4 Periodic Coefficients 1997

Downloaded 03/13/13 to 141.213.236.110. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.
php
Chapter 4
Periodic Coefficients
4.1 Introduction
Human fascination with periodic phenomena governed by differential equa-
tions, such as the periodic waxing and waning of the moon, extends back into
prehistory. In this chapter we consider linear systems of differential equations
whose coefficients are periodic functions. This class of equations is much larger
than the systems with constant coefficients, and in general explicit descriptions
of the solutions will not be obtained. Nonetheless, the periodicity does impose
some structure on solutions, a structure which can be described using a bit of
linear algebra.
Before launching into the theoretical development, we describe two models
which use linear systems with periodic coefficients. The first model is a modi-
fication of the toxic discharge problem in Chapter 1 that is pictured in Figure
1.1 and led to equations (1.1) and (1.2). That case considered the build up of
toxic material in a system of lakes after a discharge of toxic material upstream.
In the modified model some of the water flowing in the channel between
Lakes 1 and 2 is removed from the system for crop irrigation. The rate at which
water is removed is not constant. Each morning at the same time a valve is
opened, and water is drained from the channel. The valve is closed later, again
at the same time each day. Thus the rate at which water is removed from the
channel is described by a function w(t) which is periodic in time. If time is
measured in days, then w(t + 1) = w(t).
When the derivation of the equations describing the lake contamination is
reexamined, we observe first that the behavior in Lake 1 is unchanged:
ii
r(t) l Ti.
dt v
However, the rate i2 at which water enters Lake 2 is no longer constant but
may now be represented by Li — w(t), where ll is the constant rate at which
93
94 Linear Ordinary Differential Equations
water flows from Lake 1 into the channel. The second equation thus becomes
Downloaded 03/13/13 to 141.213.236.110. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.php
dr2 Li — w(t) 02
V T2.
dt v 1 T1 2
In the matrix formulation the nonhomogeneous system of equations has as-
sumed the form
X' = A(t)X + B(t),
where A(t) is no longer constant but is periodic with period 1,
A(t + 1) = A(t).
Since the variation of parameters formula reduces the problem of solving a
nonhomogeneous system to that of solving the associated homogeneous system,
the fundamental problem is the analysis of
X' = A(t)X, A(t +1) = A(t).

A simple but essential observation initiates the study of such systems: shifting
the solutions of the homogeneous system by a period is a linear transformation
on a finite-dimensional vector space. Linear algebra thus offers a key with
which to open our investigation.
Periodicity can also enter linear systems through the coefficient B(t). A
mechanical example, the mass and spring system considered in section 1.3,
leads to the equation
mx"+cx'+kx= f(t).
Recall that in this model the coefficients m, c, and k are constant functions,
which are thus periodic with every period. When the external forcing function
f (t) is periodic,
f (t + w) = f (t), w > 0,
the whole equation is periodic with period w. A natural question in this setting
is whether the equation has solutions with period w. This problem is taken up
after consideration of the homogeneous system with periodic coefficients.
4.2 Floquet's theorem

Consider the system
X' = A(t)X, (4.1)
where A E C(R, M(C)), and A is periodic with period w > 0,
A(t + w) = A(t), t E R.
A basic observation about such a system is that if X(t) satisfies (4.1), then so
does Y(t) = X(t + w), since
Y'(t) = X'(t + w) = A(t + w)X (t + w) = A(t)Y(t), t E R.

Periodic Coefficients 95
Thus if X is a basis for the solutions of (4.1) then so is Y, where Y(t) _

X (t +w). This fact leads to a fundamental representation result for a basis X.

Theorem 4.1 (Floquet): Each basis X for the solutions of (4.1) can be
represented as
X(t) = P(t)e Rt , t E R, (4.2)
where P E C'(R, Mn (C)), P(t) is invertible, P(t + w) = P(t) for all t E R,
and RE M,,(C).
Proof: Since Y(t) = X(t + w) gives a basis for the solutions of (4.1),
X(t + w) = X(t)C for some invertible C E Mn (C). Theorem 4.2 below shows
that for every invertible C there exists a matrix R E Mn (C) such that e R = C;
`'
that is, Rw is a logarithm of C. It follows that
X(t + w) = X(t)e', t E R. (4.3)
If P(t) = X(t)e -Rt we see that for all t E R,

X(t)e —Rt = P(t).
P(t + w) = X(t + w)e —Rat+w) = X(t)e RW e —Rat+-) =
Since both X(t) and e -Rt are invertible, so is P(t), and P E C'(R, Mn (C)). ❑
Since a basis X of solutions for (4.1) satisfies X' = AX, substitution of
X(t) = P(t)e Rt leads to
P'(t)e Rt + P(t)Re R t = A(t)P(t)e Rt

'
or P' + PR = AP. Putting X = PY in (4.1) then gives
PY' + P'Y = APY = P'Y + PRY,
so that
Y'=RY. (4.4)
Thus the substitution X = PY transforms the system (4.1) with a periodic
coefficient matrix A to the constant coefficient system (4.4). In fact the map
Y —* X = PY is an isomorphism between the set of solutions of (4.4) and the
set S of solutions of (4.1).
4.3 The logarithm of an invertible matrix

A matrix B E M(C) such that eB = C is said to be a logarithm of C.
Theorem 4.1 made use of the fact that every invertible matrix C E M(C)
has a logarithm, which is the next result. The proof makes use of some power
series techniques which are discussed more fully in Chapter 5.
Theorem 4.2: If C E M(C) is invertible, there exists a B E M,,(C) such
that eB = C.
Proof: Let J = Q -1 CQ be a Jordan canonical form for C. If there is a
K such that e K = J, then it will follow from the power series definition of the
matrix exponential (section 3.1) that C = QJQ -1 = e B , where B = QKQ -1 .

