect to the solution pit, rlo).
The coefficients matrix of (IX.3.2) is periodic in t of period 1.
Let pi, p2, ... , p be the multipliers of (IX.3.2) (cf. Definition IV-4-5). Since
2 p-(t,
(pit, o)) d p1t, ijo), linear system (IX.3.2) has a nontrivial periodic
solution fjt, im). This implies that one of the multipliers must be 1. Set pl = 1.
The following theorem gives a basic sufficient condition for orbitally asymptotic
stability.
Theorem IX-3-4. If n - 1 multipliers p2,... , p, satisfy the condition ipiI < 1
(j = 2,... , n), then the periodic orbit C(rjo) is orbitally asymptotically stable as
t-4+oo.
Prof.
We prove this theorem in five steps. It suffices to find an (n - 1)-dimensional
manifold M in a neighborhood of the point so that
(1) M is transversal to the orbit C(i) at ilo; i.e., the tangent of C(yo) is not in
the tangent space of M at ilo,
(2) there exist two positive numbers K and r such that
FIGURE 1. FIGURE 2.
Step 1. There exists an invertible n x n matrix P(t) whose entries are real-valued,
continuously differentiable, and periodic in t of period 1 (or 2) such that the trans-
formation
(1X.3.3) 6 = P(t)v"
3. ORBITAL STABILITY 285
changes (IX.3.2) to
(IX.3.4)
dt =
AV, A = [ 0 B,
where B is a constant (n - 1) x (n - 1) matrix (cf. Theorems IV-4-1 and IV-4-3).
Since the absolute values of n -1 multipliers are less than 1, there exist two positive
numbers Co and a such that
(IX.3.5) Iexp[tB] I < Coe-tot for t > 0.
A fundamental matrix solution of (IX.3.2) is given by
0
exp[tA] = [1
1
1/(t) = P(t) exp[tA],
o exp[tBj J
This implies that the first column vector of P(t) is a periodic solution of (IX.3.2)
and all the other columns of W(t) are not periodic. Hence, assume that
Se of
I1+ K( (1 + ) < ko1 1e-OL
Next, set lIJt ({) = sup (eot 1 (t, 4)1 : t > 0) if the entries of an R"-valued func-
tion t(i(t, £) are continuous for (IX.3.16) and l}} < kolle-ot for (1X.3.16).
Then,
I
J+9i(s,t,ii(s,())ds - ft 9i(s,tG2(s,&ds l
+oo
< K(ko at
< 2IItLI - 1211(&-ot
3. ORBITAL STABILITY 287
and
f exp[(t-s)B]92(s,))ds
t
- f+ exp((t-s)B](s,(IV)
<
Q
<- 2[[i1 - &IOe-oc
for (IX.3.16) if the entries of R'-valued functions 1(t, )) and (t, t; are continuous
for (1X.3.16) and that J ,(t,£)I < (j = 1,2) for (IX.3.16).
ko1&-OC
15m(t, =
L gm(t,
f= g1(s,(s,))ds,
ao
Then, it can be shown without any difficulties that lim >im(t, exists
m --+oc
uniformly for (1X.3.16) and the limit %i(t, { is a solution of integral equations
(1X.3.14) satisfying condition (IX.3.15) for (IX.3.16). Note that i(O,t;)
0
where a(t) = f gl (s, z/i(s, })ds. This implies that
00
Step 5. Set
00,6 = Plt,vio)
Then, is a solution of system (IX.3.1) such that
(IX.3.18) At, f) - pjt, i o)[ 5 Ko{f [e-°` for (IX.3.16),
where Ko is a positive constant. Define an (n - 1)-dimensional manifold M by
M = (g=4(0'6: 01<<601-
Using (IX.3.6), (IX.3.17), (IX.3.18), and (1X.3.11), we obtain
Thus, the manifold M is transversal to the orbit C(fjo) at rp. Note that (t, t
At, ¢(0, 6).
Moreover, by continuity of the solutions of (IX.3.1) with respect to the initial
conditions, there exist two positive numbers 61 and ro such that for i satisfying the
condition
11 -401
there is a real number r(q such that 1r(q-)l < ro and p(r(g71,'7) E M. Therefore,
(IX.3.18) implies that
(i) the entries of vectors 4"4(t) (j = 2,3) are continuously differentiable and peri-
odic in t of period 1,
(ii) three vectors d
Q , 4'2(t), and 43(t) form an orthogonal system.
dp(ttrlo)
Note that = f (plt, no)).
is the plane which is perpendicular to the orbit at piT,, o). Letting t, u1, and
u2 be three functions of r to be determined, set
dge(r)
dt {/,-/
f(f(t))dT = f(P(r>io)) +
du1
42(r) +
due _
+ u1 + u2 dQ3(r)
dT dT 93(r) d-, dT
y'
This yields
(IX.3.20)
df11 dt ( 42(T) 1
((t) d93(r)1
W- 2(T) f(y(t))dr - 2(t)' dr l u1 - dT J u2,
due dt
((t) dQ2(T)\
dr u1 - 93(t) dg3(r))
dr
and
where a" b denotes the usual dot product and we assumed that q"1(r} = 1 (j =
2, 3).
Note that
dt
= 1 + 00U11 + Iu21)
77.
290 IX. AUTONOMOUS SYSTEMS
and
(IX.3.22)
dul
42(r) ' l ul + [i(T). (r))1 U2
dr
d9drr))
- (o) . _d_r) Jut - (0) - u2 + O(lul l + 1U21),
Let Q(t) be the 3 x 3 matrix whose column vectors are (at, qo)), fi(t), and
,fi(t), i.e., Q(t) = [ f (p1t, rlo)) 6(t) q"3(t)! . The transformation w = Q(t)v changes
the linear system
dtr,
(S2) _ L(PIt,rl *6
to
0 a21(t) a22(t)
Using these notations, write (IX.3.22) in the form
dul
d7-
= all(r)ul + a12(r)u2 + 91(T,ul,u2),
(IX.3.23)
1 dug
= a21(r)u1 + a22(r)u2 + 92(T,u1,u2),
dr
where
j
du' = [all(t) a12(t)
a21(t) a22(t)
U
are also multipliers of system (S2). Therefore, using system (IX.3.23), Theorem
IX-3-4 can be proven. In general, we obtain more precise information concerning
the behavior of solutions in a neighborhood of a periodic solution in this way.
The materials of this section are also found in 1CL, Chapter 13, §2, pp. 321-3271.
4. THE POINCARE-BENDIXSON THEOREM 291
Observation IX-4-3. For every transversal t and every point i , the set tnC+(
contains at most one point (cf. Figures 4-1 and 4-2).
I
PU,
If C+ (, ) does not contain any stationary points, then (i) follows (cf. Lemma
IX-4-4). If C+(no) consists of stationary points only, we obtain (ii), since C+(ilo) is
connected. If f-+ (Q contains stationary and nonstationary points, then C+(7-) for
any point it E C+(ilo) does not contain nonstationary points (cf. Lemma IX-4-4).
This is true also for t < 0. Hence, (iii) follows.
Observation IX-4-5. In cases (i) and (iii), the set 1R2 - C+(ilo) is not connected.
Furthermore, if an orbit C(i)) is contained in C+(i o), two sides of the curve p1t, tl")
belong to two different connected components of R2 - C+(t ). In fact, if we consider
a simple Jordan curve C which intersects with the orbit C(tl at n transversally, then
the curve p(t,,o) intersects with C in a neighborhood of two distinct points on C
infinitely many times (cf. Figure 5).
Theorem IX-4-6. If C(ijo) n C+(ilo) 0 0, then C+(ijo) = C(rjo) and either no is
a stationary point or the orbit C(ilo) is periodic.
Proof
In this case, C(ilo) C C+(ilo). If C+ (8o) contains nonstationary points, the orbit
C(i o) consists of nonstationary points. Choose a transversal a at ilo. Then, it can
be shown that C(ilo) is periodic in a way similar to the proof of Lemma IX-4-4,
since r1o E C+(tlo).
- __11
n2=0
(0.0)
FIGURE 7. FIGURE 8. FIGURE 9.
For 0<r<s<1,set
if T = s,
Then, v(T, s) is continuous in (T, s) for 0 <_ r < s < 1. Hence, studying how 16(0, s)
changes from s = 0 to s = 1 and how i6(T,1) changes from T = 0 to T = 1, we
obtain Ij.(.)(C) = 1. 13
5. INDICES OF JORDAN CURVES 295
di
(IX.5.1) f(+1+
dt
then 1 f-(C) = 1.
(IX.5.2) 1 f-(C) =
E-H+2
2
Sketch of proof.
Let rj(r,) and fj(rk) be two consecutive contact points such that T. < rk. Then,
(i) if both of these two points are exterior contact points, then the tangent to C
changes 7r in angle more than the vector field f does from the point #(T.) to
the point fj(rk) (cf. Figure 11-1),
(ii) if both of these two points are interior contact points, then the vector field f
changes it in angle more than the tangent to C does from the point #(T,) to
the point it(rk) (cf. Figure 11-2),
(iii) if these two points are an exterior contact point and an interior contact point,
then the tangent to C and the vector field f change in the same amount in
angle (cf. Figures 12-1 and 12-2).
Since the total amount of change of the tangent to C in angle is 27r (cf. Observation
IX-5-6), we arrive at formula (IX.5.2). 0
q(
(IX.5.3)
E--H+2
2
on the y"-plane. It is easy to see that, if zo = 171 +ir12 is a zero of p(z) of multiplicity
m, then if#?) = m, where AM _ R(p(yl + iy2)] and i) = [112, 111
. For example,
%P(YI + iy2)]
if p(z) = iz2, system (IX.5.4) becomes
dr
r dO rPF2(0),
dt = rPFI (0), dt
where
F1(0) = f 1(cos 8, sin o) cos 0 + f2(cos 0, sin 0) sin 0,
Sl F2(0) = -f1(cos0,sin6)sin0 + f2(cos0,sin0)cos0
(cf. Exercise VIII-3). If a ray to is defined by F2(0) = 0, then to is an orbit of the
system dff = f (r))). Thus, the entire y"-plane can be divided into sectorial regions
by those orbits QB determined by equation F2(0) = 0. For example, in the case of
system (IX.5.5), we obtain Fl (0) = - sin 0 and F2(0) = cos 0. Observe that
EXERCISES IX
IX-1. For each of the following three systems, using the given function V (x, y),
show that all orbits are bounded as t -. +oo .
EXERCISES IX 299
Y2;
(1) _ -xy2 - 4y, _ -yx2 + 3x, V(x,y) = 3x2 +4
Y2;
(2) = y, = -x3 - y, V(x,y) = x4 +2
dy
= y, = - (xs - 3x4 + 2x3 + 120x2 - 23x + 5) - (1 + x2)y,
IX-2. Show that every solution y(t) and its derivative L(t)
(t) of the differential
where f j and f2 are continuously differentiable on the entire (yj, y2)-plane. Assume
that
(1)
fl(yl,y2)>0 for y2>0 and -oo<yl<+oo,
fi(yi,y2) < 0 for y2<0 and - oo < yl < +oo,
(2) fj(yi,1) > 1 and fi(yi, -1) < -1 for -oo < yl < +00,
(3) f2(yi,1) = 0 and f2(yi, -1) = 0 for -oo < y1 < +oo,
(4)
2 - 1)
f2(11j,y2) <_ YI(112 for jy2j < 1 and 0 < yi < +oo,
1
f2(Y1,Y2) ? Y1 (Y22 - 1)
for jy21 < 1 and - oo < yi < 0,
is bounded as t -+ +oo.
Case 2. The solution (y1(t),y2(t)) of (SI) that satisfies the initial condition (C) is
unbounded as t +oo.
In Case 1, G+((nj, n2)) is either {(0, 0)} or a periodic orbit. In Case 2, G+((nj, 92))
is{(x,±1):-oo<x<+oo}.
300 IX. AUTONOMOUS SYSTEMS
Examples.
(a) Every orbit in Iy21 < 1 of the system with fl(yl,y2) = y2 and f2(yi,y2) _
yl(y2 - 1) is periodic.
(b) The stationary point (0,0) is a stable spiral point if fl(yl,y2) = y2(2 -
sin(yly2)) and f2(yj,y2) = 2yi(y2 - 1).
(c) The stationary point (0, 0) is an unstable spiral point if ff(yi, y2) = y2(2 +
sin(yiy2)) and f2(yi,y2) = 2y, (y22 1). -
Verify these statements by using the function V(yl, y2) = yi - ln(1 - y2).
IX-4. Let us consider a system
where
(1) the entries of the 1R2-valued function f (y-) is continuously differentiable on the
entire y-plane,
EXERCISES IX 301
Y1 - y2 y2
find the total number E of elliptic sectorial regions, the total number H of hyper-
bolic sectorial regions, and the total number of parabolic sectorial regions in the
neighborhood of the isolated stationary point 0 of the system = f (y, E) in the
dt
following two cases: (i) E = 2 and (ii) E _ 2. Also, find I y(0) for e 96 0.
IX-8. Let us consider a system
(S3) dt = f (y),
where the entries of the l 3-valued functionf is continuously differentiable on the
entire y-space R3. Assume that (S3) has a periodic orbit pit,ip) of period 1 such
that f (r'(t, o)) 0 0. Assume also that the first variation of system (S3) with respect
to the solution p(t,' o), i.e.,
d9 Of
(r(t,io))u,
dt
has three multipliers p1 = 1, p2i and p3 satisfying the condition: 1P21 < 1 and jp3j >
1, respectively. Construct the general orbits pit, r) of (53) such that distance(At,17),
C(s'7o)) tends to0ast-+too.
drt
IX-9. Show that the differential equation d-t2 + (x + 3)(x + 2) + x(x + 1) = 0
does not have nontrivial periodic solutions.
Hint. Two stationary points are a node (0,0) and a saddle (-1,0). Furthermore,
setting f(x1,x2) = x2 , we obtain divf(x1,x2) _
-(xt + 3)(x1 + 2)x2 - xt(xt + 1) J
I
-(x1 + 3)(x1 + 2) < 0 if x1 > -1. Also, use index in §IX-5.
IX-10. For the system
IX-13. Let f (x, y) and g(x, y) be real-valued, continuous, and continuously dif-
ferentiable functions of two real variables (x, y) in an open, connected, and sim-
ply connected set D in the (x, y)-plane such that 8f (x, y) + Lf (x, y) j4 0 for all
dxt
(x, y) E D. Show that the system = f (x, y), d = g(x, y) does not have any
nontrivial periodic orbit that is contained entirely in D.
IX-14. Let p(z) be a polynomial in a complex variable z and deg p(z) > 0. Set
x = 3t(z] and y = Q [z]. Verify the following statements.
(i) In the neighborhood of each of stationary points of system
there are no hyperbolic sectors. Also, system (A) does not have any isolated
nontrivial periodic orbit.
(ii) In the neighborhood of each of stationary points of system
(B)
dt
= 3R[p(z)], L = -`3'[p(x)],
there are no elliptic sectors. Also, system (B) does not have any nontrivial
periodic orbits.
EXERCISES IX 303
where the entries of the R'-valued function f are analytic with respect toy in a
domain Do C R'. Assume that
(i) system (S2) has a periodic orbit pit, ijo) of period 1 which is contained in the
domain Do,
(ii) AP-Tt,no)) 0
(iii) for any open subset V of Do which contains the periodic orbits p(t, i"p), there
exists an open subset U of V which also contains p'(t, o) such that for any
point i in U, the orbits p'(t, i of (S4) is contained in V and periodic in t.
Show that if U is sufficiently small, for any point i in U, we can fix a positive period
of p1t, y) so that is bounded and analytic with respect to i in any simply
connected bounded open subset of U.
Hint. Apply the following observation.
Observation. Let Do be a connected, simply connected, open, and bounded set in
1Rk and let T2 (j = 1, 2, ...) be analytic mappings of Do to l . Suppose that, for
any pointy E Do, there exists a j such that T3 [yl = y, where j may depends on Y.
Then, there exists a jo such that Tjo [yj = y for all y E Do.
Proof.
Set
E j _ {y E D o : T j [ y 1 = Y - ) . j = 1, 2, ... .
Then,
(1) E. is closed in Do,
+00
(2) Do = U E,,
3=1
(3) Do is of the second category in the sense of Baire.
Hence, for some jo, the set Ego contains a nonempty open subset (cf. Baire's
Theorem). Since Tea is analytic, we obtain Tao [ 7 = y" for all y" E Do. O
For Baire's Theorem, see, for example, [Bar, pp. 91-921.
CHAPTER X
In this chapter, we explain the basic results concerning the behavior of solutions
of a system
112
[-h(yi)112 - 9(111)
as t -- +oo. In §X-2, using results given in §IX-2, we show the boundedness of
solutions and apply these results to the van der Pol equation
(E) x + E(x2 - 1)
d +x=0
(cf. Example X-2-5). The boundedness of solutions and the instability of the unique
stationary point imply that the van der Pol equation has a nontrivial periodic
solution. This is a consequence of the Poincar&$endixson Theorem (cf. Theorem
IX-4-1). In §X-3, we prove the uniqueness of periodic orbits in such a way that it
can be applied to equation (E). In §X-4, we show that the absolute value of one
of the two multipliers of the unique periodic solution of (E) is less than 1. The
argument in §X-4 gives another proof of the uniqueness of periodic orbit of (E). In
§X-5, we explain how to approximate the unique periodic solution of (E) in the case
when a is positive and small. This is a typical problem of regular perturbations. In
§X-6, we explain how to locate the unique periodic solution of (E) geometrically as
e - +oo. In §X-8, we explain how to find an approximation of the periodic solution
of (E) analytically as a +oc. This is a typical problem of singular perturbations.
Concerning singular perturbations, we also explain a basic result due to M. Nagumo
[Na6] in §X-7. In §X-1, we look at a boundary-value problem
Using the Kneser Theorems (cf. Theorems 1II-2-4 and III-2-5), we show the exis-
tence of solutions for this problem in the case when F(t, y, u) is bounded on the
entire (y, u)-space. Also, we explain a basic theorem due to M. Nagumo [Na4] (cf.
Theorem X-1-3) which we can use in more general situations including singular
perturbation problems (cf. [How]).
For more singular perturbation problems, see, for example, [Levi2], [LeL], (FL],
[HabL], [Si5], [How], [Wasl], and [O'M].
304
1. TWO-POINT BOUNDARY-VALUE PROBLEMS 305
Theorem X-1-1. If the function F(t, yt, y2) is continuous and bounded on a region
11 = {(t, Y1, y2) : a < t < b, lyt I < +oo, Iy2I < +oo}, then problem (X.1.1) has a
solution (or solutions).
Proof.
For any positive number K, the set Aa = {(a, or, y2) : Iy21 < K} is a compact
and connected subset of ft We shall show that A0 satisfies Assumptions 1 and 2
of §111-2 for every positive number K. In fact, writing the second-order equation
(X.1.1) as a system
dyt _ dye
(X.1.2) - Y2, = F(t, yt, y2),
dt dt
we derive
y1(t) = y1(a) + J y2(s)ds,
a
where c is a certain point in the interval 10. Since I ddt2 (c) M, the quantity
yI(b)I can be made as large as we wish by choosing Iy2(a)I sufficiently large. Thus,
there are two points (nt, n2) and ((1, (2) in Sb such that (X.1.3) is satisfied.
Since the set Sb is compact and connected, there must be a point (Q, () in the
set Sb. This implies the existence of a solution of problem (X.1.1). 0
306 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
and
(X.1.5)
+ p-+-, =
0, y(a) = a, y(b) = Q
dt2
However, Theorem X-1-1 does not apply to
For more general cases, the following theorem due to M. Nagumo (Na4] is useful.
Theorem X-1-3. Assume that
09 f
(i) a real-valued function f (t, x, y) and its derivatives az and are continuous
in a region V = {(t, x, y) : (t, x) E A, -oo < y < +oo}, where 0 is a bounded
and closed set in the (t, x) -space;
(ii) in the region D, the function f satisfies the condition
(I) 1f(t,x,y)1 : 0(IyI),
where 0(u) is a positive-valued function on the interval 0 < u < +oo such
that
+°° udu
(II) +00;
(iii) two real-valued functions wi (t) and w2(t) are twice continuously differentiable
on an interval a < t < b and satisfy the conditions
wi(t) < w2(t) for a < t < b,
Ao = {(t x)-a<t<b wi _(t) < x<w2(t)}co,
and
d2wi (t) > f 1 t, wi (t), d
2 dt t) )
(IV)
d2dt d
2 (t) <f t, W2 (t), dt t)) , for a < t <b;
(iv) two real numbers A and B satisfy the condition
(V) wi(a) < A < w2(a), and wi(b) < B < w2(b).
Then, the boundary-value problem
Lemma X-1-4. Let x(t, to,rl) be the solution of the initial-value problem
d2x d .)
= f (t, x, , x(to) x'(to) = n,
where a < to < b, (to, l;) E Ao. Then, for any given positive number M, there exists
a positive number a(M) such that l x'(t, to, l;, il) j < a(M) if
(X.1.7) jr71 < M and (T, x(r, to, e, rl)) E Do for to < r < t or t:5 r < to.
Proof.
Letting L be a positive number such that
Suppose that there exist ri and r2 such that to < Ti < r2 < t and that
x'(r1) = M < x'(r) < x'(r2) = a(M) for r1 < r < T2,
where x(T) = x(r, to, t ,17). Then, since x'(r) > 0 for ri < r < r2, it follows that
xO(X'x
x/ (7-) for T1 < -r < -r2.
(T ))) <
Hence,
a(M) u/du
f 0u) -
This contradicts the choice of L by (X.1.9). Therefore, Lenuna X-1-4 is true for
to < r < t. We can treat the case t < r < to similarly, since if we change t by -t,
the differential equation x" = f (t, x, x') becomes x" = f (-t, x, -x'). 0
Lemma X-1-5. Set
where e is an arbitrarily fixed positive number. Let also x(t, c) be the solution to the
initial-value problem
d2x
= f (t, x, , x(a) = A, i (a) =
)
308 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
Then,
(1) two curves x = x(t, c) and x = w2(t) meet for some t on the interval a < t < b
if c>'Y;
(2) two curves x = x(t, c) and x = wl (t) meet for some ton the interval a < t < b
if c < -ry.
