Volume 2008, Article ID 316207,20pages doi:10.1155/2008/316207
Research Article
Semilinear Evolution Equations of Second Order via Maximal Regularity
Claudio Cuevas1and Carlos Lizama2
1Departamento de Matem´atica, Universidade Federal de Pernambuco, Avenue Prof. Luiz Freire, S/N, Recife, 50540-740 PE, Brazil
2Departamento de Matem´atica, Facultad de Ciencias, Universidad de Santiago de Chile, Casilla 307-Correo 2, Santiago, Chile
Correspondence should be addressed to Carlos Lizama,clizama@usach.cl Received 26 October 2007; Revised 23 January 2008; Accepted 4 February 2008 Recommended by Alberto Cabada
This paper deals with the existence and stability of solutions for semilinear second-order evolution equations on Banach spaces by using recent characterizations of discrete maximal regularity.
Copyrightq2008 C. Cuevas and C. Lizama. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
1. Introduction
LetAbe a bounded linear operator defined on a complex Banach spaceX. In this article, we are concerned with the study of existence of bounded solutions and stability for the semilinear problem
Δ2xn−Axnfn, xn,Δxn, n∈Z, 1.1 by means of the knowledge of maximal regularity properties for the vector-valued discrete time evolution equation
Δ2xn−Axnfn, n∈Z, 1.2 with initial conditionsx00 andx10.
The theory of dynamical systems described by the difference equations has attracted a good deal of interest in the last decade due to the various applications of their qualitative properties; see1–5 .
In this paper, we prove a very general theorem on the existence of bounded solutions for the semilinear problem1.1onlpZ;Xspaces. The general framework for the proof of this statement uses a new approach based on discrete maximal regularity.
In the continuous case, it is well known that the study of maximal regularity is very useful for treating semilinear and quasilinear problemssee, e.g., Amann6 , Denk et al.7 , Cl´ement et al.8 , the survey by Arendt9 , and the bibliography therein. Maximal regularity has also been studied in the finite difference setting. Blunck considered in 10,11 maximal regularity for linear difference equations of first order; see also Portal12,13 . In14 , max- imal regularity on discrete H ¨older spaces for finite difference operators subject to Dirichlet boundary conditions in one and two dimensions is proved. Furthermore, the authors inves- tigated maximal regularity in discrete H ¨older spaces for the Crank-Nicolson scheme. In15 , maximal regularity for linear parabolic difference equations is treated, whereas in16 a char- acterization in terms ofR-boundedness properties of the resolvent operator for linear second- order difference equations was given; see also the recent paper by Kalton and Portal 17 , where they discussed maximal regularity of power-bounded operators and relate the discrete to the continuous time problem for analytic semigroups. However, for nonlinear discrete time evolution equations like1.1, this new approach appears not to be considered in the litera- ture.
The paper is organized as follows.Section 2provides an explanation for the basic nota- tions and definitions to be used in the article. InSection 3, we prove the existence of bounded solutions whose second discrete derivative is inlp1 < p < ∞for the semilinear problem 1.1by using maximal regularity and a contraction principle. We also get some a priori es- timates for the solutionsxnand their discrete derivativesΔxn andΔ2xn. Such estimates will follow from the discrete Gronwall inequality1 see also18,19 . InSection 4, we give a cri- terion for stability of1.1. Finally, inSection 5we deal with local perturbations of the system 1.2.
2. Discrete maximal regularity
LetXbe a Banach space. LetZdenote the set of nonnegative integer numbers and letΔbe the forward difference operator of the first order, that is, for eachx:Z →Xandn∈Z,Δxn xn1−xn.We consider the second-order difference equation
Δ2xn−I−Txnfn, ∀n∈Z,
x0x, Δx0x1−x0y, 2.1
whereT ∈ BX, Δ2xn ΔΔxn,andf:Z→X.
DenoteC0 I, the identity operator onX, and define
Cn n/2
k0
n 2k
I−Tk, forn1,2, . . . , 2.2
andCn C−n, forn−1,−2, . . . .We define alsoS0 0,
Sn
n−1/2
k0
n 2k1
I−Tk, 2.3 forn1,2, . . . ,andSn −S−n,forn−1,−2, . . . .