A further reduction is possible because the matrix J has block diagonal form,
J = diag(Ji, ... ,J 4 ). If K3 satisfies e K i = J 3 and if K = diag(Ki, ... ,K),
then
eK = diag(e K ., ... , e K q) = diag(J1, ... J q ) = J.
Since C is invertible, none of its eigenvalues is 0. Therefore it suffices
to find a K such that e K = J when J is the 1 x 1 matrix
(i) J = (µ), z0
or when J is the r x r matrix

µ 1 0 ... 0 0
0 µ 1 ... 0 0
(ii) J= =yIr+N, r> 1, It 0.
0 0 0 ... p 1
0 0 0 ... 0 p
In case (i) choose K to be any value of log(µ). For example, if y = ^ µ l e ie

—ir < 0 < 7r, then we choose log(µ) = log(i) + iO, the principle value of
log(µ).
In case (ii), write
J = µ(I r + N/p),
and note that (N/µ)T = 0. We will find a matrix l (N/p) such that
= I T + N/µ, (4.5)
and then the matrix

K = log(µ)I r + l(N/p)
will be such that e K = J. The determination of l(N/µ) is based on the power
series
1(z) = log(1 + z) = z — z 2 /2 + z 3 /3 — ... _

k=1
which is absolutely convergent for Izi < 1, z E C. Since
1+z=exp(l(z))=1+(z—z 2 /2+...)+ 1 (z—z 2 /2+•••) 2 +•••, (4.6)
the rearrangement of the terms on the right obtained by collecting ascending

powers of z must agree with 1 + z; that is, the coefficients of the first two
powers of z sum to 1, while the remainder sum to zero.
Since (N/µ)T = 0, the sum
l(N/µ) = (N/µ) — (N/µ) 2 / 2 + (N/u) 3 / 3 — ... _ E(-1)k+l(N/µ)k/k

k=1
is a finite sum, hence is convergent. Now compute e l( ' ) as in (4.6), with the
same rearrangement of terms. Since the coefficients after rearrangement agree

with those in (4.6), (4.5) is obtained. ❑
Equation (4.3) shows that
C = e = X - 1 (0)X(w),
so R can be computed once we know X at 0 and w. Also the periodicity of

P means it is uniquely determined by its value on [0, w] . Thus a basis X is
determined on all of R by (4.2) once it is known on [0, w].
4.4 Multipliers
The matrices C and R depend on the basis X. If Xi is another basis for S,
the set of solutions of (4.1), then X 1 = XT for some invertible T E M(C),
and if
X1(t+w) =Xl(t)C1, tER,
then
C1 = X1'(0)Xl(w) = T — 'X -1 (0)X(w)T = T -1 CT
so that C1 is similar to C. Thus the eigenvalues of C, together with their mul-
tiplicities, depend only on A(t) and are independent of the basis chosen. The
eigenvalues of C are called the multipliers of (4.1). The name is suggested
by the following property of the multipliers.
Theorem 4.3: The system (4.1) has a nontrivial solution X satisfying
X(t + w) = pX (t), t E R, (4.7)
if and only if it is a multiplier for (4.1). In particular, (4.1) has a nontrivial

solution of period w if and only if u = 1 is a multiplier for (4.1).
Proof: If X is a basis for (4.1), then the nontrivial solution X = X^, 0,
satisfies (4.7) if and only if
X(t + w) = X(t + w) = x(t)Ce = µ x(t) = px(t) ,
or C y^. ❑
Now let X be the basis for (4.1) satisfying X(0) = I. Then C = e R``' _
X(w), and the multipliers for (4.1) are the eigenvalues of X(w). Let S(µ) be
the set of all X E S satisfying (4.7). The structure of S(µ) is given in the
following result.
Theorem 4.4: For each u E C,
(i) S(µ) is a vector subspace of S,
(ii) dimS(p) = dimE(X(w), p) = dimN(iI n — X(w)).
Proof: If X, Y E S(µ) and a, 0 E C, then aX + /3Y E S and
(aX +,3Y)(t + w) = aX(t + w) + 3Y(t + w)

= caµX (t) + O/tY(t) = µ(c X + /3Y) (t), t E R,

so that ozX + 13Y E S(µ), proving (i). As to (ii), we note that the proof of
Theorem 4.3 shows that the map
S: £(X(w),µ) —fX=X^
takes the eigenspace of X(w) for µ onto S(tc). It is clearly linear, for a, /3 E C
and ,'q E 9(X (w), p),
S(o< + 077) = X( + /3i) = aX + ,3Xi7 = aS(^) + /S(]),
and it is one to one since X is invertible; that is, S(^) = S(r^), or X = Xi7, if
and only if = rt. Thus S is an isomorphism of £(X(w), µ) onto S(µ), and (ii)
follows, since two finite-dimensional vector spaces are isomorphic if and only
if they have the same dimension. ❑
If µl, ... , p n is an enumeration of all the multipliers, where each multiplier
is repeated as many times as its multiplicity, then (see Chapter 3, exercise
39(a))
µ,...µn=det(X(w)).
Since the Wronskian WX = det(X) satisfies (Theorem 2.6)
det(X(w)) = det(X(0)) exp VOW tr(A(s)) ds

] ,
it follows that
PI ... An = exp tr(A(s)) ds (4.8)
[fo1 ]
In particular, none of the ,i is zero, and if n — 1 of them are known, the

remaining one can be obtained from (4.8).
4.5 The behavior of solutions for large t

We continue using the basis X for (4.1) such that X(0) = I. The eigenvalues
of a matrix R such that X(w) = e Rw are called the characteristic exponents
for (4.1). Let J be a Jordan canonical form for R, so that J = Q — 'RQ for
some invertible Q and
J = diag(J1,...,J q ),
with the J; being r2 x ri, Ji = (.\i) if ri = 1, and
Ai 1 0 ... 0 0
0 Ai 1 ... 0 0
Ji= ri>1.
0 0 0 ... Ai 1
0 0 0 ... 0 Ai
Then X(w) = e R, = exp(QJQ -1 w) = Q exp(Jw)Q -1 , and since

exp(Jw) = diag(exp(J i w), ... , exp(Jw)),
the eigenvalues of X(w), that is the multipliers for (4.1), are the same as the
eigenvalues of exp(Jw), which are given by exp(A 3 w). Thus
µj=exp(\jw), j=1,...,n,
for some enumeration of the ,u and A. If Rl is another matrix such that

exp(R l w) = X(w), with eigenvalues vj, then
pj = exp(vjw), j = 1, ... , n,
and it follows that

A^ = v3 + (2irik)/w
for some integer k. Thus the characteristic exponents are not uniquely deter-
mined by (4.1), but their real parts are since
µ^ I = exp(Re(Aj)w), Re(A) = W logI i I, j = 1,... ,n.
From the representation (4.2) for X and a Jordan canonical form J =

Q -1 RQ for R, we can obtain a basis X1, where
X1 ( t) = X (t ) Q = Pi(t)eJt,
Pi(t) = P(t)Q, Pi(t+w) = Pi(t).
From the explicit structure of e Jt as given in Theorem 3.10, we see that the
columns Xi,. . . , X n of X1 have the form
X^ (t) = e \t P(t),
where A is an eigenvalue of R and P is a vector polynomial with coefficients

which are periodic functions of period w. For example, if P 1 = (PI , ... , P,)
and the first block of J is the r l x r l matrix
Al 1 0 ... 0 0
0 A l 1 ... 0 0
Ji = ri > 1 ,
0 0 0 ... A l1
0 0 0 ... 0 Al
then
Xi (t) = exp(A1t)Pi (t),
X2(t) = exp(Ait)[tPi(t) + P2(t)],