Proof.
La
For part (1), by virtue of Lemma X-1-4, x'(r,c) > if (r,x(t,c)) E Ao for
a < r < t. The proof of part (2) is similar.
Proof of Theorem X-1-3.
Now, let us complete the proof of Theorem X-1-3. The main point is that when
two curves x = x(t,c) and x = w2(t) or two curves x = x(t,c) and x = w, (t) meet,
they cut through each other. So look at Figure 1.
(b, B)
(a. A)
t=a r=b
FIGURE 1.
Example X-1-6. Theorem X-1-3 applies to the boundary-value problem
d2x _
(X.1.10) x(0) = A, x(1) = B
dt2 Ax'
d [yj __ Y2
(X.2.1)
dt y[-h(yi)y2 - 9(yi)
as t
l)y2- 9(y1)
+oo. Set y = I yl
J
and f (yl = [_h(yi
h(x), g(x), and dd(x) are continuous with respect to x on the entire real line R.
JLet us assume that
Also, we denote by p(t,r) the solution of (X.2.1) satisfying the initial condition
y(0) = n
Fi rst set V (y) = + G(yl ), where G(x) = fox g(s)ds. Then,
2Y2
Y2 = 0.
Observation X-2-1. Denote by M the set of all stationary points of system
(X.2.1), i.e., M = {9: g(yl) = 0, y2 = 01. Then, M is the largest invariant set in
S if the following three conditions are satisfied:
(1) h(x) > 0 for -oo < x < +oc,
(2) h(x) has only isolated zeros on the entire real line IR,
(3) g(x) has only isolated zeros on the entire real line R.
The proof of this result is left to the reader as an exercise (cf. Figure 2, where
0, 0 and 0).
By using Theorem IX-2-2, we conclude that lim p(t, y) = 17 E M if conditions
t-+00
(1), (2), and (3) are satisfied and if the solution p(t,g) is bounded for t _> 0. Note
that G+(rt) is a connected subset of M.
In Observation X-2-1, the boundedness of the solution p(t, i) for t _> 0 was
assumed. In the following three observations, we explore the boundedness of all
solutions of (X.2.1). Set
G(x) = o
Jo
g(s)ds and H(x) = f
o
X
h(s)ds.
(X.2.2)
d [zl] Z2 - H(z1)
dt 22 -g(z1)
by the transformation
Y1 = Z1, Y2 = z2 - H(zi).
Denote by qqt,() the solution of (X.2.2) such that z(0) _ . Set V, (z-) = zz2 +
G(z1). Then,
L9 V,
ay - [g(zl), z21, ,P = -g(z1)H(zl)
8z"
Note that g(x)H(x) > 0 for -oo < x < +oo and that
dt V1(ggt,C))
- z(g-(t,C)) F(glt,S)) S 0 for t > 0.
-
Y2=0 z2=0
{s2, 0) (43,0) (41.0)
di-1 z1=0
FtcuRE 2. F1cuRE 3.
2. APPLICATIONS OF THE LIAPOUNOFF FUNCTIONS 311
Observation X-2-4. Every solution p(t, >)1 of (X.2.2) is bounded for t > 0 if
(i) = lim
+00
H(x) = +oo, (ii) = --Cc
lim H(x) = -oo, (iii) g(x) > a for x > as > 0,
and (iv) g(x) < -a for x < -ao < 0, where a and ao are some positive numbers.
Proof.
In Observation X-2-3, the Liapounof function
OV2
09
_ [[y2 + H(yi) - k(yi )J {h(yi) - dyyt )1 + 9(yi ), 1
Y2 + H(1h) - k(yi) I
and J
(Y ')
J
az {y22
dy, + [H(yi) - k(yi)J y2}
- 9(yi) [H(yi) - k(yl)J
Using (i) and (ii), three positive numbers M, a, and c can be chosen so that
fa [H(x) - cJ > M for x > a > ao,
-a [H(x) + c] > M for x < -a < -ao.
Also choose a function k(x) so that
{c for x > 2a,
(1) k(x) =
-C for x!5 -2a,
dk(x)
(II} Jk(x)J < c and >0 for - on < x - +oo ,
dx
and
dk(x) >
m > 0 for JxJ < a
dx
for some positive number m (cf. Figure 4).
k
k=c
k=-c
x=-2a x=-a x=O x=a x=2a
FIGURE 4.
312 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
In fact,
2 + e(x2 - 1) dt + x
= 0,
where a is a positive number, h(x) = e(x2 - 1) and g(r) = x. Hence,
t
lim
2Since V (y-)
+oo
a
ev. f =
WY=
exists for every solution p-(t, ii) of (X.2.4). Now, observe that
(1) We must have a < 1. Otherwise we would have y2(t,,)2 > 2(a+cos(yi(t,rte))
for t > 0. This implies that dt V (pl t, r7-')) < -2a(o - 1) < 0. This contradicts
(X.2.5).
(2) If -1 < a < 1, the solution p(t, 7) must stay in one of connected components
of the set {y" : V(y) < a + e < 1} for large positive t. Those connected
components are bounded sets.
(3) In case a = 1, we can show the boundedness of p(t, t) by investigating the
behavior of solutions of (X.2.4) on the boundary of the set {g: V(y-) < 1).
dt V
(zI = 9(zl)[z2 - H(zi)] - z29(z1) = -g(z1)H(z1)
along an orbit of (X.3.2). Hence,
dz2 g(zl) dzi z2 - H(zi)
dzi z2 - H(zl)' dz2 9(z1)
dV 9(z1)H(z1) dV _
H(zi)
dz1 z2 - H(z1)' dz2
that
[
z1(ao, ao) = 0]
and z2 (t, ao) >0 for 0:5t < oo
for some positive number co, where a is the unique positive zero of H(x) given in
condition (v) (cf. Figure 5).
0
z(r(a), a) and 0 < zi (t, a) :5a for 0< t < r(a)
-Q(a)
if 0 < a < ao. Also, since H(x) < 0 for 0 < x < a, we obtain
(A) V(z"(r(a),a)) - V(z(0,a)) > 0 for 0 < a < a0 (cf. Figure 6).
(0.ao)
zI = 0
FIGURE 5. FIGURE 6.
Observation 3. If ao < a, there exists a positive number r0(a) such that
z2(t,a) = Z(zl(t,a),a),
where Z(x, a) is a continuous function of (x, a) for 0 < x < a and a > a0, and
continuously differentiable for 0 < x < a and a > ao except for x = a and a = a0.
Furthermore, Z(x, al) < Z(x, a2) for 0:5 x:5 a if a0 < a1 < 02 (cf. Figure 7).
Set 2(x, a) = I Z(z, a) J . Then,
g(x)H(x)
,
dx
d
V(Z(x,a)) - Z(x,a) - H(x) > 0 for 0<x<a
and, hence,
Observation 4. If a0 < a, there exists a positive number ri(a) such that ri(a) >
ro(a), zl (r1(a), a) = a, and zi(t, a) >a for ro(a) <t <7-1 (a) (cf. Figure 8).
(a, z2(rp(a), a)) (a. z2(r0(a), a))
(0. a) O,a
(O.a0) I
z2=H(zt) (
(O.ao)
) z2=H(z0
z2=0 =0
zi =0 zi=a Z, =0
FIGURE 7. FIGURE 8.
Note that z2(ri(a),a) < 0 and that H(zi(t,a)) > 0 for ro(a) < t < 7-1(a) since
zi (t, a) > a on this t-interval. Regarding zi (t, a) as a function of z2(t, a) for
z2(r1(a), a) < z2 < z2(ro(a), a), we obtain
lim
dV =0 uniformly for 0 < z1 < a,
jz21-+oo dzi
z1lim00
a = +oo uniformly for - oo < z2 < +oo,
3. EXISTENCE AND UNIQUENESS OF PERIODIC ORBITS 317
it follows that
lim (V(i(ro(a),a)) - V(z(0,0J = 0,
a+oo
lim (V(z(rl(a),a)) - V(z(ro(a),a))J = -oo,
a+oo
lim JV(z(r(a),a)) - V(i(ri(a),a))J = 0.
a-.+00
Therefore,
Thus, we conclude that g has exactly one positive zero a+, i.e.,
> 0, 0<a<a+,
(X.3.3) 9(a) Zz2(r(a), a)2 - 2a2
= 0, a = a+,
< 0, a>a+
X-4. Multipliers of the periodic orbit of the van der Pol equation
In Example X-2-5, we looked at the van der Pol equation
(X.4.1) LX + e(x2 - 1) d + x = 0,
where a is a positive number. Using Observation X-2-4, it was shown that every
orbit of the system
(X.4.2) d i
[y112 / 112
] dt
[ -E(y1 1)112 - 111 -
is bounded for t > 0. It was also remarked that 0 is the only stationary point of
system (X.4.2) and that the stationary point 0 is not stable as t - +oo. In fact, the
linear part of the right-hand side of (X.4.2) at y = 0 is Ay, where A = [ 01 and
[i]. Since trace[A]=f and det (A) = 1, the stationary point is an unstable
Y21-
node for e > 2, while 0 is an unstable spiral point for 0 < e < 2. Therefore, using
the Poincare-Bendixson Theorem (cf. Theorem IX-4-1), we conclude that there
exists at least one limit cycle. Now, Theorem X-3-1 implies that system (X.4.2)
has exactly one limit cycle and all the other orbits except for the stationary point 6
approach this limit cycle asymptotically as t - +oo. In fact, since h(x) = e(x2 -1),
3
H(x) = e I - x] , and g(x) = x, the seven conditions (i) - (vii) of Theorem X-3-1
3
are satisfied. In particular, the positive zero of H(x) is a = J > 1.
In this section, we prove the following theorem concerning the multipliers of the
unique periodic solution x = x(t, e) of the van der Pol equation (X.4.1).
Theorem X-4-1. The multipliers of the periodic solution x(t, e) of (X.4.1) at 1
and p such that )pI < 1.
Proof.
z
xz
d , then dt = -e(x2 - 1)y2. This implies
11
If we! set v = , wwhere 11 =
2
that eJ (x(t)2 - 1)dt = - / . Now
Now look at Figure 11.
FIGURE 11.
5. THE VAN DER POL EQUATION FOR A SMALL PARAMETER 319
On each of two curves Cl and C2, let us denote y as a function of v by yl (v) and y2(v)
respectively. Note that x2 > 1 on C1, but x2 < 1 on C2. Hence, yl(v)2 < y2(v)2.
This implies that
fT
e (x(t)2 - 1)dt > 0,
where T is a period. Therefore, we can complete the proof by using the following
lemma.
Lemma X-4-2. Assume that two functions f (x) and g(x) are continuously differ-
entiable in R. Assume also that the differential equation
d2
2 + f(x) dt + g(x) = 0
r1
has a nontrivial periodic solution x(t) of period 1 such that f (x(t))dt > 0. Then,
J0
the multipliers of the periodic solution x(t) are 1 and p such that jpI < 1.
Remark X-4-3. If system (X.4.2) has more than one periodic orbit, then at least
one of them must be orbitally unstable. Therefore, the proof of Theorem X-4-1 is
another proof of the uniqueness of periodic orbit of (X.4.2).
(X.5.1)
dt[y2J LyYi
Every orbit of (X.5.1) is a circle and of period 2n in t. We expect that the periodic
orbit of (X.4.2) must be approximated by one of those circles as e -+ 0. The main
problem is to find the radius of the circle which approximates the periodic orbit of
(X.4.2) for a sufficiently small e > 0.
Since the periodic orbit and its period are functions of e, we normalize the
independent variable t by the change of independent variable
(X.5.2) t = (I +eW)T
so that the period of the periodic orbit becomes 27r for every e > 0. Transformation
(X.5.2) changes differential equation (X.4.1) to
x722
+ e(1+eW)(x2-1)d + (1+&,d)2x = 0
that can be written in the form
d 2X
(X.5.3) + x = e(1 + ew)(1 - x2) - e(2W + ew2)x.
22
320 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
d2x
(X.5.4) -T' + x = f(r)
d
is given by
/'r+2n
(X.5.5) x(r) = K cos(r + m) + a- J sf(s)sin(r - s)ds
r
as it is shown with a straight forward calculation, where K and 0 are arbitrary
constants. This solution is periodic in r of period 2ir if and only if
2w 2w
(X.5.6) f (s) sin(s)ds = 0 and f (s) cos(s)ds = 0.
10 J0
Observation X-5-2. In the case when condition (X.5.6) is not satisfied, set
27r 2, f (S)
(X.5.7) S[ f ] = 1 fo f(s) sin(s)ds and C[J] = 1 cos(s)ds.
i 7t J
o
Then,
v(T) = Kcos(r + 0)
(X.5.8) jr+21r 1
2w
I C(K,, E) = -
7r
f (1T cos(s)ds.
5. THE VAN DER POL EQUATION FOR A SMALL PARAMETER 321
m=0
where Sm(K,w) and C,,,(K,w) are polynomials in (K,w) with constant coefficients.
These two power series also converge uniformly for
Observation X-5-4. Inserting series (X.5.11) and (X.5.12) into system (X.5.10),
we obtain
d2vi
+ vi = (1 - K2cos2(r))(-Ksin(r)) - 2wKcos(r)
dr2
- So(K,w) sin(r) - Co(K,w) cos(r)
= K3 cos2(r) sin(r) - K sin(r) - 2wK cos(r)
- So(K,w) sin(1T) - Co(K,w) cos(r)
1
= 4 K sin(3r) + - K - So(K, w)J sin(r)
4
- [2wK + Co(K, w)1 cos(r).
322 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
d [$2 Il
(X.6.2)
dr
1
where a = -
E
6. THE VAN DER POL EQUATION FOR A LARGE PARAMETER 323
point (wl, w2) far away from the curve C : w2 = 3 - w!. This implies that every
orbit moves toward the curve C almost horizontally (cf. Figure 12).
Observation X-6-2. From Observation X-6-1 a rough picture of orbits of (X.6.2)
is obtained (cf. Figure 13).
Note that the slope of an orbit given by (X.6.3) is negative on C1 (O) and decreasing
to -oo as wt increases to a(f3). Also, is 0 on C1(f3).
dw2
Step 2. Let us consider a curve
where
(i) µ(w1,3) is continuous for 1 < wl < a(0) and continuously differentiable for
1 < wl < a($),
(ii) µ(w1, 0) > 0 for 1 < w1 < a(0), and (iii) µ(a(,3), 0) = 0.
On the curve C2(f3),
3
= -0µ(w1,13)
\ 31 - wl
and, hence,
_ /32w1 Owl
W2 _ f wl _ wl /f l µ(w1, 0)
3
On the other hand, the slope of the curve C2($) is
dw2 _ w1
2
- 1 - ,8 dEc(w1,f)
dwl - dwl
This implies that if µ is fixed by the initial-value problem
(X.6.4)
dwl
1, µ(a(0),,8) = 0,
we obtain
/32w1 dw2
for 1 < wl < a($),
1 dwl
W2 - -
C 3 w'/
where dw2 denotes the slope of C2(0). The unique solution to problem (X.6.4) is
i
given by
a(/3)2 - wl.
Thus, we choose the curve
3 - 31 - wl l - /3 a(/3)2 - wi
1w3 2
W2
31 wl)
and, hence,
/32w1 0'w1
(,w3
31 wl/
/ a(Q)2 - w1 + 3
+
(31 - wl
dw2
On the other hand, the slope of the curve C3(/3) is This
dw, 1 a(Q)2 - wl
implies that
02W, <dm2
for 0<w1<1,
w2-
(u, -wI ) dw,
3
where denotes the slope of C3(0) (cf. Figure 17). Note that on the curve
awl
C3(3), W2 = - 3 - Qa(Q) at wI = 0.
Step 4. Now, let us fix the positive number a(/3) > 2 by the equation
3
2
(X.6.5) + 3a(3) = a(3) - a(13).
Denote by C+(/3) the curve consisting of Cl(/3), C2(13), and C3()3), i.e., C+(Q) _
C1(Q) UC2(/3) UC3($). Let C_(0) be the symmetric image of C+(Q) with respect to
the point (0, 0). Then, C+(3) U C_ (/3) is a closed curve if a(,Q) satisfies condition
(X.6.5). We use the closed curve C+(/3) UC_(/3) as a part of the boundary 8V(/3)
of V(/3) (cf. Figure 18).
326 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
b(3) 3
Step 5. Fixing a positive number b($) < 2 so that 3 = - b(j3) +,A(0), we
choose the curve
b(3)3-
r!(,3) = {twiw2) : w2 = b(j3) +,6b($)2 - w;, 0 < w, < b(13)}
as a part of the boundary 8V(j3) of V($) (cf. Figure 19). Note that, on r,(j3),
2
w2=3 at w, = 0.
/32w1 02w1
b(3) 3
w2 Q b(B)2 - wl + - b(8) - (3' w,)
Q wl
>-
b(3) - w,
On the other hand, the slope of the curve 171(3) is dullw2
ap)2'-
wi
Step 6. We use the two curves
r2(0) = {(wiw2): w2
3
3
-wi, 1 <w, <b(S)
ll
r3($) = {(wltw2): U2 -2,
3
0 < wl <
JJ
as parts of the boundary 8V(j3) of V(8) (cf. Figure 20). Note that on r3(Q), the
slope of an orbit given by (X.6.3) is positive and dd is negative.
Step 7. Denote by r+(j3) the curve consisting of r,(8), r2(8), and r3(8), i.e.,
r+w) = r1(Q) u r2(a) u r3(8).
Let r_ (j3) be the symmetric image of r+(j3) with respect to the point (0, 0). Then,
r+(p) U r_(8) is a closed curve. We use the closed curve r+(j3) U r_ (f3) as a part
of the boundary OV(0) of V(8) (cf. Figure 21).
328 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
F (x, Y, I = 0,
d'Yx(x)
and M for a positive constant M on the interval 0 < x < t.
Let y(x) be any solution of the differential equation
to the systern
du dv
(X.7.1)
dx = v, F(X, u, v, a),
where
Do = {(x, u, v) : 0 < x < t, Jul <- a(x), lv) < pe-`= + b(x)}.
Also,
F (riY(x)v + d ()) > Lv for v > 0,
F x, Y(x). v + d ()) < Lv for v < 0,
in Do. Hence, in Do,
(X.7.2))
F(x, u, v, A) < -Lv + Klul + (e + AM) for v > 0,
F(x, u, v, A) > -Lv - Klul - (E + AM) for v 0.
8. A SINGULAR PERTURBATION PROBLEM 329
Step 2. Suppose that two functions wl (x, .A) and w2(x, A) satisfy the following
conditions:
(X.7.3)
10 < wl(x,.A) < a(x), 0 < w2(x,A) < pe-' + b(x),
wl(0,A) > 0, w2(0,.1) > p,
wi (x, A) > W&, A), Aw2(x, A) > -Lw2(x, A) + Kwl (x, A) + (E + AM)
ApA e + AM + K62(e + 1)
where bl = L2 + L , satisfy the requirements (X.7.3) if f, A,
and a positive constant 62 are sufficiently small. Observe that two roots of AX2 +
LX -K=0 are -L -(o and (o = L +O(A).
Step 4. Note that
dy
dx
dY(x) < pe-Lx/A +
dx
fE
L
+
L
(h + M1 ] ex'/L.
L J
Note that
lim a-Lx/A =0 if x > 0.
330 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
(X.8.2) (x2 - 1) d + x = 0.
Solving (X.8.2) with an initial value x(O) = xo < -1, we obtain x = 0(r), where
2 )2 - In I0(r) I = -r + 2 - In(-xo).
Observe that
d0(r)
dr
0(r)
0(7)2-1 >0 if d(r) < -1.
2
Note also that setting ro = 2 - In(-xo) -- 2 > 0, we obtain ,(ro) = -1 and
45(7-)2 - 1 > 0 for 0:5 r < ro. The graph of 0(r) is given in Figure 22.
FIGURE 22.
(i) Behavior for 0 < r < 7-o -bo, where do > 0: Let us denote by x(r, A) the solution
to the initial-value problem
d2z
(X.8.3)
a dr2 +
(x - 1) dx
2
dr- +
x = 0, x(O, A) = xo, 40,A) = 7-7,
where the prime is d7- and q is a fixed constant. Using Theorem X-7-1 (Na6J, we
derive
Ix(r, A) - O(T)I 5 AK,
I Ix'(r, A) - 4'(r)l 5 I7-7 - 0'(0)le-P'1x
+ AK,
8. A SINGULAR PERTURBATION PROBLEM 331
for 0 < T < To - 60, where K and a are suitable positive constants.
(11) Behavior for lx+1l< o: First note that lim p0'(T) _ +oo. Set ¢'(ro-2bo) _
1Pi > 0. Then, there exists T1(A) for sufficiently small A > 0 such that
{ O<ri(A)<ro_So, X
llm T1(A) _ 7.0 - 280,
x(T1(A),A) _ O(Tp - 26p), x'(T, A) > for r1(A) < T < TO - 60.
2P1
1A)
Set p = Then,
0 < p(x, A) < 2p' for ¢(ro -26o) < x < x(ro-b0, A) < -1, 0 < A < AO,
where A0 is a sufficiently small positive number. It is important to notice that if
we make 60 small and if we make A0 also small accordingly, we can make p1 small.
Set
Then,
0 < p(z, A) < 2Mo for ¢(r0-25o) < x < x(ro - 50, A), 0 < A < A0.
Now, we shall show that
(X.8.7) 0 < p(x, A) < 2Mo for Ii + xt < o, 0 < A < A0.
If (X.8.7) is not true, there must exist a such that
p({, A) = 2M0, 0 < p(x, A) < 2Mo for -1-a:5 x<
Then, this implies that
This is impossible.
(iii) Behavior for -1 + or < x < 2 - al: To begin with, it should be remarked that
8
J(2_1)d
[..-_]2 i
= -2- ( -3+1J = 0,
(X.8.8)
1)d. = -x)-3<0, for -1<x<2.
C3
Fix a positive number al so that
x,,\-
Integrating (X.8.4), it follows that
(X.8.10) 0 < p(x,A) < 2K for -1+a < x < 2 - al, 0 < A < A0.
