Está en la página 1de 7

Exponential Dichotomy

Zhisheng Shuai Department of Mathematical and Statistical Sciences University of Alberta


April 11, 2005

This lecture note is only used for my presentation on the course Math 525 Ordinary Dierential Equations IIB taught by Prof. Michael Li. Most materials are chosen from W.A. Coppels book. Consider linear autonomous system x (t) = Ax(t), and nonhomogeneous system x (t) = Ax(t) + f (t),
n 1

(L) (N L)

where A is a constant n n matrix, x R , t R and f C . Then we have equivalence of following three statements: (a) Re (A) = 0; (b) (L) has no nontrivial bounded solution for t R; (c) For each f bounded for t R, (N L) has a unique solution x(t) that is bounded for all t R, where
t

x(t) =

K (t s)f (s)ds =

eA(ts) P f (s)ds
t

eA(ts) Qf (s)ds.

()

Here, K ( t s) = eA(ts) P, ts eA(ts) Q, t < s,

and Re (AP ) < 0, Re (AQ) > 0, P + Q = I , P 2 = P . Clearly, (a) (b). Now want to show (b) (c). First, if (b) holds, then (a) holds. Furthermore, we can get a solution of the form (). Now verify x(t) is a bounded solution of (N L). In fact,
t

x (t) = P f (t) +

AeA(ts) P f (s)ds + Qf (t)


t t t

AeA(ts) Qf (s)ds = f (t) + Ax(t),

and |x(t)|

Ke(ts) ds +

Ke(st) ds

||f ||

2K ||f ||,

where ||f || = suptR |f (t)| since from Re (AP ) < 0 and Re (AQ) > 0, we have respectively |eA(ts) P | Ke(ts) , t s, |eA(ts) Q| Ke(st) , s t, where > 0 and K > 0. From the uniqueness in (b), by superposition principle, we can get the uniqueness in (c). Conversely, if (c) holds, by superposition principle, we can get the conclusion (b). 1

Note that for nonautonomous system, (a) doesnt work. However, we can generalize it to get the concept of exponential dichotomy. Then, using exponential dichotomy, we study the existence of bounded solution under small perturbation of corresponding linear nonautonomous system. Let X (t) be a fundamental matrix for the linear dierential equation x (t) = A(t)x(t), (1)

where the n n coecient matrix A(t) is continuous on an interval J . The equation (1) is said to posses an exponential dichotomy if there exists a projection P and positive constants K, L, , such that |X (t)P X 1 (s)| Ke(ts) for t s, (2) and |X (t)(I P )X 1 (s)| Le (st) for s t. (3)

It is said to posses an ordinary dichotomy if the inequalities (2) and (3) hold with and not positive but zero. The autonomous equation x (t) = A0 x(t) (4) has an exponential dichotomy on J0 = (0, ) if and only if no eigenvalues of the constant matrix A0 has zero real part. It has an ordinary dichotomy on above J0 if and only if all eigenvalues of A0 with zero real part are of simple type. The equation (1) has an exponential dichotomy with P = I if and only if it is uniformly asymptotically stable, and an ordinary dichotomy with P = I if and only if it is uniformly stable. The equivalent denition of exponential dichotomy is to change conditions (2) and (3) to the following three conditions |X (t)P | K0 e(ts) |X (s)P | for t s, |X (t)(I P ) | L0 e (st) |X (s)(I P ) | for s t, and |X (t)P X 1 (t)| M0 for all t, (5) (6) (7)

where K0 , L0 , M0 are positive constants and is an arbitrary constant vector. Suppose the projection P has rank k . Then the rst condition (5) says that there is a k -dimensional subspace of solutions tending to zero uniformly and exponentially as t . The second condition (6) says that there is a (n k )-dimensional subspace of solutions tending to zero uniformly and exponentially as t . The third condition says that the angle between these two subspaces remains bounded away from zero since when the angle 0 the norms of two project tend to innity. The proof for equivalence of denition of exponential dichotomy is following: First, (2), (3) = (5), (6) and (7). Since P is a projection, i.e., P 2 = P , and (I P )(I P ) = I P , we have |X (t)P | = |X (t)P X 1 (s)X (s)P | |X (t)P X 1 (s)| |X (s)P | Ke(ts) |X (s)P |, t s, |X (t)(I P ) | = |X (t)(I P )X 1 (s)X (s)(I P ) | |X (t)(I P )X 1 (s)| |X (s)(I P ) | Le (st) |X (s)(I P ) |, s t, |X (t)P X 1 (t)| Ke0 = K.

