A NOTE ON STABILITY BY SCHAUDER'S THEOREM

2 downloads 0 Views 78KB Size Report
(3b). |g(t, x) − g(t, y)| ≤ L(δ)f(t)|x − y|, t R+, |x|, |y| ≤ δ, and where K, M and α > 1 are ..... |(P2φ)(t0) − (P2φ)(t1)| ≤ ϵ/3 if 0 ≤ t0 < t1 ≤ T and t1 < t0 + δ3, which ...
A NOTE ON STABILITY BY SCHAUDER’S THEOREM

T.A. Burton Northwest Research Institute 732 Caroline Street Port Angeles, WA 98362

Tetsuo Furumochi Department of Mathematics Shimane University Matsue, Japan 690-8504

Abstract. In this paper we consider a nonlinear perturbation of an asymptotically stable linear differential equation. The standard perturbation theory using Gronwall type inequalities or Liapunov’s direct method do not seem to yield asymptotic stability of the perturbed equation. Our purpose here is to give an example of a stability result using fixed point theory.

0. Introduction Consider a system x0 = A(t)x + B(t, x)

(a) where solutions of (b)

y 0 = A(t)y

are known to be bounded and, perhaps, tend to zero as t → ∞, while B(t, x) is small compared to x in some sense for small x. There is a large classical theory concerning the asymptotic behavior of solutions of (a). An early account is found in Chapters 2 and 3 of Bellman [1], with progressive treatments in Coddington and Levinson [2], Hartman [4], Yoshizawa [7], and Hale [3], to name just a few. Frequently it is assumed that either: i) The zero solution of (b) is uniformly asymptotically stable, 1

or ii) A is constant or periodic and all solutions of (b) are bounded. R∞ Then the additional assumption that |B(t, x)|/|x| ≤ g(t) where 0 g(t)dt < ∞ can often

be used to conclude that the zero solution of (a) is stable and, possibly, that small solutions tend to zero. If B is independent of t, much more can be said. To place our work here in perspective, the reader might consider the following result found in Bellman [1; p. 91]. THEOREM. Let A be a constant n × n matrix, f(t, x) be continuous, all solutions of y 0 = Ay be bounded, and let |f(t, x)|/|x| ≤ c1 g(t) where

Z

∞ 0

g(t)dt < ∞.

Then the zero solution of x0 = Ax + f(t, x) is stable. Methods of proof usually involve Liapunov functions or Gronwall type inequalities which depend so much on properties i) and ii) above. In a series of projects we are investigating ways in which fixed point theory can be used to obtain stability results which have eluded investigators using the above mentioned methods. Thus, we are particularly interested in examples and fine detail. Here, we consider an equation

(1)

x00 + 2f(t)x0 + x + g(t, x) = 0,

t ∈ R+ ,

with a prototype being f(t) = 1/(t + 1) and g(t, x) = x2 /(t + 1). The linear part is asymptotically stable, but not uniformly asymptotically stable; and this makes it difficult to obtain asymptotic stability for a perturbed system. The question we propose to answer here is: ”How can we effectively use fixed point theory to prove that the zero solution of (1) is asymptotically stable?” 2

1. Asymptotic Stability Consider the scalar equation

x00 + 2f(t)x0 + x + g(t, x) = 0,

(1)

t ∈ R+ ,

where R+ := [0, ∞), f(t) and g(t, x) are continuous, f(t) > 0, f(t) → 0 as t → ∞, Rt 0 f(s)ds → ∞ as t → ∞, and (2)

|f 0 (t) + f 2 (t)| ≤ Kf(t), t ∈ R+ , K < 1,

(3a )

|g(t, x)| ≤ Mf(t)|x|α , t ∈ R+ ,

(3b )

|g(t, x) − g(t, y)| ≤ L(δ)f(t)|x − y|, t ∈ R+ , |x|, |y| ≤ δ,

and where K, M and α > 1 are constant, and L(δ) is continuous and increasing. The linear part of equations such as (1) has been extensively studied, as may be seen in Hatvani [5], for example. Change (1) to a system

x0 =y − f(t)x y 0 =(f 0 (t) + f 2 (t) − 1)x − f(t)y − g(t, x) and write it as

0

z =



−f(t) −1

1 −f(t)



z+



0 0 f (t) + f 2 (t) 3

0 0



z+



0 −g(t, x)



,

or z 0 = A(t)z + B(t)z + F (t, z).