2
X3 (t) = exp(Alt) 2, P1(t) + tP2 (t) + P3(t)
tri -1
Xrl (t) = exp(Ait) (rj — 1)! Pi (t) + ... + Pry (t) ].
From this representation of Xi (t) = Pi(t)e Jt we have the following ana-

logue of Theorem 3.11. The proof is entirely similar once we note that 1µi
exp(Re(7^)w).
Theorem 4.5: Let µl, ... , µk be the distinct multipliers for (4.1), with
multiplicities m1,. . . , mk, respectively. Then
(i) all solutions of (4.1) are bounded on [0, oo) if and only if µj < 1, j =
1,... , k, and for those t, such that ' pj = 1 we have m = dim(t'(X(w), p ));
(ii) all solutions of (4.1) tend to zero as t — oo if and only if pj ^ < 1,
j = 1,...,k.
4.6 First-order nonhomogeneous systems

Consider a nonhomogeneous system
X' = A(t)X + B(t), (4.9)
where A E C(7Z, M,,(C)), B E C(R, Cn), and A,B are periodic with period
w>0,
A(t + w) = A(t), B(t + w) = B(t), t E R.
It is interesting to determine when (4.9) has periodic solutions of period w,
and what the structure of the set of all such solutions is. Let Sp be the set of
all solutions of X' = A(t)X which are periodic with period w, and let Sp (B)
denote the set of all solutions of (4.9) which have period w.
Theorem 4.6: The set Sp (B) is an affine space,
Sp(B) = V + Sp = {X JX = V + U, U ES},
where V is a particular solution in SP (B).

Proof: If V E Sp (B) and U E SP , then X = V + U is a solution of (4.9) and
it clearly is periodic of period w, so that X E Sp (B). Thus V + Sp C Sp (B).
Conversely, if X, V E SP (B), then U = X — V satisfies X' = A(t)X and has
period w. Hence U E SP , and we have Sp (B) C V + S. ❑
As an example, consider the equation of period w = 27r,
x' _ — cos(t)x + cos(t),
with n = 1, A(t) _ — cos(t), B(t) = cos(t). The solution x satisfying x(0) _

is given by
x(t) = 1 — exp(— sin(t)) + exp(— sin(t)), E C,

and this is periodic of period 27r for every E C. In this case x = v + u, where
v(t) = 1 - exp(- sin(t)), u(t) = exp(- sin(t))e.
On the other hand, consider
x = - cos(t)x + 1,
with A(t) _ - cos(t), B(t) = 1, having period 2ir. Here the solution x such
that x(0) _ is given by
t
x(t) = exp(- sin(t))
fo exp(sin(s)) ds + exp(- sin(t)), E C,
and
27r
x(2ir) - x(0) =
fo exp(sin(s)) ds > 0.
Therefore this equation has no solution of period 21r for any E C.

Thus a central problem is to determine when Sp (B) is nonempty. Just as
in the homogeneous case, if X is a solution of (4.9) then so is Y(t) = X(t+w),
for
Y'(t) = X'(t + w) = A(t + w)X(t + w) + B(t + w) = A(t)Y(t) + B(t).
Thus a solution X of (4.9) has period w if and only if X(t) = Y(t) = X(t + w)
for all t E R, and this is true if and only if X(0) = X(w). If X is the basis for
the solutions of X' = A(t)X such that X(0) = I,,,, the variation of parameters
formula gives the solution X of (4.9) satisfying X(0) = Z; as
X(t) = X(t) + X(t) f t X -1 (s)B(s) ds, t E R.

Thus X(0) = X(w), that is, X E Sp (B), if and only if
[I n -X(w)] = X(w) X -1 (s)B(s) ds. (4.10)

JOLO
If I,, - X(w) is invertible, then clearly
ds
_ [I _ X(w]-1X(w)
J0
W X - 1 (s)B(s)
is the unique = X(0) which yields a periodic solution X of (4.9) of period

w. Now I,,, - X(w) is invertible if and only if µ = 1 is not a multiplier for
X' = A(t)X, that is, the homogeneous system has no nontrivial solution of
period w. We have shown the following result.
Theorem 4.7: There exists an X E Sp (B) if and only if there is a E C"''
satisfying (4.10). If dim(Sp ) = 0 there exists a unique X E Sp (B) for every
BE C(R, Cn) of period w.
The analysis of the case when dim(Sp ) > 0 depends on a result for linear
equations
T^=rl, TE M(C), ,rlEC n .
Note that (4.10) has this form, where
T = I n — X(w), rl = X(w) —' (s)B(s) ds.

fow X
This result relates the solvability of T = rl to the solutions ( of the adjoint ho-
mogeneous equation T*c = 0. Recall that the adjoint T* of T is the conjugate
transpose of T = (tip), so that T* _ (t^2).
The inner product (^, rj) of , r, E Cn is given by (^, rl) = rl*^. If S C Cn, its
orthogonal complement S 1 is the set
S 1 = {^ E Cn (, ^) = 71*e = 0, ES}.
The set S I is a vector subspace of Cn, and if S is a subspace of Cn, then
C n = S ®S 1 , (4.11)
a direct sum, so that every E Cn can be written uniquely as + rl, where

^ E S, 77 E S 1 • Moreover, we have (S')' = S. A consequence of (4.11) is that
if S is a subspace of Cn, then
dim(S) + dim(S 1 ) = n. (4.12)
Theorem 4.8: If T E Mn (C), then
(i) Ran(T) = (N(T*)) 1 ,
(ii) dim(N(T)) = dim(N(T*)).
Proof: For every , c E C we have
(Te, ^) = (*T = (T*()* = (e, T*).
From this it follows that (E (Ran(T))', that is, (T^, () = 0 for all E Cn
if and only if (^, T*O = 0 for all E C n or T*( = 0, that is, E N(T*).
Thus (Ran(T))' = N(T*), and hence Ran(T) = ((Ran(T))')' = (N(T*)) 1 ,
which is (i).
From (i) and (4.12), with S = N(T*), we have
dim(N(T*)) + dim(Ran(T)) = n.
But we know that
dim(N(T)) + dim(Ran(T)) = n,
and therefore (ii) is valid. ❑

Applying Theorem 4.8 (i) to T = I n — X(w) in (4.10), we see that there is

an X E Sp ( B) if and only if
(*X(w) f W X- '( s)B(s) ds = 0 (4.13)
for every (E C"' such that
[I , — X *(w)]c = 0.
n (4.14)
This condition can be interpreted in terms of the solutions S* of the adjoint

homogeneous differential equation
Y' = — A* (t)Y. (4.15)
The basis Y for S* such that Y(0) = I n is just Y = (X*) -1 where X is the ,
basis for the solutions S of X' = A(t)X satisfying X(0) = I. This follows
since (see Chapter 2, exercise 13)
Y' = _(X*)-1(X*)'(X*) -1 = —( X*) - 1 X*A*(X*) -1 = —A *Y,
Y(0) = (X * ) - 1 (0) = I.
Let Sp denote the set of all Y E $* which have period w. The equality
I n —X*(w) = — X *(w)[I — Y(w)] (4.16)
shows that (E Cn satisfies (4.14) if and only if
[In — Y(w)]( = 0 ,
and this is true if and only if Y = Y( E S. Now X*(w)(_ implies

(*X(w) = (*, and then (4.13) becomes
(* J X -1 (s)B(s) ds = fow Y*(s)B(s) ds = 0
for every Y = Y( E S. Moreover, the equality (4.16) and Theorem 4.8 applied
to I n — X(w) imply that
dim(SS) = dim(N(I n — Y(w))) = dim(N(I n — X*(w)))
= dim(N(I n — X(w))) = dim(Sp ).

