International Journal of Pure and Applied Mathematics ————————————————————————– Volume 22 No. 2 2005, 199-205
SOME NONLINEAR ERGODICS THEOREMS ON HILBERT SPACES A. Ghomari1 , A. Ouahab2 , A. Mbarki3 S. Lahrech4 § , A. Benbrik5 1,2,3,4,5 Department
of Mathematics Faculty of Science Mohamed First University Oujda, MOROCCO 2 e-mail:
[email protected] 3 e-mail:
[email protected] 4 e-mail:
[email protected] 5 e-mail: benbrik @sciences.univ-oujda.ac.ma
AMS Subject Classification: 47A35 Key Words: ergodic theorems, nonlinear ergodic theory
1. Introduction In [1] Baillon has shown the first nonlinear ergodic theorem for nonexpansive mapping. Recently Kakutani et Rouhani [3] studied the problem of the convergence the Ces` aro means of contractive sequence. Later, Oukili proved in [4] that there is an equivalent between the two works and studied the problem of the convergence the means of contractive family. In this paper we establish several results involving hypotheses weak enough to include a number of ergodic theorems as special cases. Furthermore, we deal with the problem relative to nonlinear ergodic theory for the mappings and semigroups are not necessarily continuous. Received:
June 3, 2005
§ Correspondence
author
c 2005, Academic Publications Ltd.
200
A. Ghomari, A. Ouahab, A. Mbarki, S. Lahrech, A. Benbrik
Throughout this paper, let C be a nonempty convex subset of a Hilbert space H endowed with the scalar product , and w − lim denote the weak limit. We introduce the following definition. Definition 1. A mapping T : C → C is called (α)-mapping on C, if there are two constants a, b such that a + 2b = 1 and k T x − T y k≤ a k x − y k +b[k T x − y k + k T y − x k] for all x, y ∈ C. Note that a nonexpansive mapping is (α)-mapping with b = 0. But the converse is not true as is shown in the following example Example 2. Define T : [0, 1] → [0, 1] as follows o if x = 1 , Tx = 1 2 if x 6= 1 . It is easy to see that T is not nonexpansive for any norm defined on R. But T is (α)-mapping with a = b = 13 . Likewise, we have the folowing definition. Definition 3. A sequence {a(n)} ⊂ C (resp. a family (a(t), t ≥ 0) of elements of C ) is called (α)-sequence (resp.(α)- family), if there are two constants a, b such that a + 2b = 1 and k a(n + 1) − a(m + 1) k ≤ a k a(n) − a(m) k +b[k a(n + 1) − a(m) k + k a(m + 1) − a(n) k] , for all n, m ∈ N (respectively k a(v + e) − a(u + e) k ≤ a k a(v) − a(u) k +b[k a(u + e) − a(v) k + k a(v + e) − a(u) k] , for all u, vet e ∈ R+ ).
2. Main Results Theorem 4. Let (a(n), n ∈ N ) be a bounded (α)-sequence in H. Then n−1
S(n) =
1X a(i) n i=0
converge weakly in H.
SOME NONLINEAR ERGODICS THEOREMS...
