- r + 1 then we still have xl(~0) # 0. Otherwise, ]Xl(~0)(0)[ = 0 and [xt(~0)(s - 1)[ = I 0(s)l -- o f o r s = - r + 1 . . . . . o, i.e. x~(~o) = o and so x(tp)(n) = 0 for n = 1, 2 . . . . . Therefore, we m a y assume Xl(tp) ;~ 0. Obviously, Ilxl(tp)]l - B. T h e n by (ii) and (3), V(x(¢)(2)) = V(x(xt(~o))(1)) <
max
V(x~(~)(s)) <_
s = -r,...,O
max
V(~(s)) < m,
s = -r,...,O
which implies Ix(~o)(2)[ < e. By induction, we can conclude that for 0 ;e [[~oH< ~ we have
V(x(tp)(n)) <
max
V(x,_l(tp)(s))
$ = --r, . . . , 0
<-...<_
max S ........
V(tp(s)) < m 0
and, thus, Ix(~p)(n)l < e This proves that x = 0 is stable.
for n = l, 2 . . . . .
(4)
Invariance principle
741
Next, we show the asymptotic stability. Let 6o > 0 be the corresponding 6(t) to the given e = / / i n the stability. Under the assumptions we may conclude by theorem 2 and remark 1 that when Iltpl] < 6o, Xn(~O) ~ M y ( G )
as n --* 0%
C Ev(G)
where G = 1~0 e X: II~ll -< P). Clearly, to prove the asymptotic stability we need only to point out M y ( G ) = 10]. In fact, by the above arguments of deriving (4) we know that for any ¢p: 0 ~ whenever IIx.(~)ll -< t~ for n e z + we always have max
V(to(s)) >
s = -r,...,O
max
V(x.(~o)(s))
I1~11 -<
for n >_ r + 1.
s = -r,...,O
Therefore, by the definition of M y , ~o ¢ My(G). Hence, M y ( G )
= [0}.
•
2. For convenience, in dealing with concrete equations we often choose Lyapunov functions as
Remark
v(x) = Ixl.
Then condition (ii) reduces to
Ixt~0)(l)l < II~ll Example
for any 0 ~ II~011 <- 8.
(5)
1. Consider x ( n + 1) = f ( x n ) ,
(6)
n ~ Z+,
where x e R k,
If(~0)l---zll¢ll
for II~0ll---
with some constants 0 < L < 1 and fl > 0. Then we may choose V ( x ) = ]x], and (5) is obviously satisfied. Hence, by theorem 3 we conclude that the zero solution x = 0 of (6) is asymptotically stable. E x a m p l e 2. More specifically, we consider the following delay difference system u(n + 1) = av(n - rO/(1 + u2(n - r2) ), v(n + 1) = bu(n - rl)/(1 + v2(n - r3)),
n • Z+, 1
(7)
where a, b are constants, rl, r2 and r 3 are integers, lal < 1, Ibl < 1, lal + Ibl ~ 0; 0 ~< rl, r2, r 3 < r for some integer r > 0. Let
x = (u, v) r,
(~ = ((PI' (P2) T,
and
Ixl = lul + Ivl,
I1~11 =
max
[[~Pl(s)l + I~p2(s)l}.
s= -r,...,O
Choose V(x) = V(u, v)=
Ixl = lul + Ivl.
742
SHUNIAN ZHANG
Then
Ix(¢o)(1)l =
[u(~0)(1)l + Iv(¢)(1)l [acp2(-r,)/(1 + (0~(-r2))l + Ibrpl(-r,)/(1
lallcz(-r0l
+
¢0~(-r3))1
+ Ibll~l(-r0l
< maxl[al, Ibl}(l~P2(-r~)l + I ~ ( - r ~ ) l ) _< zll~ll
f o r all ~o,
where 0 < L ---- maxllal, Ibll < 1. Therefore, it follows f r o m example 1 that the zero solution o f (7) is (globally) asymptotically stable.
R e m a r k 3. It is remarkable that in theorem 3 A(,) V = V(f(~o)) - V(~p(0)) = V(x((o)(1)) - V(¢(0)) m a y not even be nonnegative to conclude the asymptotic stability, which m u c h i m p r o v e d the k n o w n results (cf., say, [6]) in the case o f a u t o n o m o u s equations. We emphasize here that even for such a simple example 1 or its special case, example 2, we were unable to show the desired conclusion before by the k n o w n results. This is one o f the motivations to establish the invariance principles for delay difference systems. N o w we discuss the conditions for solutions o f (1) tending to constant limits. For L y a p u n o v function V = V(x), we define
K v = {¢P • X : V(~o(s)) = V(¢(O)) for s = - r . . . . . 0}. THEOREM 4. A s s u m e that there exist a L y a p u n o v function V = V(x) and a positively invariant, closed set G C X w.r.t. (1) such that for any ~0 • G whenever x(fp)(.) is b o u n d e d for n > - r , lim { max
V(xn(rp)(s))}
exists.
n ~oO $ ~ - - r , . . . ~ O
Then ~((o) C K v implies that
V(x(~o)(n)) = V(x.(~p)(0)) --, const,
as n ~ oo.