Notice that p(2 - al, A) is independent of 5o. Then, it can be shown that
A
0+ as a, -+ 0+ and A 0+.
p(2 - a,, A)
0 < p(x, A) < 1 for 2 - al < x < 2 + a2i 0 < A < Ao.
Then,
2+02
g2 - 1) + P( A)fl > Q.
P(2 + 2, A) p(2 -- o , A) 12-,,
Hence,
2+0a
g2 - 1) + < p(2 - Aal, A) for 2 - al <x < 2 + a2.
J -o,
This is a contradiction if al and A are small. Therefore, if Ao > 0 is sufficiently
small, there exists an x(A) such that
TO
FIGURE 23.
334 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
EXERCISES X
()
has a solution.
(2) Does the boundary-value problem
dtY
dt2
= F t, y, dY
dt
dy(a)
dt = a'
dY (b)
dt
_3
have a solution?
(3) Show that the boundary-value problem
d2x
= tx + x3, x(-2) = A, x(3) = B
dt2
has a solution for any real numbers A and B.
Hint. A counterexample for (2) is
d2 = 0, (n) = 0, (b) = 1.
For (3), apply Theorem X-1-3 [Na4j with w,(t) = a and w2(t) where -a
and 3 are sufficiently large positive constants.
X-2. Find the global phase portrait of each of the following two differential equa-
tions:
d2x
(i) + x2(1 - x)2 dt + (x - 1)2(x + 1)x = 0;
2
d2x dx
dx2
+
dt
- 1
l + x2
= 0.
Hint. Set V (x, y) = 2 + G(x). Then, d = - f (x, y, t)yz + e(t)y. There exists a
positive number ro > r such that V(x, y) > 4 for x2 + y2 > ro. Hence, if an orbit
(x(t), y(t)) satisfies conditions that x(t)2 + y(t)2 > ro for to < t < ti, we obtain
d V(x(t),y(t)) < Ie(t)Iiy(t)I <2(e(t)j V(x(t),y(t)) for to <t <t,.
This yields
V(x(t),y(t)) < V(x(to),y(to)) + ft., (e(r)!dr for to <t < t,.
and Theorem IX-2-2, show that if 0 < 6< 3, every orbit of system (E.2) tends to
one of the stationary points as t -' +oo,
(d) discuss the uniqueness of periodic orbits.
X-7. Show that the differential equation
d2x
dz
+ f (x)
d+ 9(x) = 0
has at most one nontrivial periodic solution if fix) and g(x) are continuously dif-
ferentiable in !R and satisfy the following conditions
<0 for 1x3 < 1,
(i)
f (x) { > 0 for IzI > 1,
(ii) 9(x)
<0 for -2<x<0,
> 0 for (x + 2)x > 0,
(iii) J g(x)dx = 0.
336 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
Then,
da(t)
_ -f(x(t))(x'(t))2 .
dt
Therefore, we obtain
Now, investigate the quantity (x'(t))2 as a function of A along the periodic orbit
(x(t), x'(t)). For a fixed value of A, compare (x'(t))2 for different values of x(t),
using the assumptions on f and g.
Step 1. First the following remarks are very important.
(a) If we set y = x', the given differential equation is reduced to the system
(S) dt = y, dt = -f (x)y -
A careful observation shows that the index of the critical point (0, 0) is 1,
while the index of the critical point (-2,0) is -1. There are only two critical
points of (S).
(c) The periodic orbit and any line {x = a constant) intersect each other at most
twice.
(d) The critical point (-2,0) should not be contained in the domain bounded by
the periodic orbit. To show this, use the fact that the index of any periodic
orbit is 1.
These facts imply that the periodic orbit should be confined in the half-plane x >
-2.
2 r:
Step 2. A level curve of V (x, y) = 2 + is a closed curve if the curve is
J0
totally confined in the strip jxj < 1. In particular, (1, 0) and (-1, 0) are on the same
level curve V (x, y) = / 1 _ 1 g(C)d{. Since A(t) is increasing as long as
0 Io
jx(t)I < 1, the periodic orbit cannot intersect the level curve V(x,y) = / ,
0
Therefore, the periodic orbit must intersect the lines x = 1 (as well as the line
x = -1) twice. Denote these four points by (1,s1(A)), (1,-rt(B)), (-1,-q(C)),
and (-1, 77(D)), where r)(A), n(B), q(C), and q(D) are positive numbers.
EXERCISES X 337
Step 3. Set
Step 4. Now, fixing a AO E Z(A) n Z(B) n 1(C) n T(D), evaluate the integral (I)
as follows:
rT {JA8 dA rc dA rAD dA AA dA 1
J
f(s(t))dt = -
w yi (A)2 + J y2 (A)2 +'Ac yl (A)2 + IA a y2 (A)2 T ,
where
dA dA r" dA
(A)2,
J\A yi (A)2 as yi (A)2 lao yi
rac dA _ ra° dA
AC
dA
+
Jib
y2(A)2,
(A)2.
ao ys (A)2 aD y2 (A) 2 Jao y2
338 X. THE SECOND-ORDER DIFFERENTIAL EQUATION
E
dt = -y + f (t)
which is periodic of period T. Also, find the uniform limit of 0(t, c) on the interval
-oo<t<+ooasE-+0.
X-9. Assuming that X(t) satisfies the condition
on an interval 0 < t < t, where a is a positive constant, and that x(t, A) is the
unique solution to the initial-value problem
d2 Q
t2Y
dt2
+ 2ydy
dt
= 0, y(0) = a, y'(0) =
where e is a positive parameter, or is a real number, and ,8 is a positive number,
find lim y(t, e) for any fixed t > 0.
f 0+
4
dt2 + 2ydty = 0, y(0) = A, y(1) = B,
in the following six cases: (1) 0 < B < A, (2) B = A, (3) B > IAA, (4) B = -A > 0,
(5) CBI < -A, and (6) B < 0 < A, assuming that e is a positive parameter. Also,
find lim y(t, e) for 0 < t < 1 in each of the six cases.
e 0+
EXERCISES X 339
Hint. Use explicit solutions together with the Nagumo Theorems (Theorems X-
1-3 and X-7-1) on boundary-value and singular perturbation problems. See also
Exercises X-10 and X-12, and [How].
X-12. Let f (x, y, t, e) be a real-valued funcion of four real variables (x, y, t, e).
Assume that
(i) 0(t) is a real-valued, continuous, twice-continuously differentiable function
on the interval Zo = {t : 0 < t < 1) and satisfies the conditions 0 =
f (¢(t), 4'(t), t, 0) and 4(1) = B on Zo, where B is a given real number,
(ii) the function f (x, y, t, e) and its partial derivatives with respect to (x, y) are
continuous in (x, y, t, e) on a region R = {(x, y, t, e) : Ix - ¢(t)I < rl, IyI <
+00, t E Zo, 0 < E < r2}, where r1 and r2 are positive nmbers,
(iii) If (4(t), 4'(t), t, e) I < Ke for t E Zo, where K is a positive number,
(iv) there exists a positive number µ such that Of (x, y, t, e) < -p on R,
(v) there is a positive-valued and continuous function '(s) defined on the interval
J+00
0 < s < +oo such that +oo and that If (x, y, t, e)I < +'(IyI) on
R,
(vi) A is a given real number.
Then, there exists a positive number co such that for each positive f not greater
than co, there exists a solution x(t, E) of the boundary-value problem
d2x
f d#2 = f (X, d , t, E) , x(0, e) = A, x(1, c) = B
such that
Ix(t,E) - di(t)I < IA - h(0)led=te-ft/` +C2e on Zo,
dx"
= f (Y, i, t, 0), 0 = g(x, y, t, 0) on Zo,
dt
(ii) f (Y, y, t, e), g(x', y, t, e), and their partial derivatives with respect to (x, y-) are
continuous on a region R = {(:e, y, t, e) : I x - m(t) I < rl, Iy - j(t) I < r2, t E
lo, 0 < e < r3 }, where rl, r2, and r3 are positive numbers,
(iii) all eigenvalues of the matrix (1(t), ti(t), t, 0) are less than a negative num-
ber M on .70.
Show that the initial-value problem
d:5
T = f (: , y, t, E), a Y
dt =
§(Y, 9, t, f), x(a) = ((e), y(a) = #(f)
has a unique solution (2(t, e), y(t, f)) for t E Zo and 0 < e < r4 if e + IC(e) -
(a)I + Ir(e) - t'(a)I is sufficiently small on 0 < e < r4i where r4 is a positive
number. Also, show that Ii(t, e) - ¢(t)I + I y(t, e) - tj,(t)I - 0 uniformly on Zo as
e + IC(E) - (a)I + I#(E) - (a)I - 0.
Hint. See [LeL].
X-15. Let x E lR", y E R"', t E R, and e E R. Also, let t, e) and g(i, y, t, e)
be respectively R"-valued and R'-valued functions of (i, y", t, e) which are periodic
in t of period T. Assume that
(i) an R"-valued function fi(t) and an R"'-valued function t%'(t) are continuous,
continuously differentiable, and periodic of period T on the interval Zo = it :
-00 < t < +oo} and satisfy the system of equations d = f (:F, y", t, 0), 0 =
g"(x, y, t, 0) on Za,
(ii) f(i, y", t, e), g(i, y", t, e), and their partial derivatives with respect to (x, y') are
continuous on a region R = {(i, y, t, e) : Ix - m(t)I <- ri, I y - 7G(t)I < r2, t E
Zo, IEI < r3}, where rl, r2, and r3 are positive numbers,
(iii) there exists a real m x m matrix P(t) such that
(iiia) the entries of P(t) and P(t)-1 are real-valued, continuous, continuously
differentiable, and periodic of period T[Bat
on Zo,
(iiib) P(t)8b(m(t),1(t),t,0)P(t) = B2(t)], where Bi(t) is a real
m1 x ml matrix with eigenvaules havingnegative real parts on 10i while B2(t)
is a real (m - ml) x (m - ml) matrix with eigenvalues having positive real
parts on Zo,
(iv) the system
EXERCISES X 341
d,F=
.iF
8iq(t), (t),t,0)
di dy"
dt - f(s,y,t,f), adt
=9'(2,y,t,e)
has a unique periodic solution (a(t, e), y(t, e)) of period T on To if 0 < Iej is suffi-
ciently small. Also, show that jT(t, e) - (t)I + lc(t, e) - iJJ(t)j 0 uniformly on To
ASYMPTOTIC EXPANSIONS
has an actual solution f (x) = el/=J t-1e-hIldt for x > 0. Integrating by parts, we
0
N
obtain f = (-1)m(m!)xm+1 + (- 1)N+1((N + 1)!)e1/x tNe-l'tdt. Since 0 <
M=0 fox
342
1. ASYMPTOTIC EXPANSIONS IN THE SENSE OF POINCAR$ 343
for all x in Do. If, moreover, KN(t) are independent of t, then the asymptotic
expansion is said to be uniform with respect to t.
Remark XI-1-4. If f (x) is holomorphic (i.e., analytic and single-valued) in a
neighborhood of x = a, by virtue of the Taylor's Remainder Theorem, f (x) admits
its Taylor's series expansion as its asymptotic expansion in any sector with vertex
at x = a.
Theorem XI-1-5. For a continuous function f (x), there is at most one asymptotic
expansion as x -+ a in a sector with vertex at x = a.
Proof
Assume that there are two asymptotic expansions of f (x) at x = a
00
(XI.1.4) f (x) ^_- c,,(x - a)14 as x -a in D
M=0
344 XI. ASYMPTOTIC EXPANSIONS
and
f(x)^, 7m(x-a)' as x -i a in D.
M=0
Then, for every non-negative integer N, there exist two constants KN and LN such
that
N
N+i ,
f (x) - > cm(x - a)m < KN Ix - al N = 0, 1, 2, .. .
m=0
N
.f (x) - ,1.(x - a)m <LNIx-aIN+1 N=0,1,2,...
m=0
Proof.
N
Fix a non-negative integer N and put f (x) _ > cm(x - a)'+Ei(x)(x-a)N+i
M=0
N
and g(x) _ > 'ym(x - a)m + E2(x)(x - a)N+1 Then, there exists two constants
m=0
KN and LN such that
(XI.1.10) IE1(x)I <_ KN, IE2(x)I <_ LN
1. ASYMPTOTIC EXPANSIONS IN THE SENSE OF POINCAR$ 345
N
(XI.1.11) If (x) ± 9(x)] - E I(Cm ± 1'.](x - a)m < (KN + LN)I x - aIN+1
M=0
for x in Do. Thus, (XI.1.8) holds. Also, it is easy to see that there exists a positive
constant k such that
ra
Since lim f (x) = co, x = a is a removable singularity. Thus, f (x) is holomorphic
in Ix - al < b. Therefore, the asymptotic series agrees with the Taylor's series (cf.
Remark XI-1-4 and Theorem XI-1-5). Thus, the asymptotic series converges in
Ix - aI < 8.
For a vector z E cm with entries (z1,z2,... zm) and p = (p1,p2,... Pm) with
non-negative integers p, (j = 1, 2,... , m), we define I6oI = P1 + p2 + + p,,,
z' = zr z2 ... zm , and Iz7 = max{Iz1I, IZ21, ... , Ixml}.
Theorem XI-1-8. Suppose that F(x, z) is a function with power series expansion
00
F(x, ) = F.(x)zY, which converges uniformly forx E D, I zl < bo, where Fp(x)
Ip1=o
00
is continuous in D and admits an asymptotic expansion Fp(x) c 1: Fpk(x - a)k
k=0
as x -, a E D. Define the formal power series in (x,
00
(XI.1.12) 4i(x, 2) = E 4ip(x)iV,
lpI=0
where
0
(XI.1.13) 4ip(x) _ E Fpk(x - a)k.
k=0
Proof.
Choose p so small that I f (x) - fol < 6o for Ix - al < p. Fix a positive integer N
and put
N 00
00
Set
and, hence,
N N
(XI.1.18) E Fp(x)(f(x) - fo)p = E Hm(x - a)m +O(Ix - aIN+1).
Ip1=0 m=o
Since this is true for all positive integers N, the theorem follows immediately. 0
Theorem XI-1-9. Suppose that f (x) is a continuous function with asymptotic ex-
pansion (XI.1.4) and co Let p(x) denote the formal power series
34 0.
(X7.1.1). Then, Pox) defines a formal power series in (x - a) and f(x) ox) as
x -+a in D.
Proof.
as x -a in D.
f aaZ f(t)dt f a=P(t)dt
Proof
N
Set pN(x) _ E c, (x - a)'. From (XI.1.4) and (XI.1.5), it follows that
m-0
m=1
Proof.
For a non-negative integer N, put f (x) = pN(x) + EN(x). Then, there exists a
positive constant KN such that
df (x)
dx
= 1 r f
2rri Jr (t - x)2
__ 1 r PN (t)
2rri Jr (t x)2 -
I
+ 2rri
Ir (t -
EN (C)
x)2
dt.
348 XI. ASYMPTOTIC EXPANSIONS
(
ei4'
2Ix-al sin 0.Then, r_(rll>I= e'4'2Ix-aIsin0,0<0<27r
l 11
2x
_ 1 1
EN(e)e"'dcb = 0 (Ix - aIN) .
7
lira Cnm = Cm
00
Proof
The assumption implies that for each non-negative integer N, there exists a
positive constant KN, independent of n, such that
N
(XI.1.23) a)'1 :5
(n = 1,2,3,... )
I
M=0
Put
N
(XI.1.26) pjN(x) =E cjm(x - a)n' (j = 1, 2, 3, ... ).
M=0
1. ASYMPTOTIC EXPANSIONS IN THE SENSE OF POINCARE 349
Then,
Thus,
Therefore, lim c)N = cN. Consequently, lim c),, = c,,, for all m.
) --00 l-00
Furthermore, since (XI.1.27) holds independently of j, we obtain
N
f(x) - > cm(x - a)m < KNlx - alN+1
M=0
them exists a function f (x) which is continuous in D and holomorphic in the inte-
rior of D, and
00
Proof.
Without loss of generality, assume that the sector D contains the ray arg(x -
a) = 0. Construct functions a,,, (x) for m = 1, 2.... such that f (x) = co +
00
E ca,,,(x)(x - a)m is continuous in D, holomorphic in the interior of D, and
m=1
satisfies (XI.1.32). Let bm and 0 be positive numbers, where bm are to be specified
later, whereas 0 is chosen to satisfy 0 < 0 < 1 and
Consider
r 1
(XI.1.34) am(x) = 1 - exp l - 2m1,m(x - a)0 J , m = 1, 2, ... .
Then, am(x) are holomorphic in a `sector containing D. Also, from (XI.1.33) and
the fact that
I1-esl=l jetdt <Izi for z<0,
it follows that
bm
(XI . 1 . 35) lam(x)I <
- 2mymlx - aIe' m= 1 , 2 ,...,
in D. Put
(XI.1.36)
bm = I Icml-' if c,,, 0 0,
0 if Cm=0.
Then,
00 00
Ix - alm-e
Icmam(x) (x - a)mI < E 2m.ym for x E D.
M=1 m=1
Hence,
Oo
To show that f (x) satisfies (XI.1.32), let N be a positive integer and observe
that
N+1 N+1
AX) -0D - E cm(x - a)m =EC exp ax - a)m
m=1 m=1
2--y-(Z - 7)01(
00
+ (x - a) N+1 i cmam(x)(x -a) m-N-1]
.
Lm=N+2
r = 0. 1, 2,... , N,
iexp [ 2'n7m(x - a)e ] Ix - aj-"I,
c-exp _
N r bm
,,=1 <Hljx-aIN+1 for xED.
m-1
I 2mrym(x - a)e] (x - a)m
Also, (XI.I.35) implies that there exists a positive constant H2 such that
00 -aim-N-1
a)m-N-1 1 [00 Ix
E cmam(x) (x - (27)N+1Ix - aj0 m=N+2 (2'r)m-N-1
m=N+2
for X E D.
- i (2y)N+2
+2
I
{x - aI - H2
2y
Theorem XI-1-14. For a given formal series p(x, t) _ 00 cm(t)(x - a)m in pour
m
ers of (x - a and a sector D given by (XI.1.31), where q. (t) (m = 0, 1, ...) are
holomorphic and are bounded in a domain S2, there exists a function f (z, t)
which is holomorphic with respect to (x, t) in D x S2 and satisfies the conditions
and
00
( X,
(XI.1.38) as x -. a in V
dt m=o dt
uniformly fort in Il.
Proof.
I dcd't(t)
As are bounded in Sl (m = 0,1, ... ), choose
{ I ax 1cm(t)I] +I if c -(t) 0,
6m = t r: I d d (t) J }-1
0 L ` ifC,,,(t)-0.
Set
00
2. The exponential integral function Ei(z) = ettdt has the following form:
J zo0
N Z
where a and b are two real numbers such that a < b and r is a positive number if
(i) ¢ is holomorphic on V (r, a, b),
+00
(ii) there exists a formal power series p = > a,,,x"' E G[[x]], such that an in-
m
equality
N-1
(XI.2.2) E amxm < Kv,a,A(N!)'(Bv,a,3)8[x[N
0(x) - M=0
holds on D(p, a, 0) for every positive integer N and every (p, a, 03) satisfying
the inequalities 0 < p < r and a < a <,3 < b, when KP,Q,o and Bp,,,,o are
non-negative numbers determined by (p, a, /3).
We denote by A, (r, a, b) the set of all functions admitting asymptotic expansions
of Gevrey order s as x -' 0 on the sector D(r, a, b). We also set J(4) = p for
0 E A. (r, a, b). In §XI-3, we shall explain the basic properties of ar[[x]] A. (r, a, b),
and the map J : A. (r, a, b) -+ G[[x]],.
In the example given at the beginning of this chapter, the formal solution p =
00
E (-1)m(m!)x"+1 of X2!Ly +y-x = 0 is of Gevrey order 1 and the solution f (x) =
M=0
eL/z Jxt-1e-lltdt admits an asymptotic expansion of Gevrey order 1. Furthermore,
JJa
J(f) = p. Also, the Maillet Theorem (cf. Theorem V-1-5) states that any formal
solutions of an algebraic differential equation belong to C[[x]], for some s, where s
depends on each solution.
It is well known that if a complex-valued function 0(x) is holomorphic and
bounded on a domain 0 < [xj < r, where r is a positive number, then 0 is repre-
sented by a convergent power series in x. The Gevrey asymptotic expansions arise
354 XI. ASYMPTOTICS EXPANSIONS
in a similar but more general situation. To explain such a situation, let us consider
N sectors
(X1.2.4)
+00
Then, there ex,sts a formal power series p = E a ..xm E C[[x)]. such that ¢t E
M=0
A. (r, at, bt) and J(41) = p for each 1.
There are various situations in which Gevrey asymptotic expansions arise. To
illustrate such a situation, let ¢(x) be a convergent power series in x with coefficients
in C. For two positive numbers r and k, set
k rr
0(t)e-('1x)}tk- 1 dt.
x J0
This integral is called an incomplete Leroy transform ofd of order k.
+oo
Theorem XI-2-4. For every ¢ = cx"j E C{x}, it holds that
m=0
R if
4,k(0) E Al/k P,-2k'2k
2. GEVREY ASYMPTOTICS 355
and
+00
J(tr,k(4)) _ r 1 1 + k Cm2'",
M=0
where k and p are any positive numbers and r is any positive number smaller than
the radius of convergence of 0.
Proof.
The following proof is suggested by B. L. J. Braaksma. For an arbitrary small
positive number e, let {Sl (e), S2(e), ... , SN(c)} be a good covering at x = 0, where
where the path of integration is the line segment connecting 0 to re{d'. Therefore,
of (x) - bt+1(x) = k
X j t
0(a)e-(ol=)'ak-1d,
where the path ryt of integration is the circular arc connecting re'd'+1 to re'd'.
Statement (3) follows, since
/
e-(°/=)` exp - t fx cos[k(arg a - arg x)J
Corollary XI-2-5. For any p E G[[x]], and any real number d, there exists a
function ¢(x) E A. d - 2 , d + ) such that J(O) = p.
(r,
2
This corollary corresponds to the Borel-Ritt Theorem (cf. Theorem XI-1-13) of
the Poincar6 asymptotics. Also, this corollary implies that the map J
A8 (r, d - s7r , d + 2) -+ c[[x]], is onto.