Considering the above notations, it was proved in16 that theuniquesolution of2.1 is given by
xm1CmxSmy S∗fm. 2.4
Moreover,
Δxm1 I−TSmxCmy C∗fm. 2.5 The following definition is the natural extension of the concept of maximal regularity from the continuous casecf.,16 .
Definition 2.1. Let 1 < p < ∞. One says that an operator T ∈ BXhas discrete maximal regularity ifKTf:n
k1I−TSkfn−kdefines a bounded operatorKT∈ BlpZ, X.
As a consequence of the definition, ifT ∈ BXhas discrete maximal regularity, thenT has discretelp-maximal regularity, that is, for eachfn∈lpZ;Xwe haveΔ2xn∈lpZ;X, wherexnis the solution of the equation
Δ2xn−I−Txnfn, ∀n∈Z, x00, x10. 2.6
Moreover,
Δ2xnn−1
k1
I−TSkfn−1−kfn. 2.7
We introduce the means
x1, . . . , xnR: 1 2n
j∈{−1,1}n
n j1
jxj
2.8
forx1, . . . , xn∈X.
Definition 2.2. LetXandYbe Banach spaces. A subsetTofBX, Yis calledR-bounded if there exists a constantc≥0 such that
T1x1, . . . , TnxnR≤cx1, . . . , xnR 2.9 for allT1, . . . , Tn ∈ T, x1, . . . , xn∈X, n∈N.The leastcsuch that2.9is satisfied is called the R-bound ofTand is denoted byRT.
An equivalent definition using the Rademacher functions can be found in7 . We note thatR-boundedness clearly implies boundedness. IfX Y, the notion ofR-boundedness is strictly stronger than boundedness unless the underlying space is isomorphic to a Hilbert space 20, Proposition 1.17 . Some useful criteria forR-boundedness are provided in7,20,21 .
Remark 2.3. aLetS,T ⊂ BX, YbeR-bounded sets, thenST:{ST :S ∈ S, T ∈ T}is R-bounded.
bLetT ⊂ BX, YandS ⊂ BY, ZbeR-bounded sets, thenS · T:{S ·T : S ∈ S, T ∈ T} ⊂ BX, ZisR-bounded and
RS · T≤RS·RT. 2.10
cAlso, each subsetM⊂ BXof the formM{λI :λ ∈Ω}isR-bounded whenever Ω⊂Cis bounded. This follows from Kahane’s contraction principlesee20,22 or7 .
A Banach spaceXis said to be UMD if the Hilbert transform is bounded onLpR, Xfor someand then allp ∈1,∞.Here, the Hilbert transformH of a functionf ∈ SR, X, the Schwartz space of rapidly decreasingX-valued functions, is defined by
Hf: 1 πPV
1 t
∗f. 2.11
These spaces are also calledHT spaces. It is a well-known theorem that the set of Banach spaces of classHTcoincides with the class of UMD spaces. This has been shown by Bourgain 23 and Burkholder24 .
Recall thatT ∈ BXis called analytic if the set nT−ITn:n∈N
2.12 is bounded. For recent and related results on analytic operators we refer the reader to25 . The characterization of discrete maximal regularity for second-order difference equations by R-boundedness properties of the resolvent operatorT reads as followssee16 .
Theorem 2.4. LetXbe a UMD space and letT ∈ BXbe analytic. Then, the following assertions are equivalent.
iT has discrete maximal regularity of order 2.
ii λ−12Rλ−12, I−T:|λ|1, λ /1
isR-bounded.
Observe that from the point of view of applications, the above-given characteriza- tion provides a workable criterion; see Section 4 below. We remark that the concept of R- boundedness plays a fundamental role in recent works by Cl´ement-Da Prato26 , Cl´ement et al.22 , Weis27,28 , Arendt-Bu20,29 , and Keyantuo-Lizama30–32 .
3. Semilinear second-order evolution equations
In this section, our aim is to investigate the existence of bounded solutions, whose second discrete derivative is inpfor semilinear evolution equations via discrete maximal regularity.