P r o o f . Let ~0 • G and x(~o)(') be b o u n d e d for n > - r . By assumption we m a y assume lim { max n~eo
V(xn(tP)(s))} = c
for some c • R.
$=--r,...DO
Suppose that Q((0)C K v but V ( x ( ¢ ) ( n ) ) ~ " const, as n ~ oo. T h e n since V(x(~o)(n)) is b o u n d e d for n > - r , there must be some sequence [nil and some real n u m b e r a # c such that ni ~ oo and V(x(~o)(ni)) ~ a as i ~ oo. By l e m m a 2 [xni(~P)} is p r e c o m p a c t and by assumption G is positively invariant and closed, hence, there exist some subsequence o f {n~}, denoted by In~} again, and some g / • G such that Xni(~o) ~ ~ as i ~ oo. Therefore, ~, • ~(~p) C Kv. By the definition o f K v , we have for some b • R,
v(~(s)) = v(~,(o)) = b,
s = - r . . . . . O.
lnvariance principle
743
Noting that xn~(~p)(0) ~ ~,(0) as i ~ oo, and the continuity o f V we derive V(x~,(~p)(0)) -- V(x(q~)(ni)) --" a = V(~,(0))
as i - ,
oo.
Thus, V(q/(s)) = V(~(0)) = a,
s = - r . . . . . 0.
O n the other hand, max
V(x.,(~o)(s))
~
c
=
S= --r,...,O
max
V(qt(s)) = a.
as n ~
oo.
-r,...,O
s=
This contradicts the fact a # c. Therefore, whenever Q(g0 C K v we have V(x(~o)(n)) -* const.
•
In the following we establish a simple criterion for ~(~p) C K v . T H E O ~ 5. Assume that there exists a L y a p u n o v function V = V ( x ) and a positively invariant, closed set G C X w.r.t. (1) such that: (i) V(x(~p)(1)) _< max V(~o(s)) for any (0 e G; S=
--r,...,O
(ii) for cp e G, V(x(tp)(1)) =
max
V(~o(s)) implies ~o e K v .
S ~ --r, ... ,0
T h e n for any ~p e G we have ~ ( ¢ ) C K v whenever x(~p)(.) is b o u n d e d for n >_ - r . Therefore, V(x(tp)(n)) ~ const,
as n ~ ~ .
P r o o f . Let ¢ e G be such that x(¢)(.) is b o u n d e d for n _> - r . Let ~v e ~ ( ¢ ) . Clearly, g / e G since G is closed and positively invariant. By theorem 2 ~ ~ E v ( G ) , hence, there is some c e R such that max s=
V(x(~,)(n + s)) =
max
-r,...,O
V(x,(gJ)(s)) =
s = -r,...,O
max
for n e Z +.
V(g,(s)) = c
(8)
s = -r,...,O
In particular, if n = r + 2, then we have max
V(x(~u)(r + 2 + s)) =
S = --r, . . . , 0
max
V(w(s)) = c.
S = --r, ...,0
Suppose that for some s o e { - r . . . . . 0} V(x(~u)(r + 2 + So)) =
max
V(x(qt)(r + 2 + s)) =
s = -r,...,O
max
V(W(s)) = c.
(9)
s = -r,...,O
Let s* = r + So + 1. Then 1 < s* < r + 1, it follows f r o m (9), and (8) with n = s* that V(x(qO(s* + 1)) = c =
max
V(x~,(q/)(s)),
S= --r,...,O
i.e. V(x(xs,(~u))(1))=
max
V(xs,(~u)(s)).
S= --r,...,O
Hence, by assumption (ii) (note that xs,(~v) we conclude that
e
G since g~ e G and G is positively invarariant),
Xs4~) e gv.
744
SHUNIAN ZHANG
Thus, for s = - r . . . . . 0.
V(xs,(Ct)(s)) = V(x~.(~u)(O))
This and (8) with n = s* imply that max S=
V(xs,(~u)(s)) = V(xs.(~)(0)) = c.
--r,...,O
Since - r _< - s * + 1 ___ 0, we have V(x(tu)(1)) = V(xs.(~u)(-s* + 1)) = c =
max S=
V(~u(s)).
-r,...,0
By (ii) again, this implies that ~/~ K V. Therefore, ~((0) C Kv. T o complete the p r o o f it suffices to mention that in this case the assumptions o f theorem 2 are certainly satisfied. F r o m the p r o o f o f theorem 2 we k n o w that whenever x((0)(.) is b o u n d e d for r/_> - r , lira [ max fl'-~
exists.
V(xn(~)(s))l = c
S=--r,...,O
Therefore, the desired conclusion follows f r o m theorem 4.