2
Theorem XI-2-3 is a corollary of the following lemma.
Lemma XI-2-6. Assume that a covering {St : e = 1, 2,... , N} at x = 0 is good
and that N functions 61(x), 62(x), ... , 6v (x) satisfy the following conditions:
(i) 6t is holomorphic on St n St+1,
(ii) I6t(x)I 7exp[-Alxl-k] on St n St+1, where -y > 0, A > 0, and k > 0 are
suitable numbers independent of e.
Define s by (XI. 2-4). Then, there exist N functions ), (x), t.b2(x), ... ,'IPN(x) and a
+oo
formal power series p = > E C[[x]], such that
m=o
(a) 4,t E A. (r, at, bt) and J(4't) = p, where St = {x : 0 < I x[ < r, at < arg(x) <
bt} (e = 1, 2, ... , N),
(b) 6t(x) _ t(x) - i/'e+l(x) onSt n St+1.
Let us prove Theorem XI-2-3 by using Lemma XI-2-6.
Proof.
Set
5t(x) = 4t(x) - 41+1(x) (e = 1, 2, ... , N).
Then, there exist N functions 4111(x), 02(z), ... , ON (x) satisfying conditions (a) and
(b) of Lemma XI-2-6. In particular, (b) implies that
Then, 0 is holomorphic and bounded for 0 < IxI < r. Therefore, 0 is represented
by a convergent power series. Since .01 = -01 + 0, Theorem XI-2-3 follows immedi-
ately. 0
We shall prove Lemma XI-2-6 in §XI-5.
Because the Gevrey asymptotics of functions containing parameters will be used
later, we state the following two definitions.
3. FLAT FUNCTIONS IN THE GEVREY ASYMPTOTICS 357
00
Proof.
Suppose that for a fixed W, there exist two non-negative numbers C and A such
that
I f ( u " , e)j < C(m!)"Am jelm for (11,f) E D x W and m=0,1,2,....
Then, by virtue of Stirling's formula
r`k) 2m
(XI-3.3)
m! = 2nm mme'm [i
l
+0(01,
Hence, I f (u", e) I < K exp[-A ej-k] for every (6,e) ED x W, where K = Coexp[s]
and A = . The converse is evident. 0
e 0
3. FLAT FUNCTIONS IN THE GEVREY ASYMPTOTICS 359
In particular,
Letting a -+ k, we derive f (x) = 0 if argx = 0 and x < p. This completes the proof
of Lemma 111-3-5.
Since use was made of the Phragmdn-Lindelof Theorem, we shall state and prove
the theorem precisely (cf. [Ne2, pp. 43-44J).
360 XI. ASYMPTOTICS EXPANSIONS
Proof.
a
Choose I > k so that Q - a < < 7r. Set g(x) = f (x) exp[-e(e-'sx)-r], where
P 1.
0= a Q
and e> 0. Then,
2
I9(x)I = If(x)Iexp[-ecos(t(argx - 0))Ixl-`]
< Kexp[-IxI-r(ecos(f(argx - 0)) - AIxIt'k)]
for x E W. Note that eI arg x - 01 < f (3- a)J < 2 for x E W. Therefore,
2
lim g(x) = 0 and I9(x)I 5 If(x)I for x E W. Since g is holomorphic in W, it
Izl-u
follows that Ig(x)I < M for x E W. Therefore,
If (x)I < 11f lexp(e(e-sex)-t)I for xEW.
Letting e - 0, we derive (XI.3.4).
We can also prove the following theorem without any complication.
Theorem XI-3-6. If f is holomorphic in e on D(r, a, b) for each fixed u" in V and
f is flat of Gevrey orders as e - 0 in D(r, a, b) uniformly on a domain D in the
u"-space, then
a, b)
j uniform'
f (u, a)do and of (a, e) are also flat of Cevney orders as e -' 0 in
ly on D.
Of
where the coefficients P,n(v") are holomorphic in D. Then, for any (a, 0) satisfying
the condition a < a < Q < b, there exists a positive number p(a, 3) such that
0 < p(a, (3) < r and that the function F(Oi (6, e), ... , thin (6,t), c) is holomorphic in
D x D(p(a, 0), a, Q) and admits the formal power series P(U, e) as its asymptotic
expansion of Gevrey order max(s, si , ... , sn) as e 0 in V(p(a, 03), a, 03) uniformly
on the domain D.
Remark XI-4-2. As a consequence of this theorem. we conclude that under the
same assumptions as in Theorem XI-4-1, the formal power series P(v, e) in a is of
Gevrey order max(s, s1, ... , sn) uniformly on D.
Proof of Theorem XI-4-1.
Fixing a pair (a, (3) satisfying the condition a < a < 0 < b, choose a suitable
good covering {S1, ... , SN } ate = 0 and N(n + 1) functions Ft(u1, ... , un, e),
¢1,e(vU, e), ... , QSn,t(U, e) (e = 1,... , N) such that
(1) for each e, the function Ft(ui,... ,un,e) is holomorphic in the domain
D(r1 i ... , r,,) x St and admits the formal power series p(u1,... , Unit) as
its asymptotic expansion of Gevrey order s as e -' 0 in St uniformly on
D(r1,... , rn),
(2) for each (j, e), the function 6,,e(17, e) is holomorphic in the domain V x St and
admits the formal power series pj (u", e) as its asymptotic expansion of Gevrey
order sJ as e -+ 0 in St uniformly on V.
(3) there exist two positive numbers K and A such that
a
,un,E) - Ft+1(ul,... Un,()! < Kexp - Iefk
362 XI. ASYMPTOTICS EXPANSIONS
1
where k, = ,
Si
(5) $t = D(p, a, E3), Ft. = F, and O,,& n) for some fo.
We can accomplish this by virtue of Corollary XI-2-5 and Theorem XI-3-2.
Observe that
Fe+1(&,e+1(v,e),...
Fl(41,t(V, E), ... , ¢n,[(ve E), E) - (v, E), ... , 41n,,+1(v, e), e) I
tl
+ JFt(¢1,1+1(Lt,E), +dn,[+1(v,e),E) - F1+1(dl.[+1(6,e), ,1+1(V,e), e)1
n
< LKexp [-I E Ik
+ Kexp [-(
IC`J'
-1 E
for (6, e) E V x (St n S[+1), where L, K, and A are suitable positive numbers.
Therefore, using Theorem XI-2-3, we complete the proof of Theorem XI-4-1. 0
As a corollary of Theorem X1-4-1, the following result is obtained without any
complication.
Theorem XI-4-3. C([x]], and A. (r, a, b) are commutative differential algebra
over C, t.e.,
(i) f + g E C((x]], (respectively A. (r, a, b)) if f and g E C((x]], (respectively
A, (r, a, b)),
(ii) f9 E C([xJ], (respectively A. (r, a, b)) if f and g E C((xi], (respectively
A. (r, a, b)),
(iii) cf E C[(x]], (respectively A. (r, a, b)) if c E C and f E C((x]J, (respectively
A. (r, a, b)),
(iv) 1 E C((xJ], (respectively A, (r, a. b)) if f E C((x)J, (respectively A. (r, a, b)),
Remark XI-4-4. Using Theorem XI-3-6, we can prove (iv) and (v) of Theorem
XI-4-3 in a way similar to the proof of Theorem X1-4-1. Also, it can be shown
1
that if f (x) E A, (r, a, b) with J(f) = >+"Qa,,,x' and ao 54 0, we obtain E
m=o
f
5. PROOF OF LEMMA XI-2-6 363
A. (p(a, 8), a, 0) and J = J{f) , where a < a < /3 < b and p(a, /3) is a
\I/
suitable positive number depending on (a,fl).
The materials in this section are also found in [Ram1], [Ram2], and [Si17, Ap-
pendices].
for x E St, t = 1, 2,... , N. The functions +Gt can be continued analytically onto
St by deforming Ct_1 and Ct without moving any of their endpoints. In doing this,
we do not change other line segments Ch (h i4 t - 1, t). Thus, the function tyt is
holomorphic on St, t = 1, 2,... , N, respectively.
Now, assuming that x E St n St+1, compute t/t(x) - tPt+1(x). To do this, write
'Pt and tLt+1 in the following forms respectively:
tt't(x) _ {2 =) / ft (2 =) Ln2
E C"
h#lJ
where the paths Ct and Ct+1 of integration are obtained by deforming C, without
moving either of its endpoints so that
(1) Ct C St n St+l and Ct+l c St n St+1,
(2) -Ct + Ct+1 is a simple closed curve whose interior contains x.
Thus, (b) follows, i.e.,
Let us derive asymptotic properties of ib1. To do this, fix an t and a closed sector
W contained in St. Let Ct (respectively Ct_1) be a path obtained by deforming Ct
(respectively CL_1) without moving the endpoints so that W is contained in the
364 XI. ASYMPTOTIC EXPANSIONS
interior of the simple closed curve Ct_ 1 + 7t - C1, where 1't is a circular arc
joining two points reist-' and rest. For x E W, it holds that
ot (x) =
{-1)
- 27ri f sew
x
(-1)
+ 2?7 f bt-1(1)
x
d{ (-1)
+ 27ri f x
It may be assumed that the path Ct is given as a union of a line segment L1 and a
curve rt defined by
pt(0) = r1e'", pt(i) = re'B', r1 < litt(r)I < r (0 < 7- < 1).
It can be also assumed that there exists a positive number a < 1 such that JxI < ar1
for x E W. Then, 1 fr, c - x
8t(-)
dC is represented by a convergent power series in x
21ri
whose radius of convergence is not less than r1.
mM+1
1
x,
x
1 1
x Q)m +
we obtain
1 M
1 6t{f) xM+lEM+1(x),
[Ymx"'
27ri L-x m=0
where
Hence,
2kirsin(0) r( k
7/N+llity+l)/k
We can estimate
-11 and
-1 (h94 Q,t-1)
2ri c,_, f - x 2iri Jc - x
in a similar manner. Thus, the proof of Lemma XI-2-6 is completed.
The material of this section is also found in [Si18].
EXERCISES XI
W = {x : 0 < JxJ < r < ro, a < a < argx < Q < b}
of S,
(b) J[f] the asymptotic expansion for f E A(S),
(c) Ao(S) the set of all those functions f (x) in A(S) such that f (x) ^-- 0 as x 0
in every W (i.e. AO(S) = If E A(S) : J[f] = 0}).
Show that C[Jxll. A(S), and Ao(S) are differential algebras over the field C and
that the map J : A(S) C[[x] is a homomorphism of differential algebras over
the field C.
XI-2. Using the same notations as in Exercise XI-1 and considering a linear dif-
N
ferential operator P = E ak(x)Dk with coefficients ak(x) which are holomorphic
k-o
in a disk JxJ < r containing S, where D = d , show that
366 XI. ASYMPTOTIC EXPANSIONS
Comment. The formal power series f E C([x]] is of Gevrey order s if and only if
f E E(s, A) for some positive number A.
XI-5. Show that f (e) = 0 identically in a sector S if f is holomorphic and flat of
Gevrey order 0 as e --+ 0 in S.
Hint. I f (e)I < KA' IeIr' fore E Sand N = 0,1,2,..., where K and A are positive
numbers independent of N.
XI-6. Show that C[[x]]o = C{x}.
Hint. f = 00E E C[[x]]o if and only if Ic,.. I < KAt If I' (m = 0, 1, 2.... ),
=G
where K andmA are positive numbers independent of m.
X1-7. Consider a system of differential equations
(S)
xy+t dy = f0 + bg +
Ip!>_o
Hint. Write system (S) in a form y" = go + x9+1 1P + E y'gp and use the
ipl>2
dx
Banach fixed-point theorem in terms of the norm II IIs.A-
XI-8. Let D(r) = {z : IzI < r}, S(r, p) = {z : 0 < IzI < r, I arg zI < p}, and a power
+w
series f (z, e) = E fn(e)zn is convergent uniformly in a domain D(ro) x S(r, p),
n=o
where ro, r, and p are positive numbers. Assume that the coefficients fn(e) of the
series f are holomorphic in S(r, p) and admit asymptotic expansions in powers of
e as e -' 0 in S(r, p). Suppose also that limo fo(e) = 0 and limo f, (c) 0 0. Show
C- 1-
that there exist a positive number rt and a function O(e) such that (1) O(e) is
holomorphic in S(rt, p) and admits an asymptotic expansion in powers of a as
e -+ 0 in S(rt, p), (2) l ¢(e) = 0, and (3) f (c(e), e) = 0 identically in S(rt, p).
368 XI. ASYMPTOTIC EXPANSIONS
(P) F(x, y, _ #,
y fa,,n2,
ip,l+IpI>o
yi z1
"
where %' _ , fn:,as E C[[x]] , pi = (pll,... ,pln), and Pz =
yn xm
(p21, ... , p2",) with non-negative integers P1k and p2k. Denote by Fg(x, y z) the
Jacobian matrix of ! with respect to W. Assume that power series (P) satisfies the
conditions (1) f;(0,6,6) = 0, (2) Fy(0, 0, 0') is invertible, (3) fn, a, E E(s, A)" for
some s > 0 and A > 0, and (4) the power series i[ fn, a, Ij, Aye' is a
ia,I+?P2I?0
convergent power series in y and F. Show that there exists a unique power series
¢(x, z) 92 in (x, z) such that (i) drr,, E C[[x]]", (ii) b(0,0) = 0, and (iii)
lack>o
F(x, (x, z), z) = 0. Also, show that (a, E E(s, B)" for some B > 0 and the power
series E II( a' i" is a convergent power series in z.
IP3i>o
&'$121 Is,A
M=0
Hint. Assume that j argx1 < 2k - b, where b is a small positive number. Then,
k
k rT tme-(t/x)ktk-Idt = xm (r'/) v"'/ke °do
J0 0
m) Jr°O
r (1 + xm - xm Om/ke-`do.
T J (t/z)k
Hence,
N
1,,k(6) _ r (i + m ) Cmxm - E cmxm Om/ke °d0
k m-0 (t/k)k
M=0
/r +00
+ k J cmtm e-(t/x)ktk-ldt.
xk 0 m=N+1
be a good covering of the sector S = {x : j argxj < 'r,0 < jxj < p}, where a1 =
,b= , and p is a positive number. Assume that v functions &1(x), 4(x),
for some positive numbers po and E. Show that if a formal power series f E C[[x]]
is k-summable in a direction arg x = 0, there exists one and only one function
F E Al/k (po, 9 - 2k - e, 0 + 2 + EJ such that J[F] = f.
Hint. Use Remark XI-3-3. For more informations concerning summability, see, for
example, [Ba13], (Ram2), [Ram3], [Si17, Appendices], and [Si19].
r e_t-s
XI-16. Consider the integral f (x) = f - x dl;, where -1 is a line segment 0 <
where a is a positive integer and f J (x, V1, v2, ... , vn, f) are holomorphic with respect
to complex variables (x, v1, v2, ... , vn, E) in a domain
(XII.1.2) Ix[ < do, 0< IEI < Po, I arg EI < ao, Iv. I < 'ro (j = 1, 2, ... , n),
6o, po, ao, and 'Yo being positive constants. Set
f) () n
(XII.1.3) f j (x, V, = fJO(x, a,h(x, e)Vh + 1 fi, (x, E)iJ ,
h=1 Icl?2
372
1. AN EXISTENCE THEOREM 373
Observation XII-1-1. Under Assumption I, fjo(x, e), a jh(x, e), and f jp(x, e) ad-
mit asymptotic expansions
00 00
and
Let A(x, e) be the n x n matrix whose (j, k)-entry is a.,&, e), respectively (i.e.,
A(x, e) = (ajk(x, e)). Then, A(x, c) admits an asymptotic expansion
where Al, µ2f ... , An are eigenvalues of Ao(0) and .,V is a lower-triangular nilpotent
matrix.
Note that [Ni can be made as small as we wish (cf. Lemma VII-3-3).
The following third assumption plays a key roll.
Assumption III. The matrix Ao(0) is invertible, i.e.,
Theorem XII-1-2. Under Assumptions I, II, and III, system (X11.1.1) has a
solution
where b, p, and a are suitable positive constants such that 0 < 6 < bo, 0 <
p<po, and 0<a<ao,
(ii) p, (x, e) admit asymptotic expansions
00
as e - 0 in the sector
(XI1.2.2) lxi < 6', 0 < jej < p', l arg el < a',
as f -. 0 in the sector
(XII.2.4) 0 < IEI < p', I argel < o .
v, = u, + q,(x,E) (j = 1,2,...,n).
Denote (ql, q2, ... , qn) and (U I, U2,. .. , un) by q" and u, respectively. Then, ii sat-
isfies the system of differential equations
dd
(XII.2.5) E° xj = gj(x,19, c) (j = 1,2,... ,n),
where
E° dq,(x,E)
93 (x,u,E) = fj(x,U + q' f) - (j = 1, 2,. .. , n).
dx
Set
n 00
In particular,
and
(XII.2.12) lxI <8, 0<lEH<p, largfl <a', lull <-y (j=1,2,... n).
Here, BN is a positive constant depending on N and 1191.
From (XII.2.5) and (XII.2.9), it follows that
I
u, = exp of( (t - x) R, (t, u, E)dt
(XII.2.14) E
where the paths of integration must be chosen carefully so that uniformly convergent
successive approximations can be defined.
Hereafter in this section, we explain how to choose paths of integration on the
right-hand side of (XII.2.14).
Observation XII-2-3. Set
(XII.2.15) w,=argµ, (j=1,2,...,n)
and suppose that
21
-2rr <w, - e < 7r and w, - e -2ir ,n).
Without loss of generality, suppose that n real numbers w, are divided into the
following two groups:
and
x(1) = be-i°, x(2) = ix(1) tan [3, x(3) = -x(1), and x(4) _ -x(2),
where b is a sufficiently small positive number. Then, a rhombus is defined by its
four vertices x( j) (j = 1,2,3,4) (cf. Figure 1). Note that the angle at x(1) is 20.
FIGURE 1.
Denote the interior of this rhombus by D(5). It is noteworthy that the domain
D(b) contains a small open neighborhood of x = 0 and is contained in the domain
{x : IxI < b}.
The basic estimates for the proof of Theorem XII-1-2 are given by the following
lemma.
Lemma XII-2-4. For each j = 1.... , n, consider the function
(x - xl)
(XII.2.18) I Ul (x, E)1 :5 c1cl, lexp '1l
L E° J
for
(XII.2.19) x E D(8), I argel <a, IEI > 0.
378 XII. ASYMPTOTIC SOLUTIONS IN A PARAMETER
Proof
We prove this lemma for the cases j = 1, 2,... , m'. The cases j = m' + 1,... , m
can be treated in a similar manner. Set f = It - x(l)I and 0 = arg(t - x(')) for
t E D(b) and set w = arg e. Then,
(XII.2.20)
11
From the definition of D(8), it follows that 7r - e - f3 < 0 < 7r - e +,6 for x E D(8).
Thus, if a is in the sector I arg eI < a, the inequalities
1 3
2a<wj +7r-(aa+(3) <wj +0-aw<wj + ir - 0 + (aa + 0) < 2a
hold f o r x E D(b) and j = 1, 2, ... , m'. Therefore, there exists a positive constant
c' such that
(XII.2.21) -cos(wj +0-aw)>c (j=1,2,...,m')
for (x, e) in (XII.2.19). By virtue of (XII.2.20) and (XII.2.21), we obtain
IUj(x,E)I <
µ,( xj) (j = 1,2,... ,m')
leXp [
u°) (x, e) 0,
I i (Z 1
(XII.3.3) Uh(xe) _ exp [__i.(t - x)} Ri (t, u"("-1)(t, e), e)dt
(j = 1,2,... ,n; h=1,2,...)1
3. PROOF OF THEOREM XII-1-2 379
where x E D(b) and the integration is taken over the straight line . For a given
positive integer N, it will be shown that
(i) for each j, the sequence I (h) (x, E) : h = 0,1, ... } is well defined,
(ii) for the given integer N, there exist positive constants KN and PN (0 < PN
p) such that
(iii) the sequence {u(') (x, f) : h = 0,1, ...) converges uniformly to (x, E) _
(01(x, E), ¢2(x, E), ... , 0n (x, E)) in (XII.3.5), where iZ (x, f) is the C"-valued
function with the entries (u(h) (x, E), ... , u,(th) (x, E)).
The limit function (x, e) is independent of N since the successive approximations
are independent of N.
If (i), (ii), and (iii) are proved, it follows that
as e - 0 in the sector S = E : 0 < lei < sup(pN), I argel < a' }, and the functions
inll N
(x, c) are holomorphic the domain D(b) x S and satisfy (XII.2.14). Setting
pf(x,E) = j(x,E) +g2(x,E)(j = 1,... ,n), we obtain a solution v. = p., (x,E)(j =
1,2,... , n) of (XII.1.1) which satisfies all of the requirements of Theorem XII-1-2.
Thus, the proof of Theorem XII-1-2 will be completed.
To show (i) and (ii), choose two constants KN and pN so that KN >
PNNKN < ry. This is possible since condition (XII.3.2) is satisfied. Now, assuming i BNOC and
that (i) and (ii) are true for ugh-1)(x,E) in (XII.3.5), let us prove that
also satisfy conditions (i) and (ii). First from (XII.3.3), it follows that
Then, from Lemma XII-2-4, (XII.2.10), (XII.3.4), and the inductive assumption,
we conclude that
for (x, e) in (XII.3.5). Thus, (i) and (ii) are true for ujh) (x, E) (j = 1, 2, ... , n).
380 XII. ASYMPTOTIC SOLUTIONS IN A PARAMETER
Suppose that Ao(O) has a distinct eigenvalues A1, A2i ... , At with multiplicities
n1, n2i ... , nt, respectively (n1 + n2 + - - + ne = n). Without loss of general-
ity, assume that Ao(0) is in a block-diagonal form
(XII.4.6) Ao(0) = diag [Al, A21 ... , At] ,
where 6, p, and a are positive numbers such that 0 < 5 < 60, 0 < p < po and
0<or <ao,
(ii) P(x, e) admits a uniform asymptotic expansion
00
(XII.4.14) B(x, e) = diag[Bj (x, e), B2(x, e), ... , B,(x, e)],
where B , (x, e) is an n, x n, matrix (j = 1, 2, ... , e),
(iv) the matrix B, (x, e) admits a uniform asymptotic expansion
(XII.4.15) Bj (x, e) ^- 00
1:
Remark XII-4-2.