Next, we consider the following second-order evolution equation:
Δ2xn−Axnf
n, xn,Δxn
, n∈Z, x00, x10, 3.1 which is equivalent to
xn2−2xn1Txnf
n, xn,Δxn
, ∀n∈Z, x00, x10, 3.2 whereT :I−A.
To establish the next result, we need to introduce the following assumption.
Assumption 3.1. Suppose that the following conditions hold.
iThe function f : Z×X×X → X satisfy the Lipschitz condition on X×X, that is for allz, w ∈X×Xandn∈Z, we getfn, z−fn, wX ≤αnz−wX×X,where α: αn∈l1Z.
iif·,0,0∈l1Z, X.
We remark that the conditionα∈l1Ziniis satisfied quite often in applications. For example, it appears when we study asymptotic behavior of discrete Volterra systems which describe processes whose current state is determined by their entire history. These processes are encountered in models of materials with memory, in various problems of heredity or epi- demics, in theory of viscoelasticity, and in solving optimal control problemssee, e.g.,33,34 . We began with the following property which will be useful in the proof of our main result.
Lemma 3.2. Letαnbe a sequence of positive real numbers. For alln, l∈Z, one has n−1
m0
αm
m−1
j0
αj
l
≤ 1 l1
n−1
j0
αj
l1
. 3.3
Proof. PuttingAm:m−1
j0 αj, we obtain l1
Am1−Am Alm
Am1−Am
AlmAl−1m Am· · ·AmAl−1m Alm
≤
Am1−Am
Alm1Al−1m1Am· · ·Am1Al−1m Alm Al1m1−Al1m .
3.4
Hence,
l1n−1
m0
Am1−Am
Alm≤n−1
m0
Al1m1−Al1m
Al1n . 3.5
Denote byW2,p0 the Banach space of all sequencesV Vnbelonging tol∞Z, Xsuch thatV0 V1 0 andΔ2V ∈lpZ, Xequipped with the norm|||V||| V∞Δ2Vp.We will say thatT ∈ BXisS-bounded ifS ∈l∞Z;X.With the above notations, we have the following main result.
Theorem 3.3. Assume thatAssumption 3.1holds. In addition, suppose thatT isS-bounded and that it has discrete maximal regularity. Then, there is a unique bounded solutionx xnof 3.1such that Δ2xn∈lpZ, X. Moreover, one has the following a priori estimates for the solution:
sup
n∈Z
xnXΔxnX
≤3Mf·,0,01e3Mα1, Δ2xp≤Cf·,0,0
1e6Mα1, 1< p <∞,
3.6 whereM:supn∈ZSnandC >0.
Proof. Let V be a sequence inW2,p0 . Then, usingAssumption 3.1 we obtain that the function g:f·, V·,ΔV·is inlpZ, X. In fact, we have
gpp∞
n0
f
n, Vn,ΔVnpX
≤∞
n0
f
n, Vn,ΔVn
−fn,0,0Xfn,0,0Xp
≤2p ∞ n0
f
n, Vn,ΔVn
−fn,0,0pX2p ∞ n0
fn,0,0pX
≤2p ∞ n0
αpnVn,ΔVnpX×X2p ∞ n0
fn,0,0pX,
3.7
where
∞ n0
fn,0,0pX∞
n0
fn,0,0p−1X fn,0,0X
≤f·,0,0p−1∞ ∞
n0
fn,0,0X f·,0,0p−1∞ f·,0,0
1.
3.8
Analogously, we have
∞ n0
αpn≤αp−1∞ α
1. 3.9
On the other hand,
Vn,ΔVnX×XVnXVn1−VnX≤2VnXVn1X≤3V∞. 3.10 Hence,
gpp≤6pVp∞∞
n0
αpn2pf·,0,0p−1∞ f·,0,0
1
≤6pVp∞αp−1∞ α12pf·,0,0p−1∞ f·,0,01,
3.11
proving thatg∈lpZ, X.