(4) Notice that

A(t)

Z

t 0

Z t A(s)ds = ( A(s)ds)A(t). 0

Hence, the principal matrix solution of z 0 = A(t)z is  Z t Z t cos t exp( A(s)ds) = exp(− f(s)ds) − sin t 0 0

sin t cos t



.

If U(t) = (uij (t)), then we define the norm by the maximum of the row sums of (|uij (t)|) and we denote that norm by kU(t)k. We then see that the norms of the matrices in (4) satisfy (5)

Z t Z t √ k exp( A(u)du)k ≤ 2 exp(− f(s)ds), t ≥ s ≥ 0 s

s

and kB(t)k = |f 0 (t) + f 2 (t)| ≤ Kf(t), t ∈ R+ . Now the solution z(t) of (4) with z(0) = z0 is Z t Rs Rt A(s)ds − A(u)du z(t) = e 0 (z0 + e 0 (B(s)z(s) + F (s, z(s)))ds). 0

We will use Schauder’s first theorem to prove that for each small z0 , a solution through z0 tends to zero as t → ∞. A statement of Schauder’s theorem can be found in Smart [6; p.15]. For reference it may be stated as follows. THEOREM. Let (C, k·k) be a normed space, and let S be a compact convex nonempty subset of C. Then every continuous mapping of S into S has a fixed point. Here, we will take C to be the Banach space of bounded and continuous functions φ : R+ → R2 with the supremum norm, kφk (which will cause no confusion with the matrix norm given above). 4

Let a be a number with 0 < a < ((1 − K)/M)β , where β := 1/(α − 1), and let |z0 | ≤ a, and define

S0 := {φ : R+ → R2 |φ(0) = z0 , |φ(t)| ≤ q(t) on R+ , φ ∈ C}, where | · | denotes the Euclidean norm on R2 and q(t) is the unique solution of the initial value problem x0 = f(t)(K − 1 + Mxα−1 )x, x(0) = a. Define a map P on S0 by Z t Rs Rt A(s)ds − A(u)du (z0 + (P φ)(t) := e 0 e 0 (B(s)φ(s) + F (s, φ(s)))ds), 0

and maps Pi for i = 1, 2, 3 by Rt A(s)ds (P1 φ)(t) := e 0 z0 , Z t Rs Rt A(s)ds − A(u)du 0 (P2 φ)(t) := e e 0 B(s)φ(s)ds, 0

and Z t Rs Rt A(s)ds − A(u)du 0 (P3 φ)(t) := e e 0 F (s, φ(s))ds. 0

Note that (P1 φ)(0) = z0 and (P2 φ)(0) = (P3 φ)(0) = 0. Next, we have the following two results. LEMMA 1. If φ ∈ S0 then |(P φ)(t)| ≤ q(t), t ∈ R+ .   x(t) Proof. Let φ(t) = . Then, for t ∈ R+ we have y(t) Rt   Rt − f (s)ds x(0) cos t + y(0) sin t A(s)ds |(P1 φ)(t)| =|e 0 φ(0)| = e 0 −x(0) sin t + y(0) cos t Rt − f (s)ds =e 0 |φ(0)| Rt − f (s)ds ≤ae 0 . 5

Next, for t ∈ R+ we obtain t Rt A(u)du |(P2 φ)(t)| =| e s B(s)φ(s)ds| 0 Z t R t   0 A(u)du 0 2 = e s (f (s) + f (s))x(s)ds 1 0 Z t R t   sin(t − s) − f (u)du 0 2 = e s (f (s) + f (s))x(s)ds cos(t − s) 0 Z t Rt − f (u)du ≤K e s f(s)|x(s)|ds.