201
Proof. Since (S(n), n ∈ N ) is a bounded sequence in Hilbert space, then there exists un subsequence (n(k), k ∈ N ) in N converges to ∞ as k → ∞ and an element x in H such that: w − lim S(n(k)) = x. k→∞
Assume that there exists another subsequence (v(k), k ∈ N ) in N , v(k) → ∞ as k → ∞ and an element y in H such that w − lim S(v(k)) = y. k→∞
So, to complete the proof it suffices to show that y = x. Similarly, we can suppose that: w − lim S(n(k)) = x , k→∞ n(k)−1 X 1 k a(i) k2 = z , n(k) i=0 n(k)−1 X 1 k a(i) − a(s) k2 = f (s) for all s ∈ N , n(k) i=0
and
w − lim S(v(k)) = y , k→∞ v(k)−1 X 1 k a(i) k2 = w , v(k) i=0 1 v(k)−1 X k a(i) − a(s) k2 = g(s) for all s ∈ N . v(k) i=0
Since (a(n), n ∈ N ) is a (α)-sequence then f and g are decreasing functions, we show for example that f is decreasing function. Indeed k a(i + 1) − a(s + 1) k2 ≤ a2 k a(i) − a(s) k2 +b2 k a(i) − a(s + 1) k2 + b2 k a(i + 1) − a(s) k2 +2ab k a(i) − a(s) k [k a(i + 1) − a(s) k + k a(i) − a(s + 1) k] + b2 k a(i) − a(s + 1) kk a(i + 1) − a(s) k ≤ [a2 +2ab] k a(i)−a(s) k2 +[2b2 +ab][k a(i+1)−a(s) k2 + k a(i)−a(s+1) k2 ] . Then
202
A. Ghomari, A. Ouahab, A. Mbarki, S. Lahrech, A. Benbrik
n(k)−1 n(k)−1 X X 1 1 2 2 k a(i + 1) − a(s + 1) k ≤ [a + 2ab] k a(i) − a(s) k2 n(k) n(k) i=0
i=0
n(k)−1
+ [2b2 + ab]
X 1 k a(i + 1) − a(s) k2 n(k) i=0
n(k)−1 X 1 + [2b + ab] k a(i) − a(s + 1) k2 . n(k) 2
i=0
Letting k → ∞, we obtain f (s + 1) ≤ (a2 + 2ab)f (s) + (2b2 + ab)[f (s) + f (s + 1)]. Since f is positive, decreasing function. Then we have n−1
1X f (i) = p. p = inf{f (s), s ∈ N }, and limn→∞ n i=0
So, using the parallelogram legality in H, we get 1 < a(i), a(j) >= [k a(i) k2 + k a(j) k2 − k a(i) − a(j) k2 ], 2 for all i, j ∈ N . Next n−1
v−1
i=0
j=0
1 1X 1X < S(n), S(v) >= [ k a(i) k2 + k a(j) k2 2 n v n−1 v−1
−
1 1 XX k a(i) − a(j) k2 ] , (1) nv i=0 j=0
for all n, v ∈ N . We replace in (1), n by n(k) and, letting as k → ∞, we obtain v−1
v−1
j=0
j=0
1X 1X 1 k a(j) k2 − f (j)] . < x, S(v) >= [z + 2 v v
(2)
Similary, we replace in (2), v by n(k) and, letting k → ∞, we get 1 < x, x >= [z + z − p]. 2
(3)
SOME NONLINEAR ERGODICS THEOREMS...
203
Using (2), we replace v by v(k). As k → ∞, we obtain 1 < x, y >= [z + w − p]. 2
(4)
By symmetry, q = inf{g(s), s ∈ N } we can write 1 < y, y >= [w + w − q], 2
(5)
1 < y, x >= [w + z − q]. 2
(6)
So from equations (3), (4), (5), (6) and the following identity < x − y, x − y >=< x, x > + < y, y > −2 < x, y > , we have < x − y, x − y >= 0. This completes the proof of the theorem. Theorem 5. Let (a(t), t ≥ 0) be a bounded (α)- family in H such that a : R+ → H is a continuous mappings. Then S(t) =
1 t
Z
t
a(u) du 0
weakly converge in H as t tend to ∞. The proof of above theorem is straightforward analogue of proof of the Theorem 4. 2.1. Applications Let Γ = (T (t), t ≥ 0) be a family of (α)-mappings (a, b independent of t) from C to C, Γ is called strongly continuous semigroup of C, if: 1) For each x ∈ C, t → T (t)x is continuous on R+ , 2) T (t1 + t2 ) = T (t1 ) ◦ T (t2 ) for all t1 , t2 ≥ 0, and 3) For all x ∈ C limt→0+ k T (t)x − x k= 0. Our next corollary is an extension of a Baillon’s Theorem, see [1].