SinceThas discrete maximal regularity, the Cauchy problem zn2−2zn1Tzngn,
z0z10 3.12
has a unique solutionznsuch thatΔ2zn∈lpZ, X, which is given by
zn KV n
⎧⎪
⎪⎨
⎪⎪
⎩
0, ifn0,1,
n−1 k1
Skf
n−1−k, Vn−1−k,ΔVn−1−k
, ifn≥2. 3.13
We now show that the operatorK:W2,p0 → W2,p0 has a unique fixed point. To verify thatKis well defined, we have only to show thatKV ∈ l∞Z, X.In fact, we useAssumption 3.1as above andM:supn∈ZSnto obtain
n−1
k1
Skf
n−1−k, Vn−1−k,ΔVn−1−k X
≤Mn−1
k1
f
n−1−k, Vn−1−k,ΔVn−1−k
−fn−1−k,0,0XMn−1
k1
fn−1−k,0,0X
≤Mn−1
k1
αn−1−kVn−1−k,ΔVn−1−kX×XMn−2
j0
fj,0,0X
≤3MV∞ n−2
j0
αjMn−2
j0
fj,0,0X
≤M
3V∞α1f·,0,01 .
3.14 It proves that the spaceW2,p0 is invariant underK.
LetV andVbe inW2,p0 . In view ofAssumption 3.1iandM <∞, we have initially as in 3.14KV n−KV nX
n−1 k1
Sk f
n−1−k, Vn−1−k,ΔVn−1−k
−f
n−1−k,Vn−1−k,ΔVn−1−k X
≤Mn−1
k1
αn−1−kV −Vn−1−k,ΔV −Vn−1−kX×X
Mn−2
j0
αjV −Vj,ΔV −VjX×X ≤3Mα1V −V∞.
3.15
Hence, we obtain
KV − KV∞≤3Mα1V −V. 3.16 On the other hand, using the fact thatS1 I, we observe first that
ΔKV nf
n−1, Vn−1,ΔVn−1 n−1
k1
Sk1− Sk f
n−1−k, Vn−1−k,ΔVn−1−k
, n≥1.
3.17
SinceS2 2I, we get Δ2KV nf
n, Vn,ΔVn
−f
n−1, Vn−1,ΔVn−1
S2−I f
n−1, Vn−1,ΔVn−1
n−1
k1
Sk2−2Sk1 Skf
n−1−k, Vn−1−k,ΔVn−1−k
f
n, Vn,ΔVn n−1
k1
Sk2−2Sk1 TSk f
n−1−k, Vn−1−k,ΔVn−1−k
n−1
k1
I−TSkf
n−1−k, Vn−1−k,ΔVn−1−k .
3.18 Taking into account thatzn1 S∗gnis solution of3.12, we get the following identity:
n−1 k1
Sk2−2Sk1 TSk f
n−1−k, Vn−1−k,ΔVn−1−k
0. 3.19
Using3.19, we obtain forn≥1 Δ2KV nf
n, Vn,ΔVn n−1
k1
I−TSkf
n−1−k, Vn−1−k,ΔVn−1−k
, 3.20
whence, forn≥1, Δ2KV n−Δ2KV n
f
n, Vn,ΔVn
−f
n,Vn,ΔVn n−1
k1
I−TSk f
n−1−k, Vn−1−k,ΔVn−1−k
−f
n−1−k,Vn−1−k,ΔVn−1−k .
3.21 Furthermore, using the fact that Δ2KV 0 f0,0,0, the above identity, and then Minkowskii’s inequality, we get
Δ2KV −Δ2KVp
f0,0,0−f0,0,0pX∞
n1
Δ2KV n−Δ2KV npX1/p
≤ ∞
n1
f
n, Vn,ΔVn−fn,Vn,ΔVnpX1/p
∞
n1
n−1
k1
I−TSk f
n−1−k, Vn−1−k,ΔVn−1−k
−f
n−1−k,Vn−1−k,ΔVn−1−k
p X
1/p
. 3.22
SinceKTis bounded onlpZ, X, usingAssumption 3.1, we obtain Δ2KV −Δ2KVp≤
1KT∞
n1
fn, Vn,ΔVn−fn,Vn,ΔVnpX1/p
≤
1KT∞
n1
αpnV −Vn,ΔV −VnpX×X1/p
≤3
1KTα1V −V∞.