Z

0

Similarly, for t ∈ R+ we have t

Z

Rt A(u)du |(P3 φ)(t)| =| e s F (s, φ(s))ds| 0 Z t R t   sin(t − s) − f (u)du e s g(s, x(s))ds = cos(t − s) 0 Z t Rt − f (u)du ≤M| e s f(s)|x(s)|α ds|. 0

Recall that x(t) is the first component of φ(t) and in the definition of S0 we have |φ(t)| ≤ q(t) so |x(t)| ≤ |φ(t)| ≤ q(t), and this will now be used. For t ∈ R+ we obtain −

|(P φ)(t)| ≤ae



≤ae

Rt 0

Rt 0

f (s)ds

f (s)ds

+K

Z

+K

Z

t −

e

s

0 t −

e 0

Rt Rt s

f (u)du

f(s)|x(s)|ds + M

f (u)du

Z

f(s)q(s)ds + M

Z

t −

e

0 t −

e 0

Rt s

Rt s

f (u)du

f (u)du

f(s)|x(s)|α ds

f(s)q α (s)ds

= : r(t). Since we have q 0 (t) = −f(t)q(t) + f(t)q(t)(K + Mq α−1 (t)), we obtain −

r(t) =ae



=ae

Rt 0

Rt 0

f (s)ds

f (s)ds

+

Z

+

Z

t −

e 0 t



e 0

Rt

f (u)du

(Kf(s)q(s) + Mf(s)q α (s))ds

Rt

f (u)du

(q 0 (s) + f(s)q(s))ds.

s

s

6

If we integrate

Rt 0



e

Rt s

f (u)du 0

q (s)ds by parts we get r(t) = q(t) on R+ , which gives the

desired inequality. LEMMA 2. There is a continuous increasing function δ = δ() : (0, 2a) → (0, ∞) with (6)

|q(t0 ) − q(t1 )| ≤  if 0 ≤ t0 < t1 < t0 + δ

and (7)

|(P φ)(t0 ) − (P φ)(t1 )| ≤  if φ ∈ S0 and 0 ≤ t0 < t1 < t0 + δ. Proof. First, it is easy to see that for any  with 0 <  < 2a there is a δ1 > 0 such

that (8)

|q(t0 ) − q(t1 )| ≤  if 0 ≤ t0 < t1 < t0 + δ1 .

Next, for any φ ∈ S0 we have Rt A(s)ds (P1 φ) (t) = A(t)e 0 z0 . 0

Let G be a number with 0 < f(t) ≤ G on R+ . Then from (5) with s = 0 we obtain Rt A(s)ds |(P1 φ) (t)| ≤kA(t)kke 0 k|z0 | R t √ − f (s)ds a ≤ 2(1 + G)e 0 √ ≤ 2(1 + G)a. 0

Thus, for any  with 0 <  < 2a there is a δ2 > 0 such that (9)

|(P1 φ)(t0 ) − (P1 φ)(t1 )| ≤  if 0 ≤ t0 < t1 < t0 + δ2 .

Now let T > 1 be a number such that q(t) ≤ /6 if t ≥ T − 1. 7

Then, since we have |(Pi φ)(t)| ≤ q(t), (i = 2, 3), it is easy to see that (10i )

|(Pi φ)(t0 ) − (Pi φ)(t1 )| ≤ /3 if T − 1 ≤ t0 < t1 .

For any t0 and t1 with 0 ≤ t0 < t1 ≤ T we have |(P2 φ)(t0 ) − (P2 φ)(t1 )| t0

=|

Z

Z R t0 A(u)du s e B(s)φ(s)ds −

t0

≤|

Z

Z R t0 R t1 A(u)du A(u)du (e s −e s )B(s)φ(s)ds| + |

0



Z

0

0 t0

0

≤Ka

t1

R t1 A(u)du e s B(s)φ(s)ds| t1

t0

Z R t0 R t1 A(u)du A(u)du ke s −e s kkB(s)k|φ(s)|ds +

R t1 A(u)du e s B(s)φ(s)ds| t1

t0

Z

t0 0

≤GKa

Z

Z R t0 R t1 √ A(u)du A(u)du ke s −e s kf(s)ds + 2Ka t0

0

R t1 A(u)du ke s kkB(s)k|φ(s)|ds t1 −

e t0

R t1 s

f (u)du

f(s)ds

R t0 R t1 √ A(u)du A(u)du s ke −e s kds + 2GKa|t0 − t1 |.