We have now proved the following theorem.
Theorem 4.9: Given B E C(R, Cn) which is periodic of period w, there
exists an X E Sp ( B) if and only if
JO LO
Y *(s)B(s) ds = 0
for every Y E S. We have

dim(SS) = dim(SP ),
and hence there is a unique X E Sp (B) if and only if dim(Sp ) = 0.

As an application of Theorem 4.9 we have the following interesting result.
Theorem 4.10: There exists an X E Sp (B) if and only if a solution X of
(4.9) which is bounded on [0, oo) exists. Equivalently, either S(B) is nonempty
or every solution of (4.9) is unbounded on [0, oo).
Proof: Clearly, if X E Sp (B), then X is bounded on R, and hence on
[0, oc). Conversely, suppose there is a solution X of (4.9) which is bounded by
M>0,
X (t) I < M, t E [0, oc). (4.17)
The variation of parameters formula gives
X(t) = X(t)X(0) + X(t) f t X -1 (s)B(s) ds, t E
where X is the basis for X' = A(t)X such that X(0) = I. Thus
X(w) = X(w)X(0) + n, 77 = X(w)

fo w
X -1 (s)B(s) ds.
Since Y(t) = X(t + w) is also a solution of (4.9), we have
Y(w) = X(w)Y(0) + ri
or
X(2w) = X(w)X(w) + T, = X 2 (w)X(o) + [ In + X(w)]j•
A simple induction then shows that for each k = 1, 2,...
X(kw) = X k (w)X(0) + [I,,, + X(w) + • • • + X k-1 (w)]ri. (4.18)
Suppose SS (B) _ 0. Then there exists a E Cn satisfying (4.14), ( = X*(w)(

such that (4.13) is not valid, that is, (*r 0. This is equivalent to the existence
of a Y E such that
Y*(s)B(s) ds 0 0.
J
Now (= X*(w)( implies
(* _ (* X 2 (w) _*X k (w), k = 1, 2, ... ,

and multiplying (4.18) by (* results in
(*X(kw) _ (*X(0) + k(* i , k =1, 2, ... .
From (4.17) it follows that
kI( *71I = ( * X(kw) — ( * X( 0 )I < 2 MI(I, k=1,2,...,

but this is clearly false since (*1])4 0. Hence Sp (B) is not empty. ❑
4.7 Second-order homogeneous equations

Important applications of the results in sections 4.1-4.4 occur in the case of

second-order equations. For example, the Mathieu equation
x" + (a + bcos(2t))x = 0,
where a, b E R, arises in the study of wave motion associated with elliptic

membranes and elliptic cylinders. The more general Hill's equation
x" + a(t)x = 0,
where a(t) is real valued and a(t+w) = a(t) for some w > 0, arises in the study
of the motion of the moon. In that case a(t) is an even function of period 'ir
and is often written as a series
a(t) = bo + bk cos(2kt).
k=1
Let us first interpret our results in the case of the equation
x" + al(t)x" + ao(t)x = 0, a 3 E C(1Z, C), (4.19)
a3 (t + w) = a^ (t), j =0,1,
for some w > 0. The system

y' = A(t)y (4.20)
associated with (4.19) has a coefficient matrix A given by
A=( ° —a l I
and X = (x1, x2) is a basis for (4.19) if and only if
X = (xl ^ x2) i i
- \ xlx2
x l x2
is a basis for (4.20). Let X be the basis for (4.19) such that X(0) = I2. Then
the multipliers for (4.20) are the eigenvalues µl, µ2 of X(w), that is, the µ
satisfying
det[PI2 — X(w)] = µ 2 — 2aµ + b = 0,
2a = tr(X(w)) = x1(w) + x2(w) = Al + µ2,

b = det(X(w)) = µ1µ2•
Note that the Wronskian WX = det(X) satisfies
( W
_fal(s)
det(X(w)) = exp ds = µ1µ2,
106
106 Linear Ordinary
Linear Ordinary Differential
Differential Equations
Equations
that ififthe
so that theaverage
average of
of a1
al over
over aa period
period isis zero,
zero,
11Jr
1
-
ww o0
—
WW
0,
al(s) ds == 0, (4.21)
(4.21)
then det(X(w)) = µ1µ2

det(X(w)) = /-ll/.J,2 = 1, and the
= 1, the equation
equation for
for the /-lj becomes
µ/-l22 — 2aµ++11== 0,
- 2a/-l XI(W)++x2(w).
0, 2a2a==x1(w) x;(w). (4.22)
Let us
Let us assume
assume (4.21),
(4.21),sosothat µ2 == 1//-l1,
that /-l2 and also
1/pi, and also assume the aaj3 are
assume that the are
real valued.
valued. Then X X real
is real and a in (4.22) is real. If Im(/-ll) =I- 0,
in (4.22) is real. If Im(µl) 0 0, then µ2
/-l2 µl
= III
l/-lll == I/1/-l21
and piI
and 2 2I = Re(/-ld ==Re(µ2)
= 1, Re(pi) Re(/-l2) ==a.a.Solving
Solving(4.22),
(4.22),we
wemay
maychoose
choose
/-ll = a +~,
p1=a+ /-l2 = a - a
a2-1, /2=a— a 22 -1.
- 1. vi
There are cases: (i) a 22 >
are three cases: > 1, (ii) aa22 << 1,
1, (ii) (iii) a 22 =
1, (iii) = 1.
1.
case (i),
In case (i), /-ll and µ2
µl and real and
/-l2 are real and distinct
distinct and µ2 =
and /-l2 = 1//-ll' Thereisisaa basis
1/µl. There basis
U = (ul,
(UI, u2)
U2) for solutions of
for the solutions of (4.19)
(4.19) having the the form
form
Atp2(t),
u1(t) =
UI(t) = eeAtpI(t), U2(t)
At pl (t), u2 = ee-—At
(t) = p2 (t), (4.23)
AW
eeAw=,
= p^
/-ll, (tPj(t
+ w)+ w)==p^ = 1,2.
(t), jj=1,2.
Pj(t),
If a> 0 wehave
Ifa>0we 0 < /-l2 < 1 < /-ll sothat
have 0</12<1</11so -Re(A) < 0 < Re(A).Now
that—Re(.\)<0<Re(A). Now
—Re (n)t (t)
I ul (t) I = e Re(A)t Ip1(t) I, Iu2 (t) I = e P2
and, since the Pj