3.23
Hence, we obtain from3.16and3.23
KV − KVKV − KV∞Δ2KV −Δ2KVp
≤3Mα1V −V3
1KTα1V −V 3
M1KTα1V −V abV −V,
3.24
wherea:3Mα1andb:1 1KTM−1.
Next, we consider the iterates of the operatorK. LetV andV be inW2,p0 . Taking into account thatS1 I,S0 0,andV0V1V0V10, we observe first that forn≥2
ΔKV n−ΔKV n n−1
k0
Sk1− Sk
fn−1−k, Vn−1−k,ΔVn−1−k
−f
n−1−k,Vn−1−k,ΔVn−1−k
n−1
k1
Sn−k− Sn−k−1 f
k, Vk,ΔVk−f
k,Vk,ΔVk ,
3.25 whence
ΔKV n−ΔKV nX≤2Mn−1
k1
f
k, Vk,ΔVk
−f
k,Vk,ΔVkX
≤2Mn−1
k1
αkV −Vk,ΔV −VkX×X.
3.26
On the other hand, from3.15we get KV n−KV nX≤Mn−2
k1
αkV −Vk,ΔV −VkX×X. 3.27
Using estimates3.26and3.27, we obtain forn≥2 KV − KV n,ΔKV − KV nX×X≤3Mn−1
k1
αkV −Vk,ΔV −VkX×X. 3.28
Next, usingKV 0 KV 10 and estimates3.28and3.10, we obtain K2V n−K2V nX≤Mn−2
j0
f
j,KV j,ΔKV j
−f
j,KV j,ΔKV jX
≤Mn−2
j1
αjKV − KV j,ΔKV − KV jX×X
≤3M2n−1
j1
αj j−1
i1
αiV −Vi,ΔV −ViX×X
≤1 23M2
n−1
τ1
ατ
2
V −V∞.
3.29
SinceK2V 0 K2V 10, we get
K2V − K2V∞≤ 1 2
3Mα1
2V −V. 3.30 Furthermore, using the identity
Δ2 K2V
n−Δ2 K2V
n
f
n,KV n,ΔKV n
−f
n,KV n,ΔKV n n−1
k1
I−TSk f
n−1−k,KV n−1−k,ΔKV n−1−k
−f
n−1−k,KV n−1−k,ΔKV n−1−k , 3.31 the fact thatΔ2K2V 0f0,0,0for allV ∈ W2,p0 , andLemma 3.2, we obtain
Δ2K2V −Δ2K2Vp Δ2
K2V
0−Δ2 K2V
0pX∞
n1
Δ2 K2V
n−Δ2 K2V
npX1/p
≤
1KT∞
n1
f n,
KV
n,Δ KV
n
−f n,
KV
n,Δ KV
npX1/p
≤
1KT∞
n1
αpnKV − KV
n,Δ
KV − KV
npX×X1/p
≤3M
1KT∞
n1
αpn n−1
k1
αkV −V k,ΔV −V kX×Xp1/p
≤32M
1KT∞
n0
αpn
n−1
k0
αk
p
V −Vp∞1/p
≤32M
1KT1 2
∞
j0
αj
2
V −V∞,
3.32
whence
Δ2K2V −Δ2K2Vp≤1 2
3Mα1
2
1KTM−1V −V. 3.33 From estimates3.30and3.33, we get
K2V − K2V≤ b
2a2V −V, 3.34
withaandb defined as above. Taking into account3.26,3.28,3.29, and3.10, we can infer that
K2V − K2V
j,Δ
K2V − K2V
jX×X≤ 3 23M2
j−1
τ1
ατ
2
V −V∞. 3.35
Next, using estimate3.35andLemma 3.2, we get K3V
n− K3V
nX ≤Mn−2
j1
αjK2V − K2V
j,Δ
K2V − K2V
jX×X
≤1
23M3n−1
j0
αj j−1
τ1
ατ 2
V −V∞
≤1 63M3
n−1
j1
αj
3
V −V∞.