For any η with 0 < η < T , let R t0 R t1 A(u)du A(u)du d(η) := sup{ke s −e s k|0 ≤ s ≤ t0 < t1 ≤ T and t1 ≤ t0 + η}. It is clear that d(η) → 0+ as η → 0 + . Let δ3 be a number such that 0 < δ3 < 1, and that √ d(δ3 ) ≤ /(6GKT a) and δ3 ≤ /(6 2GKa). Then we have |(P2 φ)(t0 ) − (P2 φ)(t1 )| ≤ /3 if 0 ≤ t0 < t1 ≤ T and t1 < t0 + δ3 , which, together with (102 ), yields (11)

|(P2 φ)(t0 ) − (P2 φ)(t1 )| ≤ /3 if 0 ≤ t0 < t1 < t0 + δ3 .

Similarly, for any t0 and t1 with 0 ≤ t0 < t1 ≤ T we have |(P3 φ)(t0 ) − (P3 φ)(t1 )| 8

Z

t0

Z R t0 A(u)du e s F (s, φ(s))ds −

t1

R t1 A(u)du =| e s F (s, φ(s))ds| 0 0 Z t0 R t0 Z t1 R t1 R t1 A(u)du A(u)du A(u)du α s s ≤M ke −e kf(s)|x(s)| ds + M ke s kf(s)|x(s)|α ds 0

t0

≤GMaα α

≤GMa

Z

t0

Z

t0

0

0

Z R t0 R t1 √ A(u)du A(u)du α ke s −e s kds + 2GMa

t1 −

e t0

R t1 s

f (u)du

ds

R t0 R t1 √ A(u)du A(u)du ke s −e s kds + 2GMaα |t0 − t1 |.

Finally, let δ4 be a number such that 0 < δ4 < 1, and that √ d(δ4 ) ≤ /(6GMT aα ) and δ4 ≤ /(6 2GMaα ). Then we obtain |(P3 φ)(t0 ) − (P3 φ)(t1 )| ≤ /3 if 0 ≤ t0 < t1 ≤ T and t1 ≤ t0 + δ4 , which together with (10)3 , yields (12)

|(P3 φ)(t0 ) − (P3 φ)(t1 )| ≤ /3 if 0 ≤ t0 < t1 < t0 + δ4 .

Thus, from (8), (9), (11) and (12), for δ5 := min{δi : 1 ≤ i ≤ 4}, we have (6) and (7) with δ = δ5 . Since we may assume that δ5 () is nondecreasing, we can easily conclude that there is a continuous increasing function δ : (0, 2a) → (0, ∞) which satisfies (6) and (7). Now we prepare an Ascoli-Arzela-like lemma for functions defined on the infinite interval R+ . LEMMA 3. Let q : R+ → R+ be a continuous function such that q(t) → 0 as t → ∞. If {φk (t)} is an equicontinuous sequence of Rd -valued functions on R+ with |φk (t)| ≤ q(t) for t ∈ R+ , then there is a subsequence that converges uniformly on R+ to a continuous function φ(t) with |φ(t)| ≤ q(t) for t ∈ R+ , where | · | denotes the Euclidean norm on Rd . Proof. It is clear that the set of functions {φk (t)} is uniformly bounded on R+ . Thus, considering intervals [0, n], n a positive integer, and using a diagonalization process there 9

is a subsequence, say {φk (t)} again, converging uniformly on any compact subset of R+ to some continuous function φ(t) with |φ(t)| ≤ q(t) for t ∈ R+ . Because φ(t) → 0 as t → ∞, it will now be possible to show that kφk − φk → 0 as k → ∞, where k · k denotes the supremum norm on R+ . From the definition of q(t), for any  > 0 there is a T > 0 with q(t) < /2 if t ≥ T, which yields (13)