pj are
are not
not identically
identicallyzero,
zero,there Tj EE [0, w]
areare
there w]such
suchthat
that Pj (Tj )
pj (Tj) i=
O.
0. Then
Re(A)— —Re(a)mw
Iu1 (rrl+mw )I = Iu1( 71)Ie Iu2(T2+mw)I = Iu2(T2)Ie
for
for every
every integer
integer m.
m. Thus
Iui(rl + mw)I -* oo, m -p oo, and Iu2(r2 +mw)I —* 00, m --^ —00,
and
and so
so UI is unbounded
ul is unbounded as
as tt ---+
— +00, whereas U2
+oo, whereas is unbounded as
u2 is as t t—j
---+ —oo.
-00.
Also
Also we
we have
have
IUI(t)l---+
ul (t) I —j0,0, tt ---+
—* -00, and
—oo, u2 (t) I —*
and IU2(t)l---+ 0, 0, tt ---+ +oo.
— +00.
If aa <
If < 00 we have00<</-ll
wehave < 11 </12
ti < < /-l2 and
and the
the roles
roles of U1,U2
of UI, are reversed.
u2 are reversed. Thus
every
every nontrivial
nontrivial solution
solution xx = CIUI ++ c2u2
= clul of (4.19),
C2U2 of (4.19), where
where CI,
cl, C2
c2are
are not
not both
0,
0, is
is unbounded
unbounded on on R.
In case
In case (ii),
(ii), /-ll and µ2
µl and are distinct
/-l2 are distinct nonreal
nonreal complex
complex numbers
numbers of of magnitude
one,
one,
µ1 = e i0w µ2 = e—i°w 0 E R, 0 ^ k7r.
There is now a basis U = (ul, u2) of the form (4.23) above, where we can take
A = iB. Since
Iu^(t)I = Ip(t)I, j = 1, 2,
all solutions x of (4.19) are bounded on 7Z. In fact U = (ul, u2) is bounded on
R.
For case (iii) there is one real multiplier p, = a = ±1 of multiplicity 2. If
p, = 1 there exists a nontrivial solution u l such that u l (t + w) = u l (t) for
all t E R, so that u l is periodic of period w. A second basis element u2 may
or may not be periodic of period w. We have u 2 of period w if and only if
U(t + w) = U(t), t E R, and this is true if and only if U(w) = U(0), or
X(w) = U(w)U -1 (0) = I 2 .
If pj = —1 there is a nontrivial solution ul satisfying ill(t + w) = —ul(t) for

all t E R, and then ul (t + 2w) = —ul (t + w) = ul (t), so that ul has period
2w. A second basis element U2 satisfies u2(t + w) = —u2(t), if and only if
U(w) = —U(0), or
-1
X(w) = U(w)U (0) = —I2.
In this case all solutions have period 2w.
4.8 Second-order nonhomogeneous equations

Finally, consider the nonhomogeneous equation
x" + a i (t)x' + ao(t)x = b(t), a3 , b E C(R, C), (4.24)
a^ (t + w) = aj (t), b(t + w) = b(t)
for some w > 0. The set S2,,(b) of all solutions of (4.24) which have period w
is an affine space
S2p(b) = v + S2p = {x I x = v + u, u E S2p},
where v is a particular solution in S2 p (b) and Se p is the set of all solutions

of the homogeneous equation (4.19) of period w. The system associated with
(4.24) is
Y' = A(t)Y + B(t), (4.25)
where
A=( a o _a l I , B=( °
A and B are clearly periodic with period w, and a solution x of (4.24) has
period w if and only if Y = x is a solution of (4.25) of period w. Theorem 4.7
then implies that if the homogeneous equation (4.19) has no solution of period
w other than the trivial one, then (4.24) has a unique solution of period w for
every b E C(R, C) which is periodic with period w.
In the general case, when (4.19) may have solutions of period w, Theorem
4.9 implies that (4.24) has a solution of period w if and only if
^ w Z*(s)B(s) ds = 0
0
for every solution Z of the adjoint homogeneous equation
Z' = — A*(t)Z
which has period w. If
Z= \z2/ ,
this says that
W
f ds = 0
for every Z satisfying

z' = aoz2 (4.26)
z2 = —zl + dlz2
which has period w. For such a Z, y = Z2 has period w, y'—aly is differentiable,
and
(y' — aly)' + aoy = 0. (4.27)
Conversely, if y = z2 has period w and satisfies (4.27) and zi = aly — y', then
Z will be a solution of (4.26) of period w. The equation (4.27) is called the
adjoint equation to (4.19). Thus (4.24) has a solution of period w if and only
if
f ow y(s)b(s) ds = 0
for every solution y of the adjoint homogeneous equation (4.27) which has
period w.
4.9 Notes
A classical discussion of Mathieu's and Hill's equations can be found in [27,
pp. 404-428]. Hill's equation is also the subject of a short and accessible book
[16]. A short historical commentary on Hill's work can be found in [12, pp.
730-732] .
In the 1970s there was a series of remarkable works related to Hill's equa-
tion. The first surprise was a close relationship between Hill's equation and
a nonlinear partial differential equation. Subsequent research revealed an ex-
tremely rich and subtle theory for certain problems related to Hill's equation.
References to this recent work can be found in [16, pp. iii—iv].
4.10 Exercises
1. Consider the equation x' = a(t)x, where a E C(R, C) is periodic with

period w > 0.
(a) Find the solution x satisfying x(0) = 1.
(b) Find the constant c such that
X (t + w) = cx(t), t E R.
(c) What condition must a satisfy in order that there exist a nontrivial
solution (i) of period w, (ii) of period 2w ?
(d) In case a is real valued, what are the answers to (c) (i) and (c) (ii)?
(e) If a(t) = a, a constant, what must this constant be in order that a
nontrivial solution of period 2w exist ?
2. Consider the nonhomogeneous equation
x' = a(t)x + b(t), (4.28)

where a, b E C(R, C) are periodic with period w > 0.
(a) Show that a solution x of (4.28) is periodic of period w if and only if
x(w) = x(0).
(b) Show that there is a unique solution of period w if there is no nontrivial
solution of the homogeneous equation x' = a(t)x of period w.
(c) Suppose there is a nontrivial periodic solution of the homogeneous equa-
tion of period w. Show that (4.28) has periodic solutions of period w if and
only if
f 0
e —a( ' b(t) dt = 0,
)
where
ft
a(t) a(s) ds.
=J
(d) Find all solutions of period 2ir for the equations
(i) x' = 3x + cos(t),
(ii) x' = cos(t)x + sin(2t),
(iii) x' = cos(t)x + 2.