3.36
Hence,
K3V − K3V∞≤ 1 6
3Mα13V −V. 3.37
Using3.35, we get
Δ2K3V −Δ2K3Vp≤
1KT∞
n1
αpnK2V − K2V
n,Δ
K2V − K2V
npX×X1/p
≤33M2
1KT1 6
∞
j0
αj 3
V −V∞,
3.38 whence
Δ2K3V −Δ2K3Vp≤1 6
3Mα1
3
1KTM−1V −V. 3.39 From estimates3.37and3.39, we get
K3V − K3V≤ b
3!a3V −V. 3.40
An induction argument shows us that
KnV − KnV≤ b
n!anV −V. 3.41
Sinceban/n! < 1 fornsufficiently large, by the fixed point iteration methodKhas a unique fixed pointV ∈ W2,p0 . LetV be the unique fixed point ofK, then byAssumption 3.1we have
VnX
n−1 k1
Skf
n−1−k, Vn−1−k,ΔVn−1−k X
≤Mn−2
k0
f
k, Vk,ΔVk
−fk,0,0XMn−2
k0
fk,0,0X
≤Mn−2
k0
αkVk,ΔVkX×XMf·,0,0
1,
3.42
hence,
VnX ≤Mf·,0,0
1Mn−1
k0
αkVk,ΔVkX×X. 3.43 On the other hand, we have
ΔVnX
n−1
k1
Sk1− Sk f
n−1−k, Vn−1−k,ΔVn−1−k X
≤2Mn−1
k0
αkVk,ΔVkX×X2Mn−1
k0
fk,0,0X,
3.44
hence
ΔVnX ≤2Mf·,0,012Mn−1
k0
αkVk,ΔVkX×X. 3.45
From3.43and3.45, we get
Vn,ΔVnX×X ≤3Mf·,0,0
13Mn−1
k0
αkVk,ΔVkX×X. 3.46
T hen, by application of the discrete Gronwall inequality1, Corollary 4.12, page 183 , we get Vn,ΔVnX×X≤3Mf·,0,0
1
n−1 j0
13Mαj
≤3Mf·,0,0
1
n−1 j0
e3Mαj
3Mf·,0,0
1e3M
n−1
j0αj
≤3Mf·,0,0
1e3Mα1.
3.47
Then,
sup
n∈Z
Vn,ΔVnX×X
≤3Mf·,0,01e3Mα1. 3.48
Finally, by3.20we obtain
Δ2Vnf
n, Vn,ΔVn n−1
k1
I−TSkf
n−1−k, Vn−1−k,ΔVn−1−k
. 3.49
Hence, using the fact thatΔ2V0f0,0,0and proceeding analogously as in3.23, we get Δ2Vpf0,0,0pX∞
n1
Δ2VnpX1/p
≤f0,0,0X ∞
n1
Δ2VnpX1/p
≤f0,0,0X ∞
n1
fn, Vn,ΔVnpX1/p
KT∞
n1
fn, Vn,ΔVnpX1/p
≤2 ∞
n0
f
n, Vn,ΔVnpX1/p
KT∞
n0
f
n, Vn,ΔVnpX1/p
≤
2KT∞
n0
f
n, Vn,ΔVnX,
3.50 where, byAssumption 3.1and3.48,
∞ n0
f
n, Vn,ΔVnX≤∞
k0
αkVk,ΔVkX×Xf·,0,0
1
≤3Mα1f·,0,0
1e3Mα1f·,0,0
1
≤f·,0,01e6Mα1.
3.51
This ends the proof of the theorem.
In view ofTheorem 2.4, we obtain the following result valid on UMD spaces.
Corollary 3.4. LetX be a UMD space. Assume thatAssumption 3.1holds and supposeT ∈ BX is an analyticS-bounded operator such that the set{λ−12Rλ−12, I−T : |λ| 1, λ /1}is R-bounded. Then, there is a unique bounded solutionx xnof 3.1such thatΔ2xn∈lpZ, X.
Moreover, the a priori estimates3.6hold.
Example 3.5. Consider the semilinear problem
Δ2xn−I−Txnqnfxn, n∈Z, x0x10, 3.52
wheref is defined and satisfies a Lipschitz condition with constantLon a Hilbert spaceH.