204
A. Ghomari, A. Ouahab, A. Mbarki, S. Lahrech, A. Benbrik
Corollary 6. Let T : C → C be a (α)-mapping with C be a closed convex subset a Hilbert space, if there exists an element x0 of C such that (T i x0 , i ∈ N ) is bounded. Then for all x ∈ C, n−1
Sn x =
1X i T x n i=0
weakly converge to a fixed point of T in C. Proof. From Theorem 4, let y = w − lim Sn x0 , n→∞
which implies, since C is weakly closed (by using the fact that C closed convex subset) that y ∈ C. The other parts since T is (α)-mapping, then k T i+1 x0 − T y k2 ≤a ˜ k T i x0 − y k2 +˜b[k T i+1 x0 − y k2 + k T y − T i x0 k2 ] , (A) with a ˜ = a2 + 2ab and ˜b = 2b2 + ab. So k T j x0 − y k2 =k T j x0 − T y k2 + k y − T y k2 +2 < T j x0 − T y, T y − y > . (B) Put βj =k T j x0 − T y k2 . From (B) and (A) we have βi+1 ≤ a ˜[βi + k T y − y k2 +2 < T i x0 − T y, T y − y >] + ˜b[βi+1 + k T y − y k2 +2 < T i+1 x0 − T y, T y − y >] + ˜bβi . An easy induction shows that 1 (˜ a + ˜b) n
n−1 X
[βi+1 − βi ] ≤ [˜ a + ˜b] k T y − y k2 +2(˜ a + ˜b) < Sn x0 − T y, T y − y >
i=0
+
2˜b [< T n x0 − T y, T y − y > − < x0 − T y, T y − y >] , n
hence 1 a + ˜b] k T y − y k2 +2(˜ a + ˜b) < Sn x0 − T y, T y − y > (˜ a + ˜b) [βn − β0 ] ≤ [˜ n
SOME NONLINEAR ERGODICS THEOREMS...
205
2˜b [< T n x0 − T y, T y − y > − < x0 − T y, T y − y >] . n Letting n → ∞, we obtain +
0 ≤ [˜ a + ˜b] k T y − y k2 −2[˜ a + ˜b] k T y − y k2 , this means that k T y − y k2 = 0. Then, to complete the proof it suffices to show that for all x ∈ C (T i x, i ∈ N ) is bounded. So, since T is a (α)-mapping and y ∈Fix(T ); then for all x ∈ C k T i x − y k≤k x − y k , therefore, (T i x, i ∈ N ) is bounded. The following result is a direct extension of a main result in [2] and it can be proved by using the same argument and the methods very similar of the proof of above corollary. Corollary 7. Let Γ = (T (t), t ≥ 0) ba a strongly continuous (α)semigroup of a convex closed subset C of a Hilbert space, if there exists an element x0 of C such that (T (t)x0 , t ≥ 0) is bounded. Then for each x ∈ C. Z 1 t T (u)x du S(t) = t 0 weakly converge to fixed point y of Γ (i.e. T (t)y = y for all t ≥ 0) as t tend to ∞. References [1]
J.B. Baillon, Un th´eor`eme de type ergodique pour les contractions non lin´eaires dans les espaces de Hilbert, C. R. Acad. Sci. Paris S´erie s´erie A., 208 (1975), 1511-1514.
[2]
J.B. Baillon H. Brezis, Une remarque sur le comportement asympotique des semi-groupes non lin´eaires, Honston J. Math., 2 (1976), 5-7.
[3]
S. Kakutani, Bd. Rouhani, Preprint.
[4]
H. Oukili, Contribution a la th´eorie ergodique des op´erateurs lin´eaires et non lin´eaires, Th`ese de Doctorat D’´etat Es Sciences, Univ S. M. Ben abdallah. Fes (1990).
206