3. (a) If A E C(1, M.M,(C)), show that A is periodic of period w zA 0, w E R,
A(t + w) = A(t), t E R,
if and only if A is periodic with period -w,
A(t - w) = A(t), t E R.
Thus there is no loss of generality in assuming that A has period w > 0 in

equation (4.1).
(b) If A E C(R, M7z (C)) is periodic of period w > 0, show that A is also
periodic with period kw for any k = 2, 3, ... .
4. Consider the equation
x"+µx=0, µE1.
For what values of will there exist a nontrivial solution x satisfying x(0) _
x(27r)? Find all solutions for each such µ.
5. Consider the system X' = A(t)X, where A has period 27r and is given
by
—2 sin 2 (t) 1 — 2 sin(t) cos(t)
A(t
) _ ( —1 — sin(t) cos(t) —2 cos 2 (t)
(a) Verify that the basis X satisfying X(0) = I2 is given by
C
cos(t) e-2t sin(t)
X(t) _ — sin(t) e -Zt cos(t)
(b) Compute X(2ir) and find a matrix R such that X(27r) = e 2 " R . Then
write X as X(t) = P(t) e Rt , where P(t + 2ir) = P(t), t E R.
(c) Find all solutions X of X' = A(t)X such that
X(t+27)=pX(t), tER,
for some u E R. In particular, find all solutions which are periodic with period
27.
6. Compute a logarithm B for the following C E M2(R), that is, a matrix
B satisfying e B = C.
3 0
(a) C= (
0 2) '
(b) c=(' 11),
(c) C = —^ 1 )
1 0-
7. Show that if there exists a nontrivial solution of
x (n) + an-ix(n-1) + ... + aox = 0, a E R,
satisfying x(0) = x(1), then there exists a nontrivial real solution which is
periodic with period 1.
8. The toxic waste problem from the introduction to this chapter leads to
a homogeneous system
dTl_
= a11T1,
dt
dr2
dt — a21(t)T1 + a22T2
where all and a22 are constant and a2 1 (t) is a nonconstant function with least
period 1. If X(t) is a basis for this system satisfying X(0) = I, find X(1).
9. Let X be the basis for (4.1) such that X(0) = I, and let µl, ... , p. be
the multipliers for (4.1), where each is repeated as many times as its multiplic-
ity.
(a) Show that
= exp Re
Ii
J tr(A(s)) ds .
(b) If Re fo tr(A(s)) ds < 0, show that there is at least one such that
''<1.
(c) If Re fo tr(A(s)) ds > 0, show that there is at least one ,u such that
1µl > 1.
10. Let u be a multiplier for (4.1), and let X be a nontrivial solution
satisfying X(t + w) = µX (t).
(a) Show that
X(t + kw) = µ k X (t), t E R,
for each k = 1, 2, 3, ... .
(b) If Iµi < 1 show that IX(s)I —* 0 as s — oo. (Hint: If s > 0 then s/w
lies between two consecutive integers, k < s/w < k+1, k = 0,1, ._ . , and hence
s = t + kw, 0 < t < w. Use (a).)
(c) If i > 1 show that X is unbounded on [0, oc).
(d) If = 1 show that
0<m<IX(s)1<M, sER,
where
m=inf{IX(t)I 0<t<w},
M = sup{^X(t)j J0 < t < w}.

11. (a) If X is the basis for (4.1) satisfying X(0) = I, show that X(t +
kw) = X(t)X k (w), k = 1, 2, 3.... .
(b) Show that X(s)I —* 0, s —* oc, if and only if X''(w)I — 0, k oo.
(c) Show that X(s)I is bounded on [0, oc) if and only if jX k (w)I is bounded
fork = 1,2,3,....
12. Suppose that El, E2 > 0 and el + E2 < 1. Define an n x n matrix-
valued function A(t) to be the constant matrix Al on intervals of the form
(K — cl, K + E2), where K = 0, ±1, ±2, ... , and another constant matrix A2
on intervals of the form (K + E2, K + 1 — El). Notice that A(t + 1) = A(t).

The differential equation
X' = A(t)X
does not have a continuous coefficient matrix A(t), but we may consider as so-
lutions those continuous functions X (t) which satisfy the equation on intervals
where A(t) is continuous.
(a) Using the existence and uniqueness theorem for linear systems, show
that the initial value problem
X' = A(t)X, X(to) = ^
has a unique solution for any to E R and E F 1 .

(b) If X(t) is the basis satisfying X(0) = I, describe X(1). (Hint: What is
X(E2) )?
(c) Describe the relevance of this problem to the toxic discharge problem
from the introduction to this chapter.
13. Consider the system
X'=AX, AEM,,(C), t e R. (4.29)
The coefficient function A(t) = A has period w for any w > 0.

(a) Show that (4.29) has a nontrivial solution of period w > 0 if and only if
A has an eigenvalue A of the form A = 27rik/w, k = 0, ±1, ±2,.... (Hint: The
eigenvalues y of E = eA`^ have the form y = e, where A is an eigenvalue of
A.)
(b) If A = 2irik/w, k = 0, ±1, ±2, ... , is an eigenvalue of A, show that
(4.29) has a nontrivial solution of period w of the form
X(t) = e )` t e, E C.
(c) Show that (4.29) has a nontrivial solution X satisfying
X(t + w) = tX (t), t E R, (4.30)
for some
i = l pl e ie , 0 < 9 < 27r,
if and only if A has an eigenvalue A of the form
A = -[log() + iO + 2irik], k = 0, f1, f2, .... (4.31)
(d) If A is given by (4.31) show that (4.29) has a nontrivial solution satis-
fying (4.30) of the form
X(t) = e at
e, E C.
14. Suppose X is a nontrivial solution of (4.1) satisfying

X(t+w) = pX(t), t E R,
for some E C. Show that
X(t) = e At P(t),
where P(t+w)=P(t),tER, and e A'=p.