In addition, supposeqn∈l1Z.Then,Assumption 3.1is satisfied. In our case, applying the preceding result, we obtain that ifT ∈ BHis an analyticS-bounded operator such that the set{λ−12Rλ−12, I−T:|λ|1, λ /1}is bounded, then there exists a unique bounded solutionx xnof3.52such thatΔ2xn∈lpZ, H.Moreover,
max
sup
n∈Z
xnHΔxnH
,Δ2xp
≤Cf0Hq1e6LMq1. 3.53
In particular, takingT Ithe identity operator, we obtain the following scalar result which complements those in the work of Drozdowicz and Popenda2 .
Corollary 3.6. Supposef is defined and satisfies a Lipschitz condition with constantLon a Hilbert spaceH. Letqn∈l1Z, H, then the equation
Δ2xnqnfxn 3.54
has a unique bounded solutionx xnsuch thatΔ2xn∈lpZ, Hand3.53holds.
We remark that the above result holds in the finite dimensional case where it is new and covers a wide range of difference equations.
4. A criterion for stability
The following result provides a new criterion to verify the stability of discrete semilinear sys- tems. Note that the characterization of maximal regularity is the key to give conditions based only on the data of a given system.
Theorem 4.1. LetXbe a UMD space. Assume thatAssumption 3.1holds and supposeT ∈ BXis analytic and 1∈ρT.Then, the system3.1is stable, that is the solutionxnof 3.1is such that xn→0 asn→ ∞.
Proof. It is assumed thatTis analyticwhich implies that the spectrum is contained in the unit disc and the point 1, see10 and that 1 is not in the spectrum, then in view of27, Proposition 3.6 , the set
λ−12R
λ−12, I−T
: |λ|1, λ /1
4.1 isR-bounded, becauseλ−12Rλ−12, I−Tis an analytic function in a neighborhood of the circle. TheS-boundedness assumption of the operatorT follows from maximal regularity and the fact thatI−T is invertible. In fact, we get the following estimate:
supn≥0Sn ≤ I−T−1KT. 4.2
By Corollary 3.4, there exists a unique bounded solution xn of 3.1 such that Δ2xn ∈ lpZ, X. Then,Δ2xn → 0 asn→ ∞.Next, observe thatAssumption 3.1and estimate3.10 imply
f
n, xn,ΔxnX≤f
n, xn,Δxn
−fn,0,0Xfn,0,0X
≤αnxn,ΔxnX×Xfn,0,0X
≤αnsupn∈Zxn,ΔxnX×Xfn,0,0X
≤3αnx∞fn,0,0X.
4.3
Sincef·,0,0∈l1Z, Xandαn∈l1Z, we obtain thatfn, xn,Δxn→0 asn→ ∞.
Then, the result follows from the fact that 1∈ρTand3.1.
From the point of view of applications, we specialize to Hilbert spaces. The following corollary provides easy-to-check conditions for stability.
Corollary 4.2. Let H be a Hilbert space. Let T ∈ BH such that T < 1. Suppose that Assumption 3.1holds inH. Then, the system3.1is stable.
Proof. First, we note that each Hilbert space is UMD, and then the concept ofR-boundedness and boundedness coincide; see7 . Since T < 1, we get that T is analytic and 1 ∈ ρT.
Furthermore, for|λ|1, λ /1,the inequality λ−12R
λ−12, I−T
λ−12 λλ−2
∞ n0
T λλ−2
n
≤ |λ−1|2
|λ−2| − T ≤ 4
1− T 4.4 shows that the set4.1is bounded .
Of course, the same result holds in the finite dimensional case.
5. Local perturbations
In the process of obtaining our next result, we will require the following assumption.
Assumption 5.1. The following conditions hold.
i∗The function fn, zis locally Lipschitz with respect toz ∈ X×X; that is for each positive numberR, for alln∈Z, andz, w∈X×X,zX×X≤R, wX×X≤R
fn, z−fn, wX≤ln, Rz−wX×X, 5.1 where:Z×0,∞→0,∞is a nondecreasing function with respect to the second variable.
ii∗There is a positive numberasuch that∞
n0n, a<∞.
iii∗f·,0,