•
R e m a r k 4. It is notable that the L y a p u n o v functions V in theorem 5 (as well as in theorems 2
and 4) are not necessarily nonnegative. This brings out m u c h convenience in applications. E x a m p l e 3. Consider the scalar equation 0
x(n + 1) = ax(n) + b x ( n - r) - b x ( n ) x ( n - r) ~
[x(n + s) - x(n)l 2,
(10)
S ~ --r
where n • Z +, a > 0, b > 0 are constants, and a + b = 1. Let G = {(0 • X : 0 _< ~o(s) _< 1, s = - r . . . . . 0}. Then it is easy to show that G is positively invariant w.r.t. (10). Clearly, it suffices to show that 0 < x(~o)(l) _< 1 for ¢ • G. In fact, if (0 • G, then Ileal[ --- 1 and x((0)(1) = a(0(0) + b ( 0 ( - r )
I
o
1 - cp(0) ~ $=
1
[(0(s) - (0(0)12 > a(0(0) > 0,
--r
and
b)ll(01[
x((0)(1) _< all(oil + bll(0H = (a +
= 11(011~ 1.
Thus, o _~ x((0)O) -< 1.
C h o o s e V(x) = x. Then for any ~ e G we have as above that V(x((0)(1)) = x((0)(1) _<
11(oil =
max
v((0(s)).
$ = --r, ..., 0
O n the other hand, if (0 e G and V(x((0)(1)) =
max
V((0(s)), then x((0)(1) = H(0[I, i.e.
S= --r,...,O 0
(0(0) - b(0(O) + b ( 0 ( - r ) - b(0(O)(0(-r) ~ $=
[(0(s) - (o(0)12 = --r
11(011.
(11)
Invariance principle
745
I f (0 = 0, t h e n o b v i o u s l y (0 ~ K v. H e n c e , we need o n l y to c o n s i d e r the case o f (0 ;~ 0. S u p p o s e (0(0) < 11(oil ~ 0. T h e n the l e f t - h a n d side o f (11) < all(011 + oil(011 = 11(oil, hence, the e q u a l i t y (11) c a n n o t h o l d . T h e r e f o r e , (0(0) = 11(011. H o w e v e r , t h e n it clearly follows f r o m (11) t h a t 0 ( 0 ( - r ) = (0(0) and (0(0)(0(-r) ~ [(0(s) - (0(0)] 2 = 0. $=
--r
T h e r e are t w o p o s s i b l e cases: (i) if ( 0 ( - r ) = (0(0) = 0, t h e n (0 = 0, a n d c e r t a i n l y (0 e K v ; (ii) if (0(-r) = (0(0) ;~ 0, then ~s°= _r[(0(s) - (0(0)] 2 = 0 implies t h a t (0(s) = (0(0)
for s = - r . . . . . 0.
This m e a n s t h a t (0 e K e. T h e r e f o r e , in a n y case, the a s s u m p t i o n s o f t h e o r e m 5 are satisfied. Hence, V(x((0)(n)) = x((0)(n) ~ const,
as n -+ ~ .
R e m a r k 5. It is easy to see t h a t x(n) = c for a n y c e R is a s o l u t i o n o f (10) if a + b = 1. C O N C L U D I N G REMARKS It is w o r t h m e n t i o n i n g t h a t we a r e n o t expected to o b t a i n all the p a r a l l e l results for d e l a y d i f f e r e n c e systems to the k n o w n results for d e l a y d i f f e r e n t i a l systems, b e c a u s e there are essential differences b e t w e e n discrete m o d e l s a n d c o n t i n u o u s m o d e l s . F o r instance, even f o r the simple d e l a y d i f f e r e n c e systems o f the f o r m x(n + 1) - x(n) = - h ( x ( n ) ) + h(x(n - 1)), if h(u) = u 2 a n d c h o o s e x ( - l ) = l a n d x(0) = 2, t h e n it is easily seen t h a t t h e c o r r e s p o n d i n g s o l u t i o n is [ - 1 , 2 , - 1 , 2 . . . . }; while if h ( u ) = u 3 which is strictly increasing, a n d c h o o s e x ( - 1 ) = 1 a n d x ( 0 ) = 0, t h e n the s o l u t i o n is {1,0, 1 , 0 . . . . 1. N o t e t h a t n o n e o f t h e m is a s y m p t o t i c a l l y c o n s t a n t as n -+ ~ . This is a distinct c o n t r a s t to the relevant results m e n t i o n e d in [5]. I n short, t h e r e are quite a few o p e n p r o b l e m s f o r d e l a y difference systems to be i n v e s t i g a t e d further. W e m a y discuss these in a s u b s e q u e n t p a p e r . Acknowledgement--The author wishes to thank the referee for making valuable suggestions and comments on the original manuscript. REFERENCES 1. LASALLE J. P., The Stability of Dynamical Systems. Society for Industrial and Applied Mathematics, Philadelphia (1976). 2. LASALLE J. P., Stability theory for ordinary differential equations, J. diff. Eqns 4, 57-65 (1968). 3. HALE J. K., Sufficient conditions for stability and instability of autonomous functional differential equations, J. diff. Eqns 1,452-482 (1965). 4. HALE J. K., Theory of Functional Differential Equations. Springer, New York (1977). 5. H A D D O C K J. R. & TERJEKI J., Liapynov-Razumikhin functions and invariance principle for functional differential equations, J. diff. Eqns 48, 95-122 (1983). 6. ELAYDI S. & ZHANG SHUNIAN, Stability and periodicity of difference equations with finite delay. Funkcialaj Ekvacioj (to appear).