15. Show that if (4.9) (with B not the zero function) has a solution X
satisfying
X(t+w) = pX(t), t E R,
then µ=1.
16. Consider the nonhomogeneous equation (4.9) when A(t) = A, a con-
stant matrix.
(a) Show that X' = AX + B(t) has a unique solution of period w if and
only if A has no eigenvalue of the form A = 2irik/w, k = 0, ±1 ± 2, ... .
(b) Show that X' = AX + B(t) has a solution of period w if and only if
f
for every ( E C' satisfying eA*w( = c.
17. Let Cp (R, C') be the set of all continuous functions f : R —^ C"' having
period w > 0, and define
(f, g) = J g * (s)f (s) ds, f,g E Cp( 7 ,C n )
(a) Show that (,) is an inner product on Cp (7Z,Ctm)

(b) Let A E C(R, M,,(C)) have period w, and put
L(X)=X'—AX
for each X E Cp (R, Cn), the set of all X E C 1 (R, C), having period w, and
define
L* (Y) = —Y' — A*Y
for Ye CP (R, Cn). Thus L, L* are operators taking C(RC) into C(R, Cn).
Define the range Ran(L) and nullspace N(L) by
Ran(L) = {B E CP (7Z,Ctm)IL(X) = B for some X E Cp(R,C n )}
N(L) = {X E C(RC)IL(X) = 0},

and similarly for Ran(L*), N(L*). With these definitions show that Theorem
4.9 can be stated as
(i) Ran(L) = (N(L*))1,

(ii) dim(N(L)) = dim(N(L*))

This is a direct analogue of Theorem 4.8.

18. From Theorem 4.7 (or Theorem 4.9) we know that if dim(S p ) = 0 then
there is a unique X E Sp (B) for every B E C(R,CTh) of period w. Show that
if there is an X E Sp (B) for every B E CP (R,C), then dim(Sp ) = 0. (In the
terminology of exercise 13, Ran(L) = CP (R.,Ct) if and only if dim(Sp ) = 0.)
19. Consider the homogeneous system
X'= A(t)X, tER,
where A E C(R, Mn (C)) has period w > 0, and let S(µ) denote the set of all
solutions X satisfying
X (t + w) = yX (t), t E R.
Similarly, let S* (v) denote the set of all solutions Y of the adjoint equation
Y'=— A*(t)Y, tER.
(a) Show that if µ 0 then

dim(S*(1/µ)) = dim(S(p)).
(Hint: Show that
[,uIn — X (w)] * = — µX * ( w)[ 1 /jiIn — Ylw)]

where X is the basis of X' = A(t)X such that X(0) = I n and Y = (X*) -1 .)
(b) Suppose A is real, A(t) = A(t), t E R. Show that
dim(S*(1/µ)) = dim(S*(l/µ)) = dim(S(µ)) = dim(S(µ))
for each p 0.
20. Show the following extension of formula (4.18):
X(t + kw) = X(t) + X (t) [I n, + X (w) + • • • + X k-1 (w)] C, t E R,
where X satisfies X' = A(t)X + B(t) and C = X(w) — X(0) = X(w)X(0) + r).
(Putting t = 0 gives (4.18).)
21. Consider the nonhomogeneous equation
X' = A(t)X + B(t), (4.32)
where A E C(R, Mn,(C)) and B E C(R, CTh) have period w > 0, together with
the boundary condition
X(0) — X (w) = /3, ,3 E C. (4.33)

Let S(B, ,0) denote the set of all solutions X of (4.32) satisfying (4.33); thus
S(B, 0) = S(B).
(a) Show that S(B, /3) is an affine space,
S(B„Q) =V+SP ,
where V is a particular solution of (4.32) satisfying (4.33).

(b) Show that there exists an X E S(B, ,Q) if and only if
f Y*( s )B( s ) ds +Y*(0)/3 = 0
for every Y E S. (Hint: A solution X of (4.32) satisfies (4.33) if and only if

there exists a E Cn satisfying
[I, - X(W)] 77+Q,
where
r^ = X(w) -1 (s)B(s) ds
fO LO X
and X is the basis for X' = A(t)X such that X(0) = I a .)
22. The following equations have coefficients with period 27r.
(i) X/I + x = 0,
(ii) (2 + cos(t))x" + cos(t)x = 0.
For both of these

(a) find the basis X = (u, v) satisfying X(0) = I2. (Hint: For (ii), x(t) _
2 + cos(t) is a solution. Try a second solution of the form y = ux.)
(b) What are the multipliers?
(c) Determine a basis for the set of all solutions of period 27r.
23. Consider the equation (4.19) with real-valued coefficients al, ao with
the condition (4.21). In case (iii), when p = —1 is the multiplier, we showed
that if there is a basis U = (ul, u2) satisfying
U(t + w) = —U(t), t E 1?, (4.34)
then U is a basis satisfying
U(t + 2w) = U(t), t E R. (4.35)
Show that if (4.19) has a basis U satisfying (4.35) (with p = —1 being the
only multiplier), then U satisfies (4.34), as follows. If X is the basis satisfying
X(0) = I2, then a Jordan form J for X(w) has one of the two forms
(i) J = —I2,
0 1
(ii) J=—I2+N, N= 0 0 ,
and X(2w) = (X(w)) 2 = X(0) = I2.

24. Consider the general linear equation (4.19) with coefficients ao, al hav-
ing period w > 0 and where we assume that al E C l (R, C), a0 E C(R, C).
Show that x satisfies (4.19) if and only if
/ l
y(t) = x(t) exp I
1
0 ,
al(s) ds
satisfies Hill's equation
y" + a(t)y = 0, a = ao — ai/2 — a 2 /4.
25. Consider Hill's equation
x" + a(t)x = 0, t E R, (4.36)
where a E C(R,1Z), a(t + w) = a(t) for some w > 0, and a(t) <0 for t E 7Z.
(a) If u is the solution of (4.36) satisfying u(0) = 1, u'(0) = 1, show that u
is real.
(b) Show that u(t) > 0 for all t >_ 0 as follows. If not, let ti be the smallest
t > 0 such that u(ti) = 0. Show that u'(ti) < 0. Then
0<^ tl u"(s) ds=u'(tl)-1,

0
which is a contradiction.
(c) Show that u(t) >_ 1+t for t >_ 0, so that u(t) —f +oo as t +oo. (Hint:
u"(t) = —a(t)u(t) > 0 for t >_ 0.)
(d) Conclude that the multipliers Al, µ2 for (4.36) are real and, if equal
(µl = µ2 = ±1), there is no basis of (4.36) consisting entirely of periodic
solutions.
26. Let u E C 2 (I, R), I = [a, b], and suppose u(a) = u(b) = 0, u(t) > 0 for
t E (a, b). Show that
(b — a)
pb
I
u ^)
(( ds > 4.
One approach is to let
max{u(t)la < t < b} = u(c) > 0, a < c < b.
The mean value theorem implies there are a, T satisfying
u(c) — u(a) = u'(a)(c — a), a < a < c,
u(b) — u(c) = u'(r)(b — a), c < r < b.