|φk (t) − φ(t)| ≤ 2q(t) <  if k ∈ N and t ≥ T,

where N denotes the set of positive integers. On the other hand, since {φk (t)} converges to φ(t) uniformly on [0, T ] as k → ∞, for the  there is a κ ∈ N with |φk (t) − φ(t)| <  if k ≥ κ and 0 ≤ t ≤ T, which together with (13), implies that kφk − φk <  if k ≥ κ. This shows that kφk − φk → 0 as k → ∞. REMARK. Lemma 3 is a generalization of Ascoli-Arzela’s lemma in the following sense. Let {φk (t)} be a uniformly bounded and equicontinuous sequence of Rd -valued functions on [0, T ], and let |φk (t)| ≤ Q on [0, T ]. Define q(t) := QeT −t for t ∈ R+ , and let φk be an extension to R of φk obtained by defining φk (t) := φk (T )eT −t for t > T . Then, q and {φk (t)} satisfy the assumptions of Lemma 3, and hence, the original sequence {φk (t)} contains a subsequence that converges uniformly on [0, T ] to a continuous function. Let S be a set of functions φ ∈ S0 such that for the function δ in Lemma 2, |φ(t0 ) − φ(t1 )| ≤  if 0 ≤ t0 < t1 < t0 + δ. Then we have the following lemma. 10

LEMMA 4. The set S is a compact convex nonempty subset of C, and the map P : S → S is continuous. Proof. Since the function φ(t) := (q(t)/a)z0 is contained in S, S is nonempty. Clearly S is a convex subset of C. Moreover, Lemma 3 implies that S is compact. Next, we prove that the map P : S → S is continuous. For any φ ∈ S, let ξ := P φ. Clearly we have ξ(0) = z0 and ξ ∈ C. Next, from Lemma 1 we obtain |ξ(t)| ≤ q(t), t ∈ R+ . Thus, we have ξ ∈ S0 , which together with Lemma 2, implies that ξ ∈ S. Hence, P maps S into S. We need to prove that P is continuous. For any φi ∈ S (i = 1, 2) and t ∈ R+ we have |(P φ1 )(t) − (P φ2 )(t)| ≤|

Z

t 0

Rt A(u)du e s (B(s)(φ1 (s) − φ2 (s)) + (F (s, φ1 (s)) − F (s, φ2 (s)))ds|

√ Z t − R t f (u)du ≤ 2 e s |Kf(s) + L(a)f(s)||φ1 (s) − φ2 (s)|ds 0

Z t Rt √ − f (u)du ≤ 2(K + L(a)kφ1 − φ2 k e s f(s)ds 0 √ ≤ 2(K + L(a)kφ1 − φ2 k, which implies that P is continuous. In view of Schauder’s first theorem we have the following result. THEOREM. Under assumptions (2), (3a ), and (3b ), let a be a number with 0 < a < ((1 − K)/M)β , where β := 1/(α − 1), and let |z0 | ≤ a. Then the solution z(t, z0 ) of (4) satisfies |z(t, z0 )| ≤ q(t). We conclude with an example. 11

EXAMPLE. Consider the equation x00 + [2/(t + 1)]x0 + x + x2 /(t + 1) = 0, t ∈ R+ .

(14)

(Thus, f(t) = 1/(t + 1) and g(t, x) = x2 /(t + 1).) Clearly, f(t) → 0 as t → ∞,

R∞ 0

f(s)ds =

∞. Moreover, it is easy to see that (2) holds with K = 0, that (3a ) holds with M = 1 and α = 2, and that (3b ) holds with L(δ) = 2δ. Change (14) to a system z 0 = G(t, z) as before. Let a be a number with 0 < a < 1, and let |z0 | ≤ a. By our result, the solution z(t, z0 ) satisfies |z(t, z0 )| ≤ a/[1 + (1 − a)t]. References 1. Bellman, R., Stability Theory of Differential Equations, McGraw-Hill, New York, 1953 2. Coddington, E. A. and N. Levinson, Theory of Ordinary Differential Equations, McGraw-Hill, New York, 1955. 3. Hale, Jack K., Ordinary Differential Equations, Wiley, New York, 1969. 4. Hartman, Philip, Ordinary Differential Equations, Wiley, New York, 1964. 5. Hatvani, L., Integral conditions on the asymptotic stability for the damped linear oscillator with small damping, Proc. Amer. Math. Soc. 124(1996), 415-422. 6. Smart, D. R., Fixed Point Theorems, Cambridge Univ. Press, Cambridge, 1974. 7. Yoshizawa, T., Stability Theory by Liapunov’s Second Method, Math. Soc. Japan, Tokyo, 1966.

12