(a) Set
(XII.4.16) diag [B1v(x), B2 ,(x), ... , Bl,(x)] -
Then, the coefficient matrix B(x, e) of system (XII.4.13) admits a uniform
0
asymptotic expansion B(x, e)
'=a
(b) When a fundamental matrix solution Z(x, e) of (XII.4.13) is known, a funda-
mental matrix solution Y(x, e) of (XII.4.1) is given by Y(x, e) = P(x, e)Z(x, e).
(c) In the case when the matrix Ao(O) has n distinct eigenvalues, by Theorem
XII-4-1, we can diagonalize system (XII.4.1).
(d) In the case when eigenvalues of the matrix Ao(O) are not completely distinct,
the point x = 0 is, in general, a so-called transition point. In order to study
behavior of solutions in the neighborhood of a transition point, we need a
much deeper analysis of solutions of system (XII.4.1). For these informations,
see, for example, [Wash], [Was2], and [Si12].
Proof of Theorem XII-4-1.
We prove this theorem in two steps. The proof is similar to that of Theorem
VII-3-1.
Step 1. We show that there exist a positive number 6 (< oo) and an n x n matrix
Po(x) such that
(i) the entries of Po(x) and Po(x)-1 are holomorphic in the domain {x : JxI < b}
and Po(0) = In,
(ii) the matrix Co(x) = Po(x)-1Ao(x)Po(x) is in a block-diagonal form
In fact, two matrices Po(x) and Co(x) must be determined by the equation
(XII.4.18) Ao(x)Po(x) - Po(x)Co(x) = 0.
Set
4(11)(x)
'4(12)(x) 4(13)(x) ... 40(2)
Ao(x)
Ao2 1) (x 4 ) (x) AA(x)
o2) ... Aou) (x)
Aat1)(x) Ao r)(x)
Aotz}(x) Ao)(x)
nd
Poll)(x) Po12)(x) Po's)(x)
Pol() (x)
Pa21)(x)
P Po2)(x) Po2'(x)
Po(x) = o )(x)
,
Poti)(x) Poti)
P ' (x) Po )(x) (x) 1
4. A BLOCK-DIAGONALIZATION THEOREM 383
k)
where A4(jk) (x) and PO (x) are nj x nk matrices. Furthermore, set PO l) (x) _
In, (j = 1,2,... , t). Then,
and
Ao(jj)(x)po
k)(x) - po k)(x)Cc(k)(x)
(XII.4.20)
+ +AD(jk)(x) =
0
(j i4 k)
h#j,k
dP(x, e)
(XII.4.23) Ea = A(x, E)P(x, E) - P(x, E)B(x, E),
where
Set
A(11)(x,E) A(12)(x,E) ... A(11)(x,E)
A(21)(,, ) A(22) (x, E) ... A(21) ('T' E)
A(x, c) _
A([I) (x, E) A(M) (x, E) ... A(t[) (x, E)
P(11)(x,E) P(12)(x,E) ... P(I[)(x'E)
P(21) (x, E) P(22) (x, E) ... P(2[) (x, E)
and
B(x, e) = diag [b(') (X, E), b(2) (X, E), . . . , B(') (x, E)] ,
where A(jk) (x) and P(Jk) (x) are nj x nk matrices and BJ (x, c) is an nj x nJ matrix
(j, k = 1,... , e). Furthermore, set
and
(XII.4.27)
EodP(jk)(x,E)
= A(J)(x)P(Jk)(x,E) - P(Jk)(x,E)A(k)(x) + A(jk)(x,E)
Eo dP(J c) (x, E)
4 J)
x P(Jk) x E - P(Jk) x, e A(k) x + A(Jk) x, E
I
+E L.:A(Jh)(x, E)P(hk)(x, E) - E p(Jk)(x, E)A(kk) (x, E)
h=1
I
- (2P(Jk) (x, c) A(kh) (x, E)P(hk) (x, E) (j0k).
h=1
E
o dX (2k) = A (J) ( x) X (jk) - X (jk) A(k)x+(X)A(Jk) x( E),
dx
I
(XII.4.28) + E E A(Jh)(x, C)X(hk) - EX (Jk)A(kk)(x, f)
h=1
I
(XII.5.3) A(x, e)
89
Oyn (x, 0, f)
8y1(x,E)
and
respectively.
We first prove the following lemma.
Lemma XII-5-1. Assume that
(i) Ax, y", e) is holomorphic with respect to (x, f, c) in a domain A(bo) x Q(po) x
S(ro, no), where 6o, po, ro, and no are positive numbers,
(ii) f (x, y", c) is bounded on A(60) x i2(po) x S(ro, no),
(iii) the matrix Ao(x) defined by (XII. 5.4) exists as e - 0 in S(ro, no) uniformly
in x E A(6o) and Ao(0) is invertible,
(iv) f (x, 0, e) is flat of Gevrey order -r as e -+ 0 in S(ro, no) uniformly in x E
A(60), where r is a non-negative number.
Then, there exist three positive numbers b, r, and or such that system (XIL5.1) has
a solution O(x, e) which is holomorphic in (x, t) E A(6) x S(r, a) and that (x, c)
is flat of Gevrey order -r as e -' 0 in S(r,a) uniformly in x E L1(6).
386 XII. ASYMPTOTIC SOLUTIONS IN A PARAMETER
Proof.
Note first that
(XII.5.5) I exp1- k1I = expJ-cJfJ-'cos(k(arge))]
for any positive numbers c and k. Note next that assumptions (i) and (iv) imply tha
in the case when r = 0, Ax, 0, e) is identically equal to 0 for (t, f) E A(b) x S(r, a).
Hence, in this case, 0 is a solution of (XII.5.1). In the case when r > 0, it holds
that
(XII.5.6) If(x,o,e)I < h exp[-2cIEl-k1
for some positive numbers K and c if (x, e) E A(b) x S(r, a) for sufficiently small
positive numbers 6, r, and a and if
k
(XII.5.7)
exp[cE-k1f(x, eXp[-cE-k]u, e)
= eXPfce-kJf (x, 0, e) + A(x, e)U + explc(1 - lpl)E-k]ul fp(x, e)'
ip1>2
Then,
(1) (XIL5.1) has a unique formal solution
Proof.
If a positive number & is sufficiently small, for every real number 6, there exists
a C"-valued function fe(x,y,e) such that
(a) fe(x, y', e) is holomorphic in (x, y", e) on a domain A(60) x f2(po) x Se(ro, &),
where
(XIL5.14) Se(ro,&) = {e: jarge - 61 < &, 0 < fej < ro},
(b) fe(x, y, e) admits an asymptotic expansion F(x, y e) of Gevrey order s as
e - 0 in Se(ro, &) uniformly in (x, y) E 0(bo) x f2(po), where s is the non-
negative number given in Theorem XII.5.2.
Such a function fe(x, y, e) exists if & is sufficiently small (cf. Corollary XI-2-5). In
particular, set
(XII.5.15) fo(x, b, e) = f (x, v, e)
(XII.5.16) e° = fe(x, y, e)
such that &(x, e) is holomorphic and bounded in (x, e) E i (b1) x Seri, al) for
suitable positive numbers 61, r1, and al. Using Theorem XII-1-2, it can be shown
that such a solution of (XII.5.16) exists.
388 XII. ASYMPTOTIC SOLUTIONS IN A PARAMETER
B(x,E)W(x,E) = fet(x,iez(x,E),E),
(XII.5.19)
6(x,0 =
Note that
and u and v are suitable positive numbers (cf. Theorem XI-3-2). From (XII.5.20),
it follows that
1=0 if s=0,
..
(XII526) 1.-.f .)l
5. GEVREY ASYMPTOTIC SOLUTIONS IN A PARAMETER 389
for (x, e) E A(S2) x Se (r2, a2), K and c are suitable positive numbers, t(x, e) is
a fundamental matrix solution of the homogeneous system
on the domain V(54) x So(r4,Q4), where 64i r4, and a4 are suitable positive
numbers.
El
Write fin the form z" _ I where z"J E C"' (j = 1, 2, ... , f). Then, (XII.5.29)
ztJ
can be written in the form
In this way, we can prove that t 5(x, e) - w'(x, e) is flat of Gevrey order 1 as e - O in
S e, (r1, a i) n Sez (ri , a 1) uniformly for x E A(S) if a positive number 6 is sufficiently
small (cf. [Si6; Proof of Lemma 1 on pp. 377-379]). Therefore, by using Theorem
X1-2-3, we can complete the proof of Theorem XII-5-2.
Materials of this section are also found in JSi22J.
(T) y= PMg
reduces system (E) to a system of the form
(Eo)
= xkB(z}v,
where B(t) is an n x n matrix whose entries are polynomials in t. Moreover, the
matrix P(t) can be chosen so that x = 0 is, at worst, a regular singular point
of (E).
Observe that if we set t = x-3, (E) becomes
dg _t-k-2
{E1) = A(t)9.
dt
6. ANALYTIC SIMPLIFICATION IN A PARAMETER 391
If k < -2, (E1) has the coefficient holomorphic in (D). Hence, there is a funda-
mental matrix solution V(t) of (E1) whose entries are holomorphic in (D). Then,
the transformation v = V (1) i reduces (E) to the system f = 6. Therefore, the
main claim of Theorem XII-6-1 concerns the case when k > -1. In this theorem
of G. D. Birkhoff, the entries of the matrix B(t) are polynomials in t. However,
even though we can choose P(t) so that x = 0 is, at worst, a regular singular point
of (Eo), the degree of B(t) with respect to t may be very large. In order that the
degree of B(t) with respect to t is at most k + 1 so that x = 0 is a singularity
of the first kind of system (Eo), we must impose a certain condition on A(t) (cf.
Exercises XII-8, XII-9, and XII-10). For interesting discussions on this matter, see,
for example, [u2], [JLP], [Ball], and [Ba12j. A complete proof of Theorem XII-6-1
is found, for example, in [Si17, Chapter 3].
In this section, we prove a result similar to Theorem XII-6-1 for a system of
linear differential equations
under the assumption that a is a positive integer, y" E C", and A(x, e) is an n x n
matrix whose entries are holomorphic with respect to complex variables (x, e) in a
domain
(XII.6.2) X E Do, [e[ < 60,
by the transformation
dil
ed En CkAk(x)+em+IB0(x)I!
.
k=0
dil
e = {Ao(x) + eB0(x)} u.
where
QdP
(XII.6.9) e = A(.x, e)P - PB(x, e).
and
o+k
dPk(x)
=A m +1+ o +k(x) + Ad+k-h(x)Ph(x)
dx
h=o
(XII.6.11) o+k
E Ph(x)Bo+k-h(x) (k = 0,1,2,...),
h=k-m
where
It should be noted that the formal power series P and B that satisfy the equation
(XII.6.9) are not convergent in general. In order to construct P as a convergent
power series in e, we must choose a suitable B. To do this, first solve equation
(XII.6.10) for B,,,+1+k(x) to derive
dPk(x)
(XII.6.15) = fk(x; P) (k = 0,1, 2, ... ),
dx
where
o+k o+k
fk(x; P) = Am+1+o+k(x) + E Ao+k-h(x)Ph - 1: PhA,,+k-h(x)
h=0 h=k-m
(XII.6.16)
a+k
- E PhHo+k-h(x;P) (k=0,1,2,...).
h=k-m
dP = F(x; P).
(XII.6.17)
dx
394 XII. ASYMPTOTIC SOLUTIONS IN A PARAMETER
Solve this differential equation in a suitable Banach space with the initial condition
P(0) = 0. Actually, we solve the integral equation
- k=0
> ek E Ph[Ak-h(x) + Hk-h(x;P)1
h=O
l
}.
Step 3. We construct the matrix P(x, e), solving the integral equation (XII.6.18)
by the method of successive approximations similar to that given in Chapter I. By
virtue of Lemma XII-6-4, we can construct a solution P(x) in a subdomain Dl of
Do containing x = 0 in its interior. Since (XII.6.18) is equivalent to differential
equation (XII.6.15) with the initial condition P(0) = 0, the solution P(x) gives
the desired P(x, e). The matrix B(x, c) is given by (XII.6.13) and (XII.6.14). 0
EXERCISES XII
00
where y E C", A is an invertible constant n x n matrix, and 5,n (x) and bm(x) are
C"-valued functions whose entries are holomorphic in x in a neighborhood of x = 0.
XII-2. Using Theorem MI-4-1, diagonalize the system
dy = 0 1+x1 where
e dx 11-X ex y' [1.
XII-3. Using Theorem XII-4.1, find two linearly independent formal solutions of
each of the following two differential equations which do not involve any fractional
powers of e.
E2d2
Y
(1) 2 + y = eq(x)y, (2) a2 + y = eq(x)y,
where q(x) is holomorphic in x for small jxj.
Hint. If we set '62 = e, differential equation (2) has two linearly independent
solutions e4=/00(x, 0) and a-1/190(x, -/3). The two solutions
= Q reiz/°4(x, Q) - -Q),
A(x, if, z, c), the functions f (x, y, z, e), and §(x, .F, e) admit asymptotic expansions as
f - 0 in the sector So = {E : 0 < IEI < ao, I arg EI < /30} uniformly in (x, y, z) in the
domain Ao = {(x, y, z) : IxI < ro, Iv'I < pl, I1 < p2} or (x, z-) in Vo = {(x, z) : I xI <
ro, Izl < p2}. The coefficients of those expansions are holomorphic with respect to
(x, y", I) in Do or (x, z-) in V0. Assume also that det A(0, 0, 0, 0) 0 0. Show that the
system
f dy = A(x, y, z, E)y + E9(x, Z, f), = Ef (x, g, Z, f)
has one and only one solution (y, z) = (¢(x, f), e)) satisfying the following
conditions:
(a) the entries of functions O(x, E) and '(x, f) are holomorphic with respect to
(x, e) in a domain {(x, f) I xI < r, 0 < lei < a, I arg EI < /3} for some positive
:
numbers r, a, and 0 such that r < ro, a < ao and /3 < /io,
(b) the entries of functions (x, c) and i/ (x, c) admit asymptotic expansions in
powers of f as c - 0 in the sector if : 0 < lei < a, I argel < (3} uniformly
in x in the disk (x : Ixi < r), where coefficients of these expansions are
holomorphic with respect to x in the disk {x: Ixi < r},
(c) 0(0,c) = O for f E {E : 0 < IEI < a, I arg EI < ,0}.
Hint. Use a method similar to that of §§X11-2 and XII-3.
r 1
at
XII-5. Find the following limits: (1) lim /
c-0 0 f (t) sin 1 dt and (2) lim
\E
j f (t) sine I I dt, where a is a nonzero real number and f (t) is continuous
and continuously differentiable on the interval 0 < t < 1.
XII-6. Discuss the behavior of real-valued solutions of the system
XII-7. Discuss the behavior of real-valued solutions of the following two differential
d3 y
equations as f O+: (1) E2 + 2 + xty = O, (2) E3 1 (1 x)y = O.
XII-8. Assume that
(i) the entries of a C"-valued function Ax, y", e) are holomorphic with respect to
(x, y", e) in a domain A1(Eo) X II(po) X O2(ro), where b0, po, and ro are positive
numbers and
0l(6o)={xE(C:IxI <bo}, I(po)={iEC":Iv7<po},
A2(ro) = {E E C : IEI < ro},
Denote by S(r, a, 0) the sector (EEC : 0 < IEI < r, I arg e - 01 < a). Show that
y
(1) the system (S) c'±dx = f (x, y, E) has a unique formal solution p(x, E) _
+00
with coefficients p"l(x), which are holomorphic in A(50),
c=1
(2) for any real number 0, there exist three positive numbers b, r, and a such that
(S) has an actual solution ¢(x, e), which is holomorphic in (x, E) E 0(b) x
S(r, a, 0), and that ¢(x, c) has the formal solution p(x, e) as its asymptotic
expansion of Gevrey order 1 as c -. 0 in S(r, a, 0) uniformly in x E 0(b)
or
Hint. This is a special case of Theorem XII-5-2.
+M
XII-9. Assume that p(x,E) _ E Emp',,,(x) is a formal solution of a system (S)
M=1
dy
ca = f (x, 17, f), where a is a positive integer, y" E C, Ax, y", e) is a C"-valued
dx
function whose entries are holomorphic in a neighborhood of (x, y, f) = (0, 0, 0),
and the entries of pm(x) (m = 0.1, ...) are holomorphic in a disk Ixi < ro, where
ro is a positive ember. Assume also that p(x, E) E {G[[E]]3}" uniformly for Ixi < ro
and that 0 < s < 1. Show that if S is an open sector in the E-plane with vertex at
e = 0 and whose opening is smaller than sir, there exists two positive numbers ri
and r2 and a solution $(x, e) of (S) such that the entries of q,(x, e) are holomorphic
in (x, E) for Ixt < r1, IEI < r2,( E S, and that (x, E) admits the formal solution
p(x, c) as the asymptotic expansion of Gevrey order s as e -+ 0 in S n {I-El < r2}
uniformly for IxI < r1.
Note. No additional conditions on the linear part of Ax, y", c) are assumed.
+00
XJI-10. Assume that p(x, c) = F E"`p""m(x) is a formal solution of a system
m--O
e = f (z, y, E), where y E C", f (x, y, e) is a C"-valued function whose entries are
holomorphic in a neighborhood of (x, y, c) = (0, 6, 0), p".. E C[(x]]" (m = 0,1, ... ),
p"o(0) = 6, and p(O, e) is convergent. Show that p(x, e) is convegent for ]x) < r and
IEI < p for some positive numbers r and p.
Hint. Regard p(ET, e) as the solution of the initial-value problem y = f (.ET, y, E),
d
00) = P(0. E.
XII-11. Let t and e be complex variables, y" E G", and the entries of a G"-
valued function f(t, y, e) be holomorphic with respect to (t, y. E) in a domain 1)o =
E) : ICI < do, -oo < tt < oo, Jyj < po, 0 < IEI < ro, I arg E) < ao). As-
00
some that f(t, y, E) E E'f,,,(t, y-) as e -+ 0 in the sector So = {E : 0 < IEI <
m=0
ro, I arg EI < ao } uniformly with respect to (t, y-) in the domain Do = {(t, y) : J 3'tI <
do, -oo < Iftt < +oc, Iyl < po}, where the coefficients fm(t,y-) are holomorphic in
398 XII. ASYMPTOTIC SOLUTIONS IN A PARAMETER
Ao. Assume also that f(t, y, e) and the coefficients fm(t, y) are periodic in t of
dy
period 1. Show that the differential equation = 27riy + e[y + e f (t, y, e)] has a
dt
periodic solution U = e) of period 1 such that
(a) the entries of (t, e) are holomorphic with respect to (t, e) in a domain D =
{ (t, e) : I `3'tJ < d, -oo < Rt < +oo, 0 < JeJ < r, I arg el < a} for some (d, r, a)
such that0<d<do,0<r<r0,and0<a<ao,
00
(b) j(t, e) ^ - - E em&m(t) as a -+ 0 in the sector S = If : 0 < Jej < r, J arg el <
M=0
a}, where coefficients pm(t) are holomorphic and periodic of period 1 in the
domain D' = It : J:33tI < d, -oo < Rt < +oo}.
Hint.
Step 1. Construct the solution +li(t, c", e) to the initial-value problem
Step 3. The particular solution ¢(t, e) = j(t, cue), e) is the periodic solution satis-
fying all of the requirements.
XII-12. Consider a system of differential equations
dy
(s) = xka(x,e)y,
dx
where am(e) E U and these quantities are holomorphic and bounded with respect
toe in a sector S = {e : 0 < je' < bo, I argel < bl}, and the series (a) is convergent
in norm for lx[ > Ra uniformly for e in S. Also, assume that a(x, e) admits an
asymptotic expansion in powers of a uniformly for x in jxl > Ro as e -. 0 in S.
Show that if a positive integer N is sufficiently large, there exist elements p(x, e),
bo(e ), ... , bM (e) of U such that
(i) p(x, e) is holomorphic and bounded in S and large [x[,
(ii) p(x, e) admits an asymptotic expansion in powers of e uniformly for large [xi
a positive integer and the quantities bo(e), ..., bkf(e) are holomorphic
in S and admit asymptotic expansions in powers of e as a -+ 0 in S,
(iv) m>k+l,
(v) the transformation y = [I + x-(N+1)p(x, e)Ju changes (s) to
N M
du E x-mam(e) + x- (N+1)
(S')
dx
xk
E x-mb,,,(e)I U.
Comment and Hint. See [Sill; in particular Theorem 2 on p. 157]. The point
x = 0 is not necessarily a regular singular point of (s') as in Theorem XII-6-1.
Step 1. The main idea is, assuming that N > k, to solve equations of the following
forms:
(1)
(N + 1) x -(k+l)Q(x) - x -k d((x) + a (x )Q( x ) - Q( x ) a ( x ) - B(x )
+ x-(N+1) [a(x)Q(x) - Q(x)B(x)] = F(x)
and
where
(1) Q(x) is an unknown quantity which should be a convergent power series
in x-1,
(2) B(x) is an unknown quantity which should be a polynomial in x-1 of
degree k,
(3) a(x) is a given polynomial in x-1 of degree N,
(4) a(x) is a given convergent power series in x-1,
(5) 0 is a given polynomial in x-1 of degree k,
(6) -r(x) is a given convergent power series in x-1,
(7) F(x) is a given convergent power series in x-1.
To solve these equations, first express B in terms of Q. In fact, setting
too k N
Q(x) _ E x-'Q,,,. B(x) _ F, x-mBm,
a(x) _ E X -0m,
m=0 m=0 m=0
we obtain
M
Bm = 1:fam-hQh - Qham-hl, m = 0,1,..., k.
h=1
Then, we can write (I) and (II) in the form
1
(III) QM
= (m>0),
where 9,,, is either quadratic or linear in Q (n > 0).
Step 2. If N > 0 is sufficiently large, we can solve (III) by defining a norm for a
+oo +oo
convergent power series P(x) = E x-'P,,, by 11P11 = E p'11 P,,, 11, where p is a
m0 m=0
sufficiently small positive number.