and

Conversely, for any , setting = X 1 (s) , we get |X (t)P X 1 (s) | |X (t)(I P )X 1 (s) | K0 e(ts) |X (s)P X 1 (s) | K0 e(ts) |X (s)P X 1 (s)| | | K0 M0 e(ts) | |, t s, L0 e (st) |X (s)(I P )X 1 (s) | L0 e (st) |X (s)(I P )X 1 (s)| | | L0 (M0 + 1)e (st) | |, s t,

thus (2) and (3) hold. Now assume the fundamental matrix X (0) = I for (1) and the interval J = J0 = (0, ). Consider the inhomogeneous equation x (t) = A(t)x(t) + f (t), (8) where vector function f (t) is locally integrable on J0 , i.e., t |f (s)|ds is bounded, or Lebesgue integrable on J0 , i.e., 0 |f (s)|ds is bounded, or bounded, i.e., supt0 |f (t)| < . Correspondingly, denote the Banach space of all locally integrable vector functions f as M , the Banach space of all Lebesgue integrable vector functions f as L, and the Banach space of all bounded vector functions f as C , with norm
t+1 t+1

||f ||M = sup


t0 t

|f (s)|ds, |f (s)|ds,
0

(9) (10) (11)

||f ||L = and

||f ||C = sup |f (t)|,


t 0

respectively. Clearly, L M and C M . Then we have conclusions as follow: Theorem 1 (W.A. Coppel). The inhomogeneous equation (8) has at least one bounded solution for every function f L if and only if the homogeneous equation (1) has an ordinary dichotomy. Theorem 2 (W.A. Coppel). The inhomogeneous equation (8) has at least one bounded solution for every function f M if and only if the homogeneous equation (1) has an exponential dichotomy. Theorem 3 (W.A. Coppel). Suppose (1) has bounded growth, i.e., there exist real constant C, such that its fundamental matrix X (t) satises |X (t)X 1 (s)| Ce(ts) , for t s. Then the inhomogeneous equation (8) has at least one bounded solution for every function f C if and only if the homogeneous equation (1) has an exponential dichotomy. Proof of Theorem 1, 2 and 3 Step 1: Suciency If (1) has exponential dichotomy or ordinary dichotomy, then consider the solution
t

x(t) =
0

X (t)P X 1 (s)f (s)ds


t

X (t)(I P )X 1 (s)f (s)ds.

(12)

Similarly, we can show that x(t) is a solution of (8) and x(t) is bounded for every f C if (1) has exponential dichotomy. Furthermore, if (1) has exponential dichotomy, then for every f M , x(t) is still a bounded solution of (8) since
t 0

e(ts) |f (s)|ds (1 e )1
t

t+1

|f (s)|ds

and
t

e(st) |f (s)|ds (1 e )1

t+1

|f (s)|ds.

If (1) has ordinary dichotomy, then for every f L, i.e., < , x(t) is still a bounded solution of (8). Step 2: Projection and a lemma Let V1 be the subspace of Rn consisting of the initial values of all bounded solutions of (1), and let V2 be the xed subspace of Rn supplementary to V1 . Set P denote the projection from Rn to V1 with nullspace V2 . Lemma 1. Suppose the equation (8) has a bounded solution for every function f B , where B denote any one of the Banach spaces L, M, C . Then there exists a least constant rB = rB (P ) > 0 such that, for every f B , the unique bounded solution y (t) of (8) with y (0) V2 satises ||y ||C rB ||f ||B . Proof. It follows from the superposition principle that, for every f B , the equation (8) does have a unique bounded solution y (t) with y (0) V2 . Moreover, the map T : f y is linear. We will show that T has a closed graph. Suppose fn f in B and yn = T fn y in C . Then, t t y (0) = limn yn (0) V2 and for any xed t we have 0 f (s)ds = limn 0 fn (s)ds. Hence
t t t

t |f (t)|dt 0

y (t) y (0) = lim

yn (s)ds = lim

(A(s)yn (s) + fn (s))ds =


0 0

(A(s)y (s) + f (s))ds.

Thus y (t) is a solution of the equation (8). Therefore, since it is bounded, y = T f . It now follows from the Closed Graph Theorem that the linear map T is continuous. This proves the existence of some constant rB > 0, and the existence of a least one can be deduced immediately. Step 3: Construction of the bounded solution Put X (t)X 1 (s) for 0 s < t G(t, s) = X (t)(I P )X 1 (s) for 0 t < s. Then G is continuous except on the line s = t, where it has a jump discontinuity. If f is a function in B which vanishes for t > t1 , then
t1 t

y (t) =
0

G(t, u)f (u)du =


0

X (t)P X 1 (u)f (u)du


t

t1

X (t)(I P )X 1 (u)f (u)du

is a solution of (8). Moreover, it is bounded from (5) since


t1

y (t) = X (t)P
0

X 1 (u)f (u)du for t t1 ,


t1

and y (0) = (I P )
0

X 1 (u)f (u)du V2

since X (t)y (0) exponentially increases from (6). Thus, by lemma 1, we have ||y ||C rB ||f ||B . Step 4: Proof of Theorem 1

Let be any constant vector and let f be the function dened by f (t) = 0 for s t s + h otherwise,

where s 0 and h > 0. Then f L and ||f ||L = h| |. Therefore,


s+ h

|y (t)| =
s

G(t, u)du rL ||f ||L = rL h| |.