Then
1 1 _ 1 [ u(c)— u(a)u(b) — u(c) 1
= u/ (
+ b—c uc c—a
c ) / ( T) < u (c)
0 -) (
-
—ca b—c
/T lu // (s)j ds <
fab
I u(( ))1 ds.
If a = c — a, = b — c, then (a + 0) 2 > 4a/@ implies

iu"(s)l
(b — a) —
4 < 1/a +b
1/Q <
Ja u(s)
ds.
27. Consider Hill's equation
x" + a(t)x = 0, (4.37)
where a E C(R, R), a(t+w) = a(t), t E R for some w > 0. Suppose p is a real
multiplier for this equation, and u is a nontrivial solution satisfying
u(t+w)=pu(t), tERZ.
Show that we may assume that u is real.

28. Assume that in Hill's equation (4.37) the following hold: (i) a is not
the zero function,
(ii) Ja(s) ds < 4.

J W a(s) ds > 0, (iii) w
J
(a) Show that any nontrivial real solution u satisfying
u(t + w) = tu(t), t E 7Z, (4.38)
for some p E R is such that u(t) = 0 for some t E [0, w]. (Hint: Suppose
u(t) > 0 for t E [0, w]. Then
fu N /s) fw r a /(sl 1 2 /w
0
I
u(l ) ds +
s ow
a(s) ds
f = I
I u(ls)1
L J ds
+J
a(s) ds
which contradicts (i) and (ii).)

(b) Show that all solutions of (4.37) are bounded on R, together with their
first derivatives.
To establish (b), show that (i)—(iii) imply that the multipliers µl, µ2 for
(4.37) are not real. If they are real, there is a nontrivial real solution u satisfying
(4.38) for some p E R. By (a) there are points a < b such that u(a) = u(b) = 0,
b — a < w, and u(t) 0, t E (a, b). Then by exercise 26 and (iii),
4< (b — a)
I b lu"(s)l
u(s)
ds < w
— f o
w Ia(s)I ds < 4,
which is a contradiction. Thus the multipliers Al, µ2 are distinct nonreal com-
plex numbers (µ2 = µ l ) of magnitude 1.

4 Periodic Coefficients 1997

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

4 Periodic Coefficients 1997

Uploaded by

Copyright:

Available Formats

Downloaded 03/13/13 to 141.213.236.110. Redistribution subject to SIAM license or copyright; see http://www.siam.org/journals/ojsa.

X' = A(t)X, A(t +1) = A(t).

4.2 Floquet's theorem

where A E C(R, M(C)), and A is periodic with period w > 0,

Y'(t) = X'(t + w) = A(t + w)X (t + w) = A(t)Y(t), t E R.

Thus if X is a basis for the solutions of (4.1) then so is Y, where Y(t) _

X (t +w). This fact leads to a fundamental representation result for a basis X.

that is, Rw is a logarithm of C. It follows that

X(t + w) = X(t)e', t E R. (4.3)

If P(t) = X(t)e -Rt we see that for all t E R,

P'(t)e Rt + P(t)Re R t = A(t)P(t)e Rt

or P' + PR = AP. Putting X = PY in (4.1) then gives

PY' + P'Y = APY = P'Y + PRY,

4.3 The logarithm of an invertible matrix

matrix exponential (section 3.1) that C = QJQ -1 = e B , where B = QKQ -1 .

or when J is the r x r matrix

In case (i) choose K to be any value of log(µ). For example, if y = ^ µ l e ie

and then the matrix

1(z) = log(1 + z) = z — z 2 /2 + z 3 /3 — ... _

1+z=exp(l(z))=1+(z—z 2 /2+...)+ 1 (z—z 2 /2+•••) 2 +•••, (4.6)

the rearrangement of the terms on the right obtained by collecting ascending

l(N/µ) = (N/µ) — (N/µ) 2 / 2 + (N/u) 3 / 3 — ... _ E(-1)k+l(N/µ)k/k

same rearrangement of terms. Since the coefficients after rearrangement agree

so R can be computed once we know X at 0 and w. Also the periodicity of

X(t + w) = pX (t), t E R, (4.7)

if and only if it is a multiplier for (4.1). In particular, (4.1) has a nontrivial

X(t + w) = X(t + w) = x(t)Ce = µ x(t) = px(t) ,

(aX +,3Y)(t + w) = aX(t + w) + 3Y(t + w)

= caµX (t) + O/tY(t) = µ(c X + /3Y) (t), t E R,

S(o< + 077) = X( + /3i) = aX + ,3Xi7 = aS(^) + /S(]),

Since the Wronskian WX = det(X) satisfies (Theorem 2.6)

det(X(w)) = det(X(0)) exp VOW tr(A(s)) ds

In particular, none of the ,i is zero, and if n — 1 of them are known, the

4.5 The behavior of solutions for large t

Then X(w) = e R, = exp(QJQ -1 w) = Q exp(Jw)Q -1 , and since

exp(Jw) = diag(exp(J i w), ... , exp(Jw)),

for some enumeration of the ,u and A. If Rl is another matrix such that

and it follows that

µ^ I = exp(Re(Aj)w), Re(A) = W logI i I, j = 1,... ,n.

From the representation (4.2) for X and a Jordan canonical form J =

where A is an eigenvalue of R and P is a vector polynomial with coefficients

X2(t) = exp(Ait)[tPi(t) + P2(t)],

From this representation of Xi (t) = Pi(t)e Jt we have the following ana-

4.6 First-order nonhomogeneous systems

X' = A(t)X + B(t), (4.9)

where V is a particular solution in SP (B).

x' _ — cos(t)x + cos(t),

with n = 1, A(t) _ — cos(t), B(t) = cos(t). The solution x satisfying x(0) _

x(t) = 1 — exp(— sin(t)) + exp(— sin(t)), E C,

v(t) = 1 - exp(- sin(t)), u(t) = exp(- sin(t))e.

On the other hand, consider

Therefore this equation has no solution of period 21r for any E C.

Y'(t) = X'(t + w) = A(t + w)X(t + w) + B(t + w) = A(t)Y(t) + B(t).

X(t) = X(t) + X(t) f t X -1 (s)B(s) ds, t E R.

[I n -X(w)] = X(w) X -1 (s)B(s) ds. (4.10)

If I,, - X(w) is invertible, then clearly

is the unique = X(0) which yields a periodic solution X of (4.9) of period

T = I n — X(w), rl = X(w) —' (s)B(s) ds.

The set S I is a vector subspace of Cn, and if S is a subspace of Cn, then

a direct sum, so that every E Cn can be written uniquely as + rl, where

dim(S) + dim(S 1 ) = n. (4.12)

Theorem 4.8: If T E Mn (C), then

(i) Ran(T) = (N(T*)) 1 ,

(ii) dim(N(T)) = dim(N(T*)).

(Te, ^) = (T = (T()* = (e, T*).

I n —X(w) = — X (w)[I — Y(w)] (4.16)

(X(kw) _ (X(0) + k(* i , k =1, 2, ... .

kI( 71I = ( X(kw) — ( * X( 0 )I < 2 MI(I, k=1,2,...,