Step 3. Using Steps 1 and 2, we can construct a formal power series q(x, e) _
+00
E e'ge(x) such that the coefficients qt(x) are holomorphic and bounded in a disk
t=0
A = {x : Ix] > R > 0} and that the formal transformation y = [I+x-(N+1)4(x, e)Ju
changes (s) to (s').
Step 4. Find a function q(x, e) such that q is holomorphic and bounded in 0 x S
and that q(x, e) q(x, e) as e --+ 0 in S uniformly for x E A. Then, transformation
Y= [I + x-(N+1)q(x, e)Iv changes (s) to
(s")
k +oo
dv
dx
= xk
F x-mam(E) + x- (N+1) E x-mbm(E) + E x-mbm(0) 1 u,
M= m=0 m=k+1
where
bm(E) ^_- b,1 (E) as E 0 in S
and
x-mam(E) ^- 0
m=k+1
as 0 in S uniformly for x E A.
EXERCISES XII 401
Step 5. Using again Steps 1 and 2, find a function r(x, e) such that
(a) r(x, e) is holomorphic and bounded for jxj > R' > 0 and e E S,
(b) r(x, e) = 0 as a - 0 in S uniformly for jx[ > R',
(c) for a sufficiently large M' > 0, the transformation v = [I +x-(h1'+1)r(x,e)jw
changes (s") to
N M
dw = xk ` x-mam(f) + x -(N+1) E ? bm(e) w,
Lm=o
M-
whereb,,,(e)-0ase-0 inSform>k+1.
XII-13. Let A(t) be a 2 x 2 matrix whose entries are holomorphic in a disk It :
is not triangular for any 2 x 2 matrix P(t) such that the entries of P(t) and P(t)-1
are holomorphic in the disk D = It : jtj < po}. Show that there exists such a 2 x 2
)
matrix P(x) for which matrix (M) has the form xk B Gl with a 2 x 2 matrix B(t)
wh ose entries are polynomials in t and whose degree in t is at most k + 1.
Hint. See [JLP]. This result can be generalized to the case when A(t) is a 3 x 3
matrix (cf. [Ball) and [Bal2J).
XII-14. Let P(t) be a 2 x 2 matrix such that the entries of P(t) and P(t)-1
are
holomorphic in a disk V = it : jtj < po} and that the transformation y" = P (!) X
u"
dil [ l +x-1 0 1
dx x-3 y =
[ i12,
1-x'1Jy'
I.
to
dil u1
ds
Blx/u, u= 112
with a 2x 2 matrix B(t) whose entries are polynomials in t. Show that the degree
of the polynomial B(t) is not less than 2. Also, show that there exists P(t) such
that the degree of the polynomial B(t) is equal to 2.
Hint. To prove that there exists P(t) such that the degree of the polynomial B(t)
-17+ at
is equal to 2, apply Theorem V-5-1 to the system t d'r = 1 +tbt ] v with
.ii L t
suitable constants a, Q, y, and b. To show that the degree of the polynomial B(t)
is not less than 2, assuming that the degree of the polynomial B(t) is less than 2,
derive a contradiction from the following fact:
402 XII. ASYMPTOTIC SOLUTIONS IN A PARAMETER
0
The transformation y = 0 u changes the system
dy rl+x'1 0 1
dx l x-3 xtJ',
to
dil 1-x-11
dx = [lx-I u, u = IuaJ.
where r is a non-negative integer and ff (x, v1, t,2,... , vn) are holomorphic with
respect to complex variables (x, v1, v2, .... vn) in a domain
h=1 Ip1?2
403
404 XIII. SINGULARITIES OF THE SECOND KIND
k
(XIII.1.4) fi(x,v1 =00>fjk(la
k=0
f}0(x) f70kx-k,
fjp(x) ti 1: fjpkx-k,
k=1 k=0
a,h(x) ti00E
ajhkx-
(j, h = 1, 2,... , n)
k=0
(XIII.1.14)
D3 (N, y, q)
l
x: Ixi > N, jwj < ao,
(2q_)r+Y<wj+(r+1)w< (2q+)1r_v}
q is an integer, N is a sufficiently large positive constant, and -y is a sufficiently
small positive constant. For each j, there exists at least one integer q, such that
the real half-line defined by x > N is contained in the interior of V., (N, y, qj). Set
,w
n
(XIIL1.15) 7)(N,7) = n V, (N,y,qj).
=1
Proof.
The main claim of this corollary is that the asymptotic expansion of the solution
is valid in a sector I argx-8 < 2k +e whose opening is greater than k So, we look
at the sector D(N, y) of Theorem XIII-1-2. In the given case, ao = +oo, r = k -1,
and µ, = Aj (j = 1,... , n). The assumptions given in the corollary imply that
for any integer p. Therefore, for each 3, there exists an integer q j such that
k=1 k=1
du,
(XIII.2.5) = x'g,(x,U-) (.7 = 1,2,... ,n),
where
dx
Set
m
(XIII.2.6) g,(x, u") = go(x) + >2 b,k(x)uk +00E b,p(x)iI (j = 1, 2,... , n).
k=1 jp1>2
In particular,
and
b3k(x)-ajk(x)=O(IxI-') (j,k=1,2,...,n)
as x -+ oo in sector (XIII.2.2). Thus,
(XIII.2.12) lxl > No, Iargxl <oo, luil <6 (j =1,2,... ,n).
Here, by, is a positive constant depending on h. Furthermore, by virtue of Assump-
tion II, it can be assumed without loss of generality that the constant c satisfies
the condition
(XIII.2.15) u, = J
t trR, (t, u) exp Ir - 1(tr+l _ xr+1)1 dt (j = 1, 2,... , n),
J
where the paths of integration L3 start from x. The paths of integration must
be chosen carefully so that uniformly convergent successive approximations can
be defined in such a way that the limit is a solution uf(x) of (XIII.2.15) which
satisfies the conditions (i) u , (x) (j = 1, 2, ... , n) are holomorphic in D(N,'y) and
(ii) u., (x) ^ _ - 0 (j = 1, 2, ... , n) as x - oo in D(N, ry) for suitable positive numbers
N and ry.
Hereafter in this section, we explain how to choose paths of integration on the
right-hand side of (XIII.2.15).
Observation XIII-2-4. Since for each j, the domain Dj (N, y, q,) contains the
real half-line defined by x > N in its interior, their common part D(N, ry) is given
by the inequalities
where t and f are positive constants. It is noteworthy that if x E D(N, y), then x
satisfies the inequalities
(XIII.2.17)
argxI <ao, - 2r+y<wj +(r+1)argx< 2r-y
(j=1,2,...,n),
provided that w1 = arg u, are chosen suitably. Therefore,
For every point E D(N), the lines Lj4 in D(N) (except possibly for their
starting points) are defined by
I s = fo + t expji arg( - £o)], 0 < t < I - toI for j E G1 u G2,
(XIII.2.36) s = t + texp[i(r + 1)t'], 0 < t < oo for j E G3,
s=C+texp[-i(r+1)t], 0<t<oo for jEG4.
Note that from (XIII.2.28) and (XIII.2.29), it follows that
Ir+ <w (j E G1),
2 2 2 2
62= j: -2r+ry'<c3-(r+1)t<Zr
(XIII.2.38.2)
2r+-y' <w) +(r+1)t' < r-ry'},
-(r + 1)t < -(r + 1)t - 27' + 17r < (r + 1)t' + 27' - 2r < (r + 1)t'.
In D(N), let
C2Nr+lexp i -(r+1)t-+2r}
1,Y ,
2. BASIC ESTIMATES 413
For every t E D(N), the paths of integration in t(N) are defined in the following
manner:
(a) For j E G1, from (XIII.2.37) and (XIII.2.38.1), it follows that -27r + 27' <
w, -(r+1)t-21+rand w, -(r+l)t-2ry'+r<w1+(r+1)f'<2r-'y'and,
hence,
-(r+1)t- 12 7 + 2
<(r+ 1)e-2lr<(r+1)(- ir+
2
lry`.
2
Let
6 =Nr+1expli{(r+1)f -2r}J.
Then, o is on arc (A-3). Let t;' be a point on arc (A-3) such that argt2 < argt' <
arg {o. Then, the tangent T4, of the circle j = Nr+1 at the point t:' is given
by s = £' + t < +oo), where 0(l;') is continuous in {' and
-(r + 1)t - 2 + r < (r + 1)e'. Moreover, the part of T. in D(N) is given
by
-(r+1)e<B(t) <(r+1)P'+ I r
414 XIII. SINGULARITIES OF THE SECOND KIND
in )(N). This implies that - 2 + 2 < wj + O(l:) < 2 - 2 The function O(C) is
defined in a way similar to the previous case.
(c) For j E C3, from (XIII.2.38.3), it follows that
I
-17r+
2 2
< wj - (r+ 1)e- 'y'+1r <
2
17r
2
- 32
2
lry'-7r< 2
7- 1>'
Let
o = N''+1 exp [i{(r+1)t'+.i'_ 32rr1l .
In the case when (XIII.2.41) holds, the lines Lj( are defined by (XIII.2.39) with
32 Y.
in D(N). This implies that - 2 + < wj + 9({) < 2 - 2 In this case, o is
not on arc (A-3), but 8(e) is still defined in a way similar to the previous cases.
For all of the cases considered above, we prove the following lemma.
Lemma XIII-2-6. There exists a positive constant H independent of h such that
the inequalities
I -
(XIII.2.43)
Isi-11 I exp[
r+ 1
a11 Ids) <Hjl:I-h l exp 1r l ]
(j = 1,2,... n)
hold in a domain D(Nh) for any positive number h, where Nh is a sufficiently large
positive constant depending only on h.
2. BASIC ESTIMATES 415
Proof
First let j E G1 U G2. Then,
exp r+l{to+texP(iarg(t-to))}JI
exp [ r+1ji - I [-
- I exp [ r + 1 to] I I exp I - r +tl exp(i arg(t - Wd
exp
L r "J 1 to] I
exp I +I1 cos(wj + g(t - to))]
for E D(N). By virtue of (XIII.2.28), (XIII.2.29), and (XIII.2.35), it holds that
27r+2Y'<w, (9EG1),
1
Therefore, - cos(w, + arg(C - co)) > sin (-') for E D(Nh) and j E G1 U G2 .
Moreover, (XIII.2.34) implies
and (XIII.2.28) and (XIII.2.29) imply (r + 1)(t + t') < n - ry'. Hence,
(XIII.2.46) dr = { Mj+ h M2 + T2 IY + 1
+ 2f Mr
416 XIII. SINGULARITIES OF THE SECOND KIND
'r + am
and Af, < - rµ+l < 0. From (XIII.2.45), it follows that
M2 + r + 2c"Mr
i 2hµJIb21),
Mb (r > 0). If M satisfies an inequality M > then Y+
2(1
T- < r + 1)
2(r + 1)
1. Since Y = 0 for r = 0, we obtain Y(r) < (r > 0). This implies that
Ipj lb
exp 1 112r+
i
1)
x ISI-h exp I exp{i arg(e - o)}J I Idsl < 2(lu
JG,c l - r +tl Ib
for E D(N),).
(i) Consider the case when 10(x) - a given point { E D(Nh). Then,
Isle =112 + t2 - cos(sr - IB(S) - argil) ? IC12.
f
Therefore,
exp exp [ -
r±1 ( +
I31-h
Idyl <
1
J,< L
r+1
p1 S 0
I dt
5).
FIGURE 4. FIGURE 5.
3. PROOF OF THEOREM XIII-1-2 417
that Y(r) < M . On the other hand, Y(r) satisfies the differential equation dY =
{
M+h
D2 + r2 }
Y -1. Since
r
D2+72
< 1
2D
, it follows that
dY
dr
>
lit, Ib
2(r + 1)
Y-1
h(r + 1)
if M > . Here, a use was made of the inequality M -< D. Since Y(0) <
1i2 1b
1 < (r + 1)
we obtain Y(r) <
2(r i 1) (r < 0). Therefore, (XIII.2.43) follows.
M2 1A2 ib 1µ, b
2(r + 1)
This completes the proof of Lemma XIII-2-6. In this case, again, H = µb ,
where µ =min{µ2 }. 0
2
as x oo in (XIII.1.15).
Observation XIII-3-1. As in the previous section, denote by Do(Nh) the domain
in the x-plane which corresponds to the domain D(Nh) in the -pllane. The domain
Do(Nh) depends on h. Set o i = Nh+1 exp {i(r + 1)(e + and denote by
virtue of (XIII.2.13), Nh, can be chosen so large that Nh, > No and b < d (cf.
(XIII.2.12)). Fix Nh, in this way. Then, by a method similar to that in §XII-3 for
each j (j = 1, 2, ... , n), successive approximations can be defined to construct a
solution u2 = O,(x) of (XIII.2.15) which is holomorphic in Do(Nh-) and 101(x)l <
(j = 1,2,... ,n), where = xT+'. In this way, the existence of a bounded
solution u2 = 02 (x) (j = 1, 2, ... , n) of system (XIII.2.15) is proved.
418 XIII. SINGULARITIES OF THE SECOND KIND
UVG1UG2)
in the domain Do(Nk). Here, u(x) denotes the Cn-valued function whose entries are
(u1(x),u2(x),... ,un(x)). Note that we can assume Do(Nk) C Do(Nh') without
loss of generality.
Upon applying successive approximations similar to those of §XII-3 together
with Lemma XIII-2-6 to (XIII.3.2), it can be proved that (XIII.3.2) has a solution
u,(x) = 0., (x) (j = 1,2,... ,n) such that IiP, (x) I < CkIxI-k U = 1,2,... ,n) in
Do(Nk), where Ck is a suitable positive number. Also, using Lemma XIII-2-6 and
(XIII.2.13), it can be shown that ¢j (x) = ikj (x) (j = 1,2,... , n) in VO(Nk), since
where j = 1, 2,... , n, and ¢(x) and tG(x) denote en-valued functions whose entries
are (01(x), 02(x), ... , On(x)) and (ti1(x), ,11 (x)), respectively. Thus, we
conclude that 0, (x) 0 (j = 1, 2,... , n) as x -' oo in Do(Nh,).
Now, let us return to Observation XIII-2-2. If we set pj(x) = Oj(x) + zj(x), the
solution v , = pj(x) (j = 1, 2, ... , n) of (XIII.1.1) satisfies all of the requirements of
Theorem XIII-1-2. 0
Remark XIII-3-3. Let Ax, y', µ, e) be a Cn-valued function of a variable (x, e)
E C x Cn x Cn x C such that
(a) the entries of f are bounded and holomorphic in a domain
Do = { (x, b, µ, e) : IxI > Ro, Iu-1 < do, lµl < p o, IeI < Eo, I arg EI < po},
where R0, 60, µo, co, and po are some positive constants,
(b) f admits a uniform asymptotic expansion in Do
+0
r/x,
s/
a/, e)
ti Ehfh(x, . p) as e - 0,
h=0
(c) if f (x, y, µ', e) = fo (x, #,c) + A(x, µ, e)y + O(I y12), then
in So, where coefficients &(µ,e) are bounded, holomorphic, and admit uni-
form asymptotic expansions in powers of a as c -+ 0 in the domain {(µ, e)
Iµ1 < {31, 0 < lei < el, I argel < po},
(iii) the solution 1'(x, N, e) admits an uniform asymptotic expansion:
+oo
(x, , e) ' ehWh(x, µ) as f -4 0
h=0
in So, where coefficients >Gh(x,#) are bounded, holomorphic, and admit uni-
form asymptotic expansions in powers of x'1 as x - oo in the domain
{(x,µ') : Ixf > N, x E S(p1,p2), Jill < p1}.
FLrther more, there exist functions tc(x, µ, e) for t = 0,1, ... such that
(a) these functions ¢'c satisfy conditions (i) and (is) given above,
(b)
h=0
A complete proof of Theorem XIII-3-4 is found in [Si7J and (Si101.
420 XIII. SINGULARITIES OF THE SECOND KIND
Remark XIII-3-5. In the proof of Theorem XHI-1-2, we used the assumption that
the matrix Ao on the right-hand side of (XIII.1.10) is invertible (cf. Assumption
III of §XIII-1). Without such an assumption, we can prove the following theorem.
Theorem XIII-3-6. Let F(x, yo, yr, ... , yn) be a nonzero polynomial in yo, yl, ,
00
yn whose coefficients are convergent power series in x'1, and let p(x) = an,x-°`
M--0
E C[(x-1)) be a formal solution/ of the differential equation
(XIII.4.8) y = P(x)zi
di ,
(XIII.4.9) = x B(x)i,
Ell E2 Eu Qtt
Q11 Q2t .
where EJk and QJk are nJ X nk matrices and F, are nJ x nJ matrices. Set
(XIII.4.16) QJJ = 0 (j = 1,2,... ,t).
From (XIII.4.4), (XIII.4.14), (XIII.4.15), and (XIII.4.16), it follows that
(XIII.4.17) F,=EJJ+FEJhQh, (j=1,2,...,t)
h#J
and
dQjk
(XIII.4.18) = x' [43Qik - QJkAk + EJk + EJhQhk - QJkAk] (j j4 k).
h*k
Substituting (XIII.4.17) into (XIrII.4.18), a system of nonlinear differential equations
dQJk
= x' LAJQJk - QJkAk + EJhQhk
dx
{ h*k
(XIII.4.19)
- QJk (Ekk + EkhQhk) + EJkJ (j # k)
h*k
is obtained. Since it is assumed that .11i ... , Al are distinct eigenvalues of Ao and
that A0 is in the block-diagonal form (XIII.4.5), upon applying Theorem XIII-1-2
to (XIII.4.19) we can construct a desired holomorphic solution Qjk(x) of (XIII.4.19)
00
which admit an asymptotic expansion Qjk(x) > x "Qjk" (j, k = 1, 2,... , 3; j
"=1
k), where Qjk" are constant nj by nk matrices. Defining Fj by (XIII.4.17) and then
B(x) by (XIII.4.13), the proof of Theorem XIII-4-1 is completed. 0
Theorem XIII-4-1 concerns the behavior of solutions of system (XIII.4.1) near
x = oo. Since it is useful to give a similar result concerning behavior of solutions
near x = 0, we consider, hereafter in this section, a system of differential equations
where d is a positive integer and the entries of n x n matrix A(x) are holomorphic
in a neighborhood of x = 0. Also, assume that A(0) is in a block-diagonal form
()III.4.21) A(0) = diag[A11,,, +N2,... +JVt],
where a1, ... , .1t are distinct eigenvalues of A(0) with multiplicities n1, n2, ... , nt,
respectively (n1 + n2 + + nt = n), and, for each j, .W is a lower-triangular and
nilpotent nj x nj matrix.
Comparing the present situation with that of Theorem XIII-4-1, we notice the
following two differences:
(a) singularity is at x = 0 in the present situation, while singularity is at x = 00
in Theorem XIII-4-1,
(b) The power series expansion of A(x) is convergent in the present situation,
while A(x) in Theorem XIII-4-1 admits only an asymptotic expansion in a
sector containing the direction arg x = 0.
We can change any singularity at x = 0 to a singularity at x = oo by changing
1.
the independent variable x by Also, any direction argx = 8 can be changed
x
to the direction arg x = 0 by rotating the independent variable x. Furthermore,
the asymptotic expansion P of Po(x) and the expansion b of Be are formal power
series satisfying the equation xd+i dP = AP - PB. This implies that two matrices
P and b are independent of P. Hence, using Corollary XIII-1-3, the following result
is obtained.
Theorem XIII-4-2. Let A(x) be an n x n matrix whose entries are holomorphic
in a neighborhood of x = 0. Also, let d be a positive integer. Assume that the matrix
A(0) is in block-diagonal form (XIII.4.21), where al, ... , at are distinct eigenvalues
of A(0) with multiplicities n1, n2, ... , nt, respectively (n1 + n2 +, + nt = n), and
for each j, JVj is a lower-triangular and nilpotent n, x nj matrix. Fix a real number
0 so that (a, - Xk)e-utO V EP for j 96 k. Then, there exists two positive numbers be
and ee and an n x n matrix Pe(x) such that
(a) the entries of Po(x) are holomorphic and admit asymptotic expansions in pow-
ers of x as x - 0 in the sector Se = Ix: 0 < IxI < be, I arg x - 01 < + to },
00
(b) if >2 xmP,,, is the asymptotic expansion of Pe(x), then this expansion is
m=0
independent of 8 and Po = I,,,
(c) the transformation PB(x)u" changes system (XIIL4.20) to a system
K=
I 9:
p E C[Ix]J, 9 E C[Ix]j, 9 34 0}
Also, denote by V the set of all row vectors (cl(x), c2(x), ... , c,,(x)), where the
entries are in the field K. The set V is an n-dimensional vector space over the
field 1C.
Define a linear differential operator C : V - V by G[vl = by + 7l(x) (v' E V ),
where b = x d
and S2(x) is an n x n matrix whose entries are in the field 1C. We
first prove the following lemma.
Lemma XIII-5-1 (P. Deligne [Del]). There exists an element iio E V such that
{v"o, Gv"o, G2v"o, ... , G"-lvo} is a basis for V as a vector space over IC.
Proof.
For each nonzero element v of V, denote by µ(v'' the largest integer t such that
{ii, Cii, C2v, ... ,.CeV) is linearly independent over K. In two steps, we shall derive
a contradiction from the assumption that max{µ(v') : v' E V} < n - 1.
5. CYCLIC VECTORS 425
(4) a sety{ {6j, 62,... , Vm } E V is linearly dependent if and only if v'1 AV2 A A 17m =
6 in K('^).