Dividing by h and letting h 0, we obtain for any t = s |G(t, s) | rL | |. Hence, since is arbitrary, |G(t, s)| rL . Namely, and |X (t)P X 1 (s)| rL , 0 s < t, |X (t)(I P )X 1 (s)| rL 0 t < s.

By continuity, when t = s, it still holds. Thus (1) has ordinary dichotomy. Step 5: Lemma from boundedness to exponential decay Lemma 2. Suppose the equation (8) has a bounded solution for every f C with rC in lemma 1. Let x(t) be a solution of the corresponding homogeneous equation (1) and put x1 (t) = X (t)P X 1 (t)x(t), x2 (t) = X (t)(I P )X 1 (t)x(t). If, for some xed s 0, |x1 (t)| N |x(s)| for s t s + r, then If, for some xed s 0, |x2 (t)| N |x(s)| for max(0, s r) t s, then Proof. Take f (t) = |x2 (t)| eN |x(s)|erC
1

|x1 (t)| eN |x(s)|erC

(ts)

for s t < .

(st)

for 0 t < s.

(t)x(t) , |x(t)|

where x(t) = X (t) is any nontrivial solution of the homogeneous equation (1) and (t) is any continuous real-valued function such that 0 (t) 1 for all t 0, and (t) = 0 for t t1 . Then ||f ||C = 1 and hence, by the arbitrary nature of , we have
t1 t0

G(t, u)x(u)|x(u)|1 du rC ||f ||C = rC for 0 t0 t1 and t 0.

(13)

Putting t1 = t, respectively, t0 = t, we obtain


t

|X (t)P |
t0

|X (u) |1 du rC for t t0 0, 5

(14)

t1

|X (t)(I P ) |
t

|X (u) |1 du rC for t t1 ,
1

(15)
(ts)

since x(t) = X (t) . Replacing by P , respectively, (I P ) , it follows by multiplying erC and integration that
s t0 t1 s

|X (u)P |1 du erC

(ts)

t t0

|X (u)P |1 du for t s t0 , |X (u)(I P ) |1 du for t s t1 .

(16)

|X (u)(I P ) |1 du erC

(st) t

t1

(17)

We can write x(t) = X (t) for some vector . Replacing t0 by s and s by s + rC in the inequality (16) we obtain for t s + rC rC |x(s)|1 N
s+rC s

|x1 (u)|1 du eerC

(ts) s

|x1 (u)|1 du.

Using the inequality (14), this gives for t s + rC


t 1

|x1 (t)| rC
s

|x1 (u)|1 du

eN |x(s)|erC

(ts)

Since the same inequality evidently holds also for s t s + r, this proves the rst assertion of the lemma. The proof of the second assertion is similar, replacing s by s rC and t1 by s in the second inequality (17). Step 6: Proof of Theorem 2 Since C M and L M , the equation (8) has a bounded solution for every f C and for every f L. This is all the information that we will actually use. By Theorem 1 and its proof, equation (1) has ordinary dichotomy with K = rL and = 0. Hence in Lemma 2 we can take N = rL , for every solution x(t) of (1) and every s 0, and obtain the exponential dichotomy with K = erL and 1 = rC . Step 7: Proof of Theorem 3 Without loss of generality, assume C 1 and > 0 in the following inequality |X (t)X 1 (s)| Ce(ts) , t s. Replacing by X 1 (s) and putting t1 = in the inequality (15), this gives for t s |X (t)(I P )X 1 (s) | rC
s 1

|X (u)X 1 (s) |1 du

rC C 1 | |1
s

e(su) du

Thus In the same way and hence

|X (t)(I P )X 1 (s)| rC C for t s. |X (t)(I P )X 1 (s)| rC Ce(ts) for t s, |X (t)P X 1 (s)| (1 + rC )Ce(ts) for t s.

(18)

Similarly, from the inequality (14) we obtain |X (t)P X 1 (s)| rC C [1 e(ts) ]1 for t > s. 6

By using this inequality for t s h, where h = 1 log and the previous inequality for t s h, we get |X (t)P X 1 (s)| (1 + 2rC )C for all t s. It now follows from Lemma 2 using (18) and (19) that |X (t)P X 1 (s)| e(1 + 2rC )CerC |X (t)(I P )X 1 (s)| erC CerC Thus (1) has an exponential dichotomy.
1

1 + 2rC 1 + rC

(19)

(ts)

for 0 s t, for 0 t s.

(st)

References
[1] W.A. Coppel, Dichotomies in Stability Theory, Lecture Notes in Mathematics, Vol 629, Springer, Berlin, 1978.

También podría gustarte