Step 2. Fix an element Vo of V such that µ(6o) = max{p(V) : v" E V} < n-1. Since
p(VO) < n-1, another element w of V can be chosen so that {VO, Cv"o, CZi3o, ... , CnOv',
w"} is linearly independent, where no = max{µ(v") : v E V}. Set v' = vo+Ax"'O E V,
where A E C and m is an integer. Then, Cwt = Cw"o + C'(Ax-ti) = Crvo +
Axm(C + m)tw. Since {V, Cv", ... , Cn0+1V} is linearly dependent, it follows that
v A DU A A CnOv' A C"O+I V = 6 for all A E C and all integers in. Note that
v A,06 A A C"OiYA C"°+16 is a polynomial in A. Since this polynomial is iden-
tically zero, each coefficients must be zero. For example, the constant term of this
polynomial is i6o A CVO A ... A 00+I V"o. This is zero since {iio, Ciio..... G"0+Iuo}
is linearly dependent. Compute the coefficient of the linear term in A of the poly-
nomial. Then,
m)n0,+1ti
w A C60 A ... A Cna+1v'o + v'Y A ... A 00 Vo A (C +
no
+ 1:60A...ACi-IVoA (C+m)lw A CJ+'A A...ACnb+Ii = 0
J=1
426 XIII. SINGULARITIES OF THE SECOND KIND
identically for all integers m. The left-hand side of this identity is a polynomial in
in of degree no + I. Hence, each coefficient of this polynomial must be zero. In
particular, computing the coefficient of mn0+1, we obtain vOAGvOA . A GnbtloAw' =
0. This is a contradiction, since {vo, Gvo, ... , L"Ovo, w} is linearly independent.
This completes the proof of Lemma XIII-5-1.
Definition XIII-5-2. An element vo E V is called a cyclic vector of G if {6o, Lu0,
L2v'o, ... , Ln-1%) is a basis for V as a vector space over X.
Observation XI II-5-3. Let uo be a cyclic vector of Land let P(x) be the n x n ma-
Uo
Gvo
trix whose row vectors are {vo,G'o,... ,L' 'io}, i.e., P(x) = . Then,
Gn-lvo
nv"o
L21 yo
o
= and, hence, setting A(x) = L[P(x)]P(x)-1 = bP(x)P(x)-l +
P(x)f2(x)P(x)-1, we obtain
0 1 0 0 .. 0
0 0 1 0 ... 0
A
0 0 0 0 ... 1
ap al a2 a3 ... an-1
with the entries a, E C. Thus, we proved the following theorem, which is the main
result of this section.
Theorem XIII-5-4. The system of differential equations
y1
(XIII.5.1) by' = fl(x)y", where g _
Y.
becomes
Example XIII-5-5.
(1) Let us consider the system
y1
(a) by" = 0, where y=
yn
The transformation
(T) u = diag[l, X'... , xn-11y
changes system (a) to
(E) bu = diag[0,1,... ,n - 1]u.
Further, the transformation
1
2
-- 22 U
(r)
2n-1
ao a1 a2 a3 ...
an-1
changes (a) to (E'). This implies that vo = (1, x, x2'. .. xn-1) is a cyclic vector in
this case.
(II) Next, consider the system
(b) by = Ay,
where A is a constant diagonal n x n matrix. Choose a transformation similar to
(T) of (a) to change system (b) to
(E") oiZ = A'u
428 XIII. SINGULARITIES OF THE SECOND KIND
so that A' is a diagonal matrix with n distinct diagonal entries. Then, a transfor-
mation similar to (T') can be found so that (E") is changed to (E') with suitable
constants ao, al, ... , an_ I .
(XIII.6.1)
where the entries of the n x n matrix A are in K (cf. §XIII-5). In order to state
this theorem, we must introduce a field extension f- of K. To define L, we first set
+a
E a.nxm1" : a.., E C and M E Z ,
M=M I
+a
where 7L is the set of all integers. For any element a = F a,nxn'/° of K,,, we
M=M
+00
define x ji by x da = F \ v) amx'nl '. Then, K, is a differential field. The field
M=M
+oo
L is given by L = U K which is also a differential field containing K as a subfield.
L=1
Furthermore, L is algebraically closed. The Hukuhara-T rrittin theorem is given
as follows.
(XIII.6.2) y' = UI
such that
(i) the entries of the matrix U are in L and det U ;j-1 0,
(ii) transformation (XIII.6.2) changes system (X111.6.1) to
(XIII.6.3) xd = Bz',
B=diag[Bl,B2,...,Bpj, Bj=diag[BJi,Bj2i...,B,,n,],
(XIII.6.4)
Bjk =AI In,,. + Jn,,,.
6. THE HUKUHARA-TURRITTIN THEOREM 429
Here, I,,,, is the njk x njk identity matrix, J.,,, is an n 1k x n3k nilpotent
matrix of the form
0 1 0
0 0 1
(XIII.6.5) Jn",
0 0 0
0 0 0
Proof.
Without loss of generality, assume that the matrix A of system (XIII.6.1) has
the form
0 1 0 0 0
0 0 1 0 ... 0
A
0 0 0 0 ... 1
00 a1 a2 a3 "' an-1
an- 1
where ak E 1C (k = 0, 1, ... , n - 1) (cf. Theorem XIII-5-4). Set E= A - n In.
Then,
(XIIi.6.8) trace [E] = 0.
Consider the system
(XIII.6.9) X E = E.
Case 1. If there exists an n x n matrix S with the entries in X such that det S 0 0
dii
and the transformation w = Sii changes (XIII.6.9) to x = A(x)u", where the
entries of A are in C[[x]], then there exists another n x n matrix S with the entries
in X such that det S 96 0 and the transformation
(XIII.6.10) w = Suu
dig
changes (XIII.6.9) to x = Aoii, where the entries of the matrix A0 are in C.
F irthermore, any two distinct eigenvalues of A0 do not differ by an integer (cf.
Theorem V-5-4). Hence, in this case, system (XIII.6.1) is changed by transforma-
tion (XIII.6.10) to
dil
xaj = [an 1 In+Ao]u.
This proved Theorem XIII-6-1 in this case.
430 XIII. SINGULARITIES OF THE SECOND KIND
1 0 0 0
- ln an_1 1 0 0
E =
0 0 0 - ln an_1 1
a cyclic vector can be found for system (XIII.6.9) by using the matrix W defined
by
w11 win
with
writ Wnn
[wll ... Win] = (10 ... 01,
[w,,1
... wJn] = V1-1[1 0 01,
triangular and the diagonal entries are {1, ... ,1}, i.e.,
1 0 0 0
1 0 0
(XIII.6.11) W=
1
Also,
0 1 0 0 0
0 0 1 0 0
x + WE] W-1 = V[ W]W-1
11
0 0 0 0 l
22 A
W i31 /32 An-1
6. THE HUKUHARA-TURRITTIN THEOREM 431
k=max(n µt t :IE91. J
Then, ut < k(n - t) for every t E J and µt = k(n - t) for some t E J. This implies
that
(I) Of = L.
Qt,mxm
(t=0,1,... n- 1)
m>-k(n-t)
and
Then,
(XIII.6.14)
where
0 1 0 0
0 0 1 0
(XIII.6.15) F= + kxkdiag [0,1, ... , n - 1]
0 0 0 0
70 71 72 73
and
7t = xk(n-t)o, =:m>-k(n-t)
$,,mxm+k(n-t) E C[ [x'1911 (0 < t < n - 1).
In particular, 7n-1 = 0.
432 XIII. SINGULARITIES OF THE SECOND KIND
+oo
Setting F = E xm/9Fm, where the entries of F,,, are in C, we obtain
m=o
0 1 0 0 0 01
Fo=
CO
C1 C2 C3 ... Cn_2 0
where the constants co, cl ... , cn_2 are not all zero. This implies that the matrix
F0 must have at least two distinct eigenvalues. Hence, there exists an n x n matrix
T such that
(1) T = XM19Tm, where the entries of the matrices Tm are in C and To is
m=o
invertible,
(2) the transformation
(XIII.6.16) y = Ti
where
if dj = 0,
-t/s if d1>0.
Step 3. Set
Then,
The matrix
with transformation (XIII.6.2) such that the entries of the matrix U of (XIII.6.2)
are holomorphic in S and each of them is in a form x,00(x) where p is a rational
number and O(x) admits an asymptotic expansion in powers of xl"° as x 0 in S,
where s is a positive integer.
Observation XIII-6-5. In the case when the entries of the matrix A(x) on the
right-hand side of (XIII.6.19) are in C[[x]l and A(0) has n distinct eigenvalues, the
matrix A(x) also has n distinct eigenvalues A1(x), A2(x), ... , which are in
C((xJJ. Furthermore, the corresponding eigenvectors p"1(x), p"2(x), ... ,15n(x) can be
constructed in such a way that their entries are in Chill and that p""1(0),7"2(0), ... ,
434 XIII. SINGULARITIES OF THE SECOND KIND
p",,(0) are n eigenvectors of A(O). Denote by P(x) the n x n matrix whose column
vectors are p1(x), p2(x), ... , p,+(x). Then, detP(O) 36 0 and P(x)-lA(x)P(x) =
diag[A1(x),A2(x),... ,An(x)]. This implies that the transformation y = P(x)ii
changes system (XIII.6.19) to
where µ1(x), µ2(x), ... , i (x) are polynomials in x of degree at most d such that
A, (x) = it) (x) + O(xd+1) ( j = 1, 2, ... , n). Therefore, in this case, the entries of
the matrix U of transformation (XIII.6.2) are in 1C.
Observation XIII-6-6. Assume that the entries of A(x) of (XIII.6.19) are in
C([x]]. Assume also that A(O) is invertible. Then, upon applying Theorem XIII-6-1
d1
to system (XIII.6.19), we obtain = d for all j. Using this fact, we can prove the
s
following theorem.
Theorem XIII-6-7. Let Q;i1(x) and Qi2(x) be two solutions of a system
where d is a positive integer, the entries of the n x n matrix A(x) and the C"-valued
function 1 *(x) are holomorphic in a neighborhood of x = 0, and A(O) is invertible.
Assume that for each j = 1, 2, the solution ¢,(x) admits an asymptotic expansion
in powers of x as x - 0 in a sector S3 = {x E C : lxl < ro, aj < arg x < b3 },
where ro is a positive number, while aJ and bi are neat numbers. Suppose also that
S1 n S2 0. Then, there exist positive numbers K and A and a closed subsector S =
{x : lxl < R,a < argx < b} of Sl nS2 such that K exp[-Alxl -d]
in S.
Proof.
Since the matrix A(O) is invertible, the asymptotic expansions of1(x) and ¢2(x)
are identical. Set 1 (x) = ¢1(x) -$2(x). Then, the C"-valued function >G(x) satisfies
system (XIII.6.19) in S, n$2 and ii(x) ^_- 6 as x 0 in S1nS2. By virtue of Theorem
XIII-6-4, a constant vector 66 E C" can be found so that r%i(x) = U4D(x)6, where
lb(x) is given by (XIII.6.17). Now, using Observation XIII-6-6, we can complete
the proof of Theorem XIII-6-7.
6. THE HUKUHARA-TURRITTIN THEOREM 435
Observation XIII-6-8. The matrix A = diag [AI In,, A2In3, ... , API,y] on the
right-hand side of (XIII.6.18) is unique in the following sense. Assume that an-
other formal fundamental matrix solution Uxcexp[A] of system (XIII.6.1) is con-
structed with three matrices U, C, and A similar to U, C, and A. Since the
matrices Uxc exp[A] and UxO exp[A] are two formal fundamental matrices of sys-
tem (XHI.6.1), there exists a constant n x n matrix r E GL(n, C) such that
Uxc exp[A] = U5C exp[A]I' (cf. Remark IV-2-7(1)). Hence, exp[A]T exp[-A] =
x-CU-IUxC. Using the fact that r is invertible, it can be easily shown that A = A
if the diagonal entries of A are arranged suitably. For more information concerning
the uniqueness of the Jordan form (XIII.6.3) and transformation (XIII.6.2), see, for
example, [BJL], [Ju], and [Leve].
Observation XIII-6-9. The quantities A.,(x) are polynomials in x1l'. Set w =
2a[!]
exp and x 1/a = wx 1/a Then, i = x. Therefore, if z 1/e in Ur Cexp[A] is
replaced by zI/', then another formal fundamental matrix of (XIII.6.1) is obtained.
This implies that the two sets {Aj (i) : j = 1, 2,... , p} and (A.,(x) : j = 1, 2,... , p)
are identical by virtue of Observation XIII-6-8.
Observation XIII-6-10. A power series p(x) in x1/' can be written in a form
a-1
Ax) = Exh1'gh(x), where ql(x) E C[[x]] (j = 0, 1, ... , s - 1). Using this fact and
h=0
Observation XIII-6-9, we can derive the following result from Theorem XIII-6-1.
Theorem XIII-6-11. There exist an integer q and an n x n matnx T(x) whose
entries are in C[[x]] such that
(a) det T (x) 96 0 as a formal power series in x,
dil
(b) the transformation y" = T(x)t changes system (XII1.6.1) to x = E(x)iZ
with an it x it matrix E(x) such that entries of x9E(x) are polynomials in x.
The main issue here is to construct, starting from Theorem XIII-6-1, a formal
transformation whose matrix does not involve any fractional powers of x in such a
way that the given system is reduced to another system with a matrix as simple as
possible. A proof of Theorem XIII-6-11 is found in [BJL].
Changing the independent variable x by x-1, we can apply Theorem XIII-6-1 to
singularities at x = oo. The following example illustrates such a case.
Example XIII-6-12. A system of the form
x F(x)
1 0 f1 1 e 0
0 x-1/2 ] l I -111 0 eE(t,a) I
436 XIII. SINGULARITIES OF THE SECOND KIND
where
m 1/2 +00
+ E ak
xk + E bk(a)
xk
i
k=1 k=1
2 (a)x(m+2-2h)/2,
E(x, a) = (m 2) x(m+2)/2 + (.m + 2 - 2h) bh
+ 1<h<(m+2)/2
+00
and F(x) = xq> x-h Fr, with an integer q and 2x2 constant matrices Fh such that
h--O
+00
det h=O x-h Fh ,-6 0 as a formal power series in x-l. For details of construction,
where b = x , at E C((xj), and an # 0 (cf. Theorem XIII 5 4). For the differential
operator
n-1
(XIII.7.4) C(>,J = anb"r) + j:atdtrt.
t=o
Definition XIII-7-4. The boundary curve C of the smallest convex set containing
P is called the Newton polygon of the operator C at x = 0.
Denote by N(C) the Newton polygon of C at x = 0.
Definition XIII-7-5. Two Newton polygons are said to be identical if the two
polygons become the same by moving one or the other upward in the direction of
the Y-axis.
Now, we prove the following theorem.
Theorem XIII-7-6. If system (XIII.6.1) and differential equation (X111.7.3) are
equivalent in the sense of Theorem X111-5-4. then the two Newton polygons N and
N(C) are identical.
Proof.
The proof of this theorem will be completed if the following three statements are
verified:
(a) If./V(C) has only one nonvertical side with slope k, then differential equation
(XIII.7.3) is equivalent to a system xk+1 = A(x)g with a matrix A(x)
whose entries are power series in x11", and A(0) is invertible if k > 0, where
s is a positive integer such that sk is an integer.
438 XIII. SINGULARITIES OF THE SECOND KIND
(b) If (XIII.7.1) gives slopes of all nonvertical sides of N(L), then the operator L
is factored in the following way:
(XIII.7.5)
where, for each h, N(Lh) has only one nonvertical side with slope kh.
(c) If L is factored as in (b) and each differential equation Lh[r)h] = 0 is equiv-
alent to a system xdih = Ah(x)uh, then the differential equation (XIII.7.3) is
Then, Statement (c) can be verified recursively on q without any complication. The
proof of Theorem XIII-7-6 in detail is left to the reader as an exercise.
Combining Observation XIII-7-1 and Theorem XIII-7-6, we obtain the following
theorem.
Theorem XIII-7-7. If the distinct slopes of the nonvertical sides of N(L) are
given by (XIII.7.1), then the n-th-order differential equation (XIIL7.S) has, at x =
0, n linearly independent formal solutions of the form
where
(i) if
{(X, Yh-I + kh(X - Xh_1)) : Xh-1 C X < Xh}
is the nonvertical side of N(L) of slope kh, then
and the quantities Oh,v(x) are polynomials in log x with coefficients in C[[x'/']]
Here, s is a positive integer such that skh (h = 1,... , q) are integers, and PI + P2 +
+ Pq = n.
A complete proof of Theorem XIII-7-7 is found in [St].
We can construct formal solutions (XIII.7.6), using an effective method with the
Newton polygon N(L). The following example illustrates such a method.
7. AN N-TH-ORDER LINEAR DIFFERENTIAL EQUATION 439
1 3
FIGURE 6.
+00
(i) For k1 = 0, set i _ E c,,,xX+m (co 0). Then,
M=0
+00
C[q] _ { (A + m)3 - (,\ + m)2} c,,,xa+m+1
M=0
t o0
(A + m + 1)C,,,xa+m = 0.
m=o
Hence,
((A + 1)co = 0,
1
(A+m+1)Cm = {(A+m-1)3-(A+m-1)2}C,,,_1 for m > 1.
Therefore,
2)2(m -3)
A = -1 and c,,, _ (m - c,,,_1 for m > 1.
m -
Thus,
A = -1, cl = -2co, c, ,, = 0 for m>2.
This implies that q = CO (x-1 - 2) is a solution of C[r7] = 0.
(ii) For k2 = 2, set n = exp[Ax-112](.
Then,
x_1 2)
b"['1] = exp[Ax-112] \a - [S] (n =0,1,2.... ).
3 )2_
ft Cb - Zx-1/2) - x I tS - -\X-112 (b - 2x-1/2) - 1 0.
440 XIII. SINGULARITIES OF THE SECOND KIND
Since
2
Ax-1/2
2x-1/21 - \x-112d + + 42x-1,
J
3
2x-1/2 =d3 - tax-1/252 + (x_1/2 4,\2x-I
/ + / d
- 18,\x-1/2
+
3,\2X-1 + 8A3x-3/21
,
it follows that
/ 2_
XId- 2x-1/2/3 - x I d - 2x-1/2 (d - x-1/2) - 1
= x63 - (xhI2 + x 62
+ (.x2 - 1 + 4Ax1/2) d
2 4
The Newton polygon of this operator has only one nonvertical side of slope 2 for
arbitrary A (cf. Figure 7-1). However, if A is determined by the equation
A \3
2 8 0,
then the Newton polygon has a horizontal side (cf. Figure 7-2).
where p1 and p2 are constants and f1 and f2 are formal power series in C[[x1/2JJ.
8. GEVREY PROPERTY OF ASYMPTOTIC SOLUTIONS 441
and G(x, yo, yl, ... , y.-I) is a convergent power series in (x, yo, yl ... y,,-,) (cf.
(SS31 and [Mal2j). Here, it can be assumed without any loss of generality that
(i) ah (h = 0,1, ... , n) are convergent power series in x and a # 0,
00
(ii) differential equation (XIII.8.1) has a formal solution p(x) = E a,,,x"' E
m=0
CI[x]I,
(iii) M is an integer such that for any differential operator At of order not greater
than n, the two Newton polygons N(C - xMIC) and N(C) are identical (cf.
Definitions XIII-7-4 and XILI-7-5).
Using Theorem XIII-3-6, we can find
(a) a good covering {S1, $2, ... , SN } at x = 0,
(b) N solutions 01(x), 02(x), ... , ON (x) of (XIII.8.1) in S1, S2,... , SN, respec-
tively such that of are holomorphic and admit the formal solution p(x) as
their asymptotic expansions as x -+ 0 in St, respectively.
Set ui = 01 - 0t+1 on Se n Sjt+1. Then, ue are flat in the sense of Poincare
in sectors Se n S(+1, respectively, where Sv+1 = S1. Furthermore, if we define
differential operators IQ by
Kt E -(x,...
h
bh (tot + (1 - t)-Ol+l I.... )dt bh,
O<h<n-1 LJO 1
then
(C - x"'K t)[ut] = 0 on se n St+l (f = 1, 2, .... N).
If the Newton polygon N(,C) has only one nonvertical side of slope 0, then x = 0
is a regular singular point of C[iI] = 0. Therefore, in this case the formal solution
p(x) is convergent (cf. Theorem V-2-7). Let us assume that./V(C) has at least one
side of positive slope. In such a case, let
be all of the positive slopes of the Newton polygon N(,C). Then, since N(1 C -xMJCe)
and N(G) are identical, we must have
Ixl-k]
lut(x)l : -yeXp(-A on Sr n Se+1
for a non-negative number -y and a positive number A, where k E {k1, k2 ... , ky}
(cf. Theorem XIII-7-7). Now, by virtue of Theorem XI-2-3, we obtain the following
theorem.
Theorem XIII-8-1. Under the assumptions given above, the formal solution p(x)
is a formal power series of Gevrey order 1 and, for each t, the solution 0e admits
p(x) as its asymptotic expansion of Gevrey order k as x -. 0 in S1, where k E
{k1ik2... ,kq}.
This theorem was originally prove in (Raml] for a linear system. For nonlinear
cases, see, for example, ISi17, §A.2.4, pp. 207-211J.
Remark XIII-8-2. In Exercise XI-14, we gave the definition of a k-summable
power series. As stated in Exercise XI-14, if a formal power series f E C((x]]
is k-summable in a direction argx = 0, there exists one and only one function
F E Al /k (Po, 0 - 2k
-e,6+ 2k + e) such that J(F] = f , where Po and a are
positive unmbers. This function F is called the sum of f in the direction arg x = 0.
If we use the idea of Corollary XIII-1-3 and Theorem XIII-6-7, we can prove the
following theorem concerning the k-summability of a formal solution of a nonlinear
system
EXERCISES XIII
dg _ x+ 5 x+ 8 where `yil
dx 3 -x+ 1 y l yz J
(a"f)(r) + rzf(x) = x
Hint. One method is to show that the given equation has a unique power series
solution in X-1 which is divergent at x = oo. Another method is to observe that
any solution of this equation has no singularity in ]x] < +oo except possibly at
x = 0. Furthermore, if p(x) is a rational solution, then some inspection shows that
p(x) does not have any pole at x = 0. This implies that p(x) must be a polynomial.
But, we can easily see that this equation does not have any polynomial solution
(cf. {Si20J).
XIII-3. Find a cyclic vector for the differential operator £[ 3 = x;ii + 'A, with a
constant n x n matrix A of the form A = I At 0 1, where for each j = 1, 2, the
444 XIII. SINGULARITIES OF THE SECOND KIND
Hint. For the given system, a cyclic vector is (1, 0, 0). This implies that the trans-
formation
1 0 0
(T) X-2 x-2 0
X-4-2x-2 x-4 + x-3 - 2x-2 x-
The Newton polygon of (E) has three sides with slopes 0, 1, and 2, respectively.
On the other hand, The standard form of the given system in the sense of
Theorem XIII-6-1 is
a(x) 0 0
3 du
X - = 0 b(x) 0 u,
dx
0 0 x2
where
a(x) = 1 + 0(x), b(x) = x + 0(x2).
Hence, this also shows that the Newton polygon of the given system has three sides
with slopes 0, 1 and 2, respectively.
EXERCISES XIII 445
XIII-6. Let A(x) be an n x n matrix whose entries are holomorphic and bounded
in a domain Ao = {x : 3x! < ro} and let j (x) be a Cn-valued function whose
entries are holomorphic and bounded in the domain Ao. Also, let Al, A2, ... , An
be eigenvalues of A(O). Assume that det A(0) # 0. Assume further that two real
numbers 01 and 82 satisfy the following conditions:
(1) 01 < 02,
(2) none of the quantities Aye-'ke (j = 1, 2, ... , n) are real and positive for a
positive integer k if 01 < 8 < 02,
(3) Ape-ike' and Aqe-'k93
are real and positive for some p and q.
Show that there exist one and only one solution f = fi(x) of the system x'`+1 ds =
A(x)f+ x f (x) such that the entries of fi(x) are holomorphic and admit asymptotic
expansions in powers of x as x - 0 in the sector S = {x : 0 < !xj < r0, 81 - <
2k
argx<82+2k
Hint. If we use Corollary XIII-1-3 at x = 0, it can be shown that for each 8 in the
interval B1 - < B < B2 + 7, a solution ¢'(x; 8) is found so that the entries of
2k
O(x; 8) are holomorphic and admit asymptotic expansions in powers of x as x - 0
'r
in the sectorial domain Sa = x : 0 < jxj < ro, I arg x - 81 < 2k + ea }, where Ee is
a sufficiently small positive constant depending on 8. If 10 - 8'l is sufficiently small,
then ¢(x; 0) = d(x; 8).
XIII-7.
(a) Show that j (x) _ (-1)m(m!)xr+L is a formal solution of
m=o
2 dy
(E) dx+y-x=0
Step 2. It is easy to see that f (x) = el/=J 2t-1e-1l`dt (x > 0) satisfies the given
0
differential equation and has the asymptotic property f (x) = j (x) as x 0. Since
¢p(x) - f (x) is a solution of the homogeneous differential equation x2dy + y = 0,
it follows s that ¢p(x) = f (x) + cell=, where c is a constant. From this, op(x) _
elI1 t-le-ll`dt follows for argx = 2pir.
XIII-8. Show that the following differential equation has a nontrivial convergent
power series solution:
3
2x
d
+ dX+
2
y = Q.
EXERCISES XIII 447
Hint.
Step 1: The given differential equation has three linearly independent formal so-
lutions
+oo +- +oo
1(x) = ell= 1+ amxm 2(x) = 1+ L. bmxm, and o3(x) = x+ amxm.
M=1 m=2 m=3
In fact, Ol can be found through some calculation with the Newton polygon. The
other two can be found by solving the equations
f ao + 2a2 = 0, al + 603 = 0,
(l a,, + (m + 1)m(m - 1)0,,,+1 + (m + 2)(m + 1)am+2 = 0 for m > 2
+00
for a formal solution E a,,,x"`.
m=o
Step 2: The given differential equation has three linearly independent actual solu-
3ir
tions such that e-'/=O1(x) ^ as x 0 in the sector j argx + 7r1 <
T,
and 02(x) fi(x) and 03(x) ^_- 3(x) as x -. 0 in the sector I argxI < 32
Step 3: In the direction arg x = -7r,
02(2e2x,) 03(xe2ai)
for some constants c2 and C3. Then, c3 (x) - c2.3(x) is a convergent power series
solution of the given differential equation.
Remark:
(1) See [HIJ.
(2) This result was originally proved for a more general case in [Per1].
(3) There is another proof based on Exercises V-18 and V-19 (cf. [HSWI).
XIII-9. Consider a system of differential equations
du
xp+1
dx = F(x, y, u),
(E)
u = V, (X, y) _ yh+Gh(x),
h=0
where coefficients zlih(x) are R'-valued functions whose entries are holomorphic in
a neighborhood of x = 0. Assume also that O0(0) = 0 and det [(oo)]
o9p
54 0.
where p and q are positive integers, x and y are two complex independent variables,
iZ, F, and d are n-dimensional vectors (i.e., E C"), the entries of F and G are
holomorphic with respect to (x, y, 11) in a neighborhood of (x, y, u) _ (0, 0, and
system (S) is completely integrable, i.e., F and d satisfy the condition
yq+I
5F (x, y, v') + (x, y, u)G(x, y, u = xp+l (x, y, u) + 5 (x, y, U-)F(x, y, u).
041
det (0, 0, 0)J j4 0. Show that system (S) has one and only one solution
L
= yi+qxl+p 8211 aF OF -
0;-ax Y
1+q
ay - &a
G(x,y,ur)
= x1+p ad ac 011
- x
+paG - ad
ax + au ax ax su F'(x, y, u)
(xl+P
ax - F(x,y,i)
811
EXERCISES XIII 449
Using this result, it can be shown that (FS) also satisfies the system
Step 3. Upon applying Exercise XIII-9 to (E), the convergence of t%i(x, y) is proved.
XIII-11. Complete the proof of Theorem XIII-7-6 by verifying rigorously state-
ments (a), (b), and (c) in the proof.
XIII-12. Show that the series
(FS)
E
+00 (3)h xexp [_x3/2}
y = P(x) = x-1/4 1 + >( -1)h
[ 54hh!r(h +) 3
d2-xZ
is a formal solution of the differential equaton -xy = 0, where r is the Gamma-
function.
d2-xZ
XIII-13. Show that the differential equation - xy= 0 has a unique solution
O(x) such that (1) b(x) is entire in x and (2) ¢(x)exp [3x3/21 admits the formal
series p(x) exp [x3I2J as its asymptotic exansion as x -oc in the sector I arg xj <
r, where p(x) is givenby (FS).
Remark. Ai(x) = 2(--X) is called the Airy function (cf. [AS, p. 446), [Wasl, pp.
v/Fr
124-1261, and [01, pp. 392-394]).
XIII-14. Using the same notations in Exercises XIII-12 and XIII-13, show that if
w =xpel(2ril
3 J then ¢(w'ix) and Q(wx) are two solutions of equation (S). Also,
(i) derive asymptotic expansions of ¢(w-lx) and m(wx),
(ii) show that {Q(x),y5(w-lx)}, {0(w-1x),O(wx)}, and {d(wx),Q(x)} are three
fundamental sets of solutions of (S),
(iii) show that if we set m(x) = c1O(w-1x)+c2y5(wx), then c2 = -w and
[ct O
1
3
w 11J
is equal to the 2 x 2 identity matrix,
(iv) using (iii), show that cl = -w-1.
XIII-15. Show that if O(x, A) is an eigenfunction of the eigenvalue problem
(EP) +00
d2 y
da l La
then
( i) Q(x) is entire in x and Q(x) exp is a polynomial,
ITJ
450 XIII. SINGULARITIES OF THE SECOND KIND
(ii) all negative odd integers are eigenvalues of (EP) and there is no other eigen-
value,
(iii) for every non-negative integer n, Hn(x) = (-1)ne=2 Un (e-y2) is a polyno-
r 2
mial, and -0n(x) = H.(x)exp I - 2 is an eigenfunction of (EP) for the
J
eigenvalue -(2n + 1), L
r+00 +00
(iv) On(x)/m(x)dx = 0 if n 9k m, and / On(x)2dx = 2nn!y'.
J 0o J o00
Remark. The polynomials Hn(x) are called the Hermit polynomials (cf. [AS, p.
775] and [01, p. 49]).
XIII-16. Construct Green's function of the boundary-value problem
- x2y = f (x),
j o0
y(x)2dx < +00. Show also that
(i)
j +00 r+00
00 J 00
G(x,)2dxd< +oo,
n-1 h
XIII-17. Consider a differential operator C[y] _ dxn + 1: an-h(x) h , where
h=0
+00
aj(x) = ajmx-'n
E C[[x1]. Also, assume that aj,_m1 # 0 if a,(x) is not
m=-mj
equal to zero identically, while mj = -oo if aj(x) is zero identically. In the (X, Y)-
plane, consider the points Pj = (j, mj) (j = 1,... , n). Construct a convex polygon
II whose vertices are (0, 0), pi, p2, ... , p, such that each pk is one of those points
Pj, and that all other points Pj are situated below the polygon. Set po = (0, 0)
and Pk = (ak, Pk) (k = 1) ... , s), where an = n. Denote by pk the slope of the
segment k k+1 (k = 0, ... , s - 1). Then, Po > Pl > P2 > ... > p,-,. Assume
that pk>-1 (k<vo)and pk<-1 (k>vo). Show that
(i) the differential equation £[y] = 0 has n - at,,, linearly independent formal so-
Me
lutions of the form ye(x) = xe" E Req(x)(logx)q (e = a,,. + 1,... , n), where
q=o
Rrq(x) E C[[x-1]], b E C, and the Mt are non-negative integers,
(ii) if k < vo, then the differential equation £[y] = 0 has ak+1 - ak linearly
me
independent formal solutions of the form yr = e^<(x)x6, E Rrq(x)(log x)q (e =
q-0
EXERCISES XIII 451
H1
f12
and J_ _ [Oq]
01
where H =
J,, +A I...
Here, 0, is the nr x oo zero matrix and I,,,, is the oc x oo identity matrix. Let
Qo=So+No and A=S+N
be the S-N decomposition of 0o and that of A in the sense of §V-3. Also, let
Po, P1, P2, ... , Pm, ... be n x n constants matrices and let AO be an n x n diagonal
matrix such that det Po 0 0 and that
A 1 0 0 0 0
0 A2 0 0 0
0 0 0 0 An
and
(2) SP=PAo,
where
P1
P2
P0
P=
P.
Note that At are eigenvalues of So. Hence, A, are eigenvalues of 00. Show that
(a) the matrix S represents a multiplication operation: y -+ o(x)y for an n x n
matrix o(x) = So + O(x) whose entries are formal power series in x,
452 XIII. SINGULARITIES OF THE SECOND KIND
(b) the matrix N represents a differential operator Lo[yM = x4+1 fy + v(x)y for
an n x n matrix v(x) = No +O(x) whose entries are formal power series in x,
(c) C[f, = £o[yl + o(x)y and Go[o(x)yj = a(x)Go[yl,
+00
(d) if we set P(x) _ F xmP,,,, then P(x)-1o(x)P(x) = A0,
m=0
(e) if we set /C[uZ = P(x)-'C[P(x)uZ and Ko[u] = P(x)-1Go[P(x)ul, then K[u"j =
Ko[ul + Aou,
(f) if we set
Vii
du
Ko[i ] = xq+1 + vo(x)u, vo(x) = .
vn1
then
vjk(x) = 0 if .\.7 # Ak
(cf. [HKS]).
REFERENCES
453
454 REFERENCES
[HsS] P. F. Hsieh and Y. Sibuya, On the asymptotic integration of second order lin-
ear ordinary differential equations with polynomial coefficients, J. Math. Anal.
Appl., 16(1966), 84-103.
[HX1] P. F. Hsieh and F. Xie, Asymptotic diagonalization of a linear ordinary differ-
ential system, Kumamoto J. Math., 7(1994), 27-50.
[HX2] P. F. Hsieh and F. Xie, Asymptotic diagonalization of a system of linear ordinary
differential equations, Dynamics of Continu. Discre. Impula. Syst., 2(1996), 51-
74.
[HX3] P. F. Hsieh and F. Xie, On asymptotic diagonalization of linear ordinary differ-
ential equations, Dynamics of Continu. Discre. Impuls. Syst., 4(1998), 351-377.
(Huki] M. Hukuhara, Sur les systemes des equation differentielles ordinaires, Proc. Imp.
Acad. Japan, 4(1928), 448-449.
(Huk2] M. Hukuhara, Sur 1'ensemble des courbes integrals d'un systeme d'equations
differentielles ordinaires, Proc. Imp. Acad. Tokyo, 6(1930), 360-362; also [Huk8,
11-131.
[Huk3] M. Hukuhara, Sur les points singuliers des equations differentielles lineaires,
Domaine reel, J. Fac. Sci. Hokkaido Imp. Univ., 2(1934), 13-88; also (Huk8,
52-1301.
[Huk4) M. Hukuhara, Sur les points singuliers des equations differentielles lineaires, II,
J. Fac. Sci. Hokkaido Imp. Univ., 5(1937), 123-166; also [Huk8, 131-175].
(Huk5] M. Hukuhara, Sur un systeme de deux equations differentielles ordinaires non
lineaires h coefficients reels, Fac. Sci. Univ. Tokyo, 6(1951), 295-317.
(Huk6] M. Hukuhara, Sur la theorie des equations differentielles ordinaires, J. Fac. Sci.
Univ. Tokyo, Sec. 1, 7(1958), 483-510.
[Huk7] M. Hukuhara, Theoremes fondamentaux de la theorie des equations differentielles
ordinaires dans l'espace vectoriel topologique, J. Fac. Sci. Univ. Tokyo, Sec. I,
8(1959), 111-138.
(Huk8] M. Hukuhara, Selected Papers, Kyoritsu Shuppan Co., Ltd., Tokyo, 1997.
[HI] M. Hukuhara and M. Iwano, Etude de la convergence des solutions formelles
d'un systeme differentielles ordinaires lineaires, Funk. Ekv., 2(1959), 1-15; also
(Huk8, 423-437).
[HN1] M. Hukuhara and M. Nagumo, On a condition of stability for a differential
equation, Proc. Imp. Acad. Tokyo, 6(1930), 131-132; also (Na7, 136-1371.
[HN2] M. Hukuhara and M. Nagumo, Sur la stability des integrals d'un system
d'equations differentielles, Proc. Imp. Acad. Tokyo, 6(1930), 357-359; also
(Na7, 138-1401.
(HN3] M. Hukuhara and M. Nagumo, Un theoreme relatif h. 1'ensemble des courbes
intygrales d'un systeme differentielles ordinaires, Proc. Phys. -Math. Soc. Japan,
(3) 12(1930), 233-239; also [Na7, 141-147].
[Hum] J. E. Humphreys, Introduction to Lie Algebras and Representation Theory,
Springer-Verlag, New York, 1972.
[I] M. Iwano, Bounded solutions and stable domains of nonlinear ordinary differ-
ential equations, Analytic Theorey of Differential Equations, Lecture Notes in
Math. No. 183, P. F. Hsieh and A. W. J. Stoddart, eds., 59-127, Springer-
Verlag, New York, 1971.
REFERENCES 457
[Na6] M. Nagumo, Uber das Verhalten der Integrale von Ay" + f (x, y, y', A) = 0 fiir
A -+ 0, Proc. Phys.-Math. Soc. Japan, 21(1939), 529-534; also [Na7, 234-239].
[Na7) M. Nagumo, Mitio Nagumo Collected Papers, Springer-Verlag, New York, 1993.
[Nell F. Nevanlinna, Zur Theorie der Asymptotischen Potenzreihen, Ann. Acad. Sci.
Fennice Sarja, Ser. A, Nid. Tom. 12(1919), 1-81.
[Ne2J R. Nevanlinna, Analytic Functions, Springer-Verlag, New York, 1970.
[NU] A. F. Nikiforov and V. B. Uvarov, Special Functions of Mathematical Physics,
Birkhauser, Boston, 1988.
[O1] F. W. J. Olver, Asymptotics and Special Functions, Academic Press, New York,
1974.
REFERENCES 459
[Sat] Y. Sato, Periodic solutions of the equation x" + f (x)x' + g(x) = 0 without the
hypothesis xg(x) > 0 for [xI > 0, Comment. Math. Univ. St. Pauli, 20(1971),
127-145.
(Sc] L. Schwartz, Theorie des Distributions, 3rd ed., Hermann, Paris, 1966.
[Sil] Y. Sibuya, Sur un systcme d'6quations differentielles ordinaires lnnaires h coef-
ficients periodiques et contenants des parami tres, J. Fac. Sci. Univ. Tokyo, I,
7(1954), 229-241.
[Si2] Y. Sibuya, Nonlinear ordinary differential equations with periodic coefficients,
Punk. Ekv., 1(1958), 77-132.
[Si3] Y. Sibuya, Sur reduction analytique d'un systeme d'i quations differentielles or-
dinaires lnnaires contenant un parametre, J. Fac. Sci. Univ. Tokyo, 7(1958),
527-540.
[Si4] Y. Sibuya, Notes on real matrices and linear dynamical systems with periodic
coefficients, J. Math. Anal. Appl., 1(1960), 363-372.
[Si5] Y. Sibuya, On perturbations of discontinuous solutions of ordinary differential
equations, Natural Sci. Rep. Ochanomizu Univ., 11(1960), 1-18.
[Si6] Y. Sibuya, On nonlinear ordinary differential equations containing a parameter,
J. Math. Mech., 9(1960), 369-398.
[Si7] Y. Sibuya, Simplification of a system of linear ordinary differential equations
about a singular point, Funk. Ekv., 4(1962), 29-56.
[Si8] Y. Sibuya, On the convergence of formal solutions of systems of linear ordinary
differential equations containing a parameter, MRC Tech. Sum. Report, No.
511, 1964.
(Si9] Y. Sibuya, A block-diagonalization theorem for systems of linear ordinary differ-
ential equations and its applications, J. SIAM Anal. Math., 14(1966), 468--475.
[Si10] Y. Sibuya, Perturbation of linear ordinary differential equations at irregular sin-
gular points, Funk. Ekv., 11(1968), 235-246.
[Sill] Y. Sibuya, Perturbation at an irregular singular point, Japan-U. S. Seminar on
Ordinary Differential and Functional Equations, Lecture Notes in Math. No.
243, M. Urabe, ed., Springer-Verlag, New York, 1971, pp. 148-168.
(Si12] Y. Sibuya, Uniform Simplification in a Full Neighborhood of a Thansition Point,
Memoirs of American Mathematical Society, No. 149, Providence, RI, 1974.
(Sil3] Y. Sibuya, Global Theory of a Second Order Linear Ordinary Differential Equa-
tion with a Polynomial Coefficient, Math. Studies 18, North-Holland, Amster-
dam, 1975.
[Si14] Y. Sibuya, Convergence of formal power series solutions of a system of nonlinear
differential equations at an irregular singular point, Lecture Notes in Math. No.
810, Springer-Verlag, New York, 1980, pp. 135-142.
[Si15] Y. Sibuya, A theorem concerning uniform simplification at a transition point
and the problem of resonance, SIAM J. Math. Anal., 12 (1981), 653-668.
[Si16] Y. Sibuya, Differential modules and theorem of Hukuhara-Turrittin, Proc. of
International Conf. on Nonlinear Phenomena in Math. Sci., Academic Press,
New York,1982, pp. 927-937.
[Si17] Y. Sibuya, Linear Differential Equations in the Complex Domain: Problems of
Analytic Continuation, Translations of Math. Monographs 82, American Math-
ematical Society, Providence, RI, 1990.
REFERENCES 461
462
INDEX 463
nonhomogeneous equation, 96 Levinson, N, 14, 28, 36. 65, 69, 108, 137,
nonlinear, 1, 16, 32, 39 138, 144, 153, 162, 191, 197, 225,
second order equation, 324, 333 233, 246, 266, 274, 281, 290, 293,
Inner product, 151 298, 304, 340, 341
Instability, 243 Levinson's diagonalization theorem, 21:3
Instability region, 189 Levitan, B. M., 181
Integral inequality, 377, 418 Levitan, Gel'fand-, integral equation, 181
Invariant set, 280, 299 Liapounoff, A., 197
Irregular singular point, 134, 441 Liapounoff function, 239, 309, 311
Iwano, M., 447 Liapounoff's direct method, 281
Iwasaki, 1., 138 Liapounoff's type number, 198
calculation of, 203
J, map of A,(r,a,b) to C[[x]] 353 multiplicities, 202
one-to-one, 359 of a function, 198
onto, 356 of a system at t = oc, 201, 204
Jacobson, N., 69 of a solution, 199
Jordan-Chevalley decomposition, 69 properties, 198
Jordan canonical form, 421, 426 Lie algebra, 90
Jordan curve, index of, 293 Limit cycle, 292
Jost solution, 168, 180, 194 Limit invariant set, 280
Jurkat, W., 136, 391, 401, 402, 435 Lin, C.-H., 370
Lindelbf, E., 1, 28, 359
K, the field of fractions of C[[x[], 424 Lipschitz condition, 3, 43, 64
K[u'J, differential operator, 123 constant, 3
Kaplan, J., 197 sufficient condition for, 5
Kato, J., 44, 66 existence without, 8
Kato, T., 103 Local blowup of solutions, 17
Kimura, II., 138 Logarithm of a matrix, 86
Kimura, T., 197 log[ 1 + ltf ], 86, 88
Kneser, H., theorem, 41, 47, 305 log[8(w)[, 88
Kohno, M., 78, 108, 130, 452 log[8(w)2], 88, 94
Komatsu, H., 8 log[s], 86, 88
log[S2], 88
C, differential operator, 120, 151, 169, Lutz, D. A., 136, 197, 391, 401, 402, 435
424, 437, 443
normal form, 121 M,,(C), set of all n x n matrices
calculation of, 130 in C, 69
C+, limit-invariant set, 280 MacDonald, 1. C., 109
Laplace transform, inverse, 103 Magnus, W., 196
LaSalle, J., 279, 281 Mahler, K., 112
Lebesgue-integrable function, 15 Maillet, E., 112, 353, 441
Lee, E. B., 68, 106 Malgrange, B., 438
Lefschetz, S., 279, 281 Manifold, 45
Legendre, equation, 141 differential equation on, 45
polynomials, 141, 192 stable, 243
Leroy transform, incomplete, 354 unstable, 246
Lettenmeyer, F., 142 Markus, L., 68, 69, 96, 106
Levelt, A. H. M., 108, 125, 435 Matrix, 69
Levin, J. J., 340 functions of a matrix f (A), 84
466 INDEX
468
Universitext (continued )
ISBN 0-387-98699-S
www.springer-ny.com 9 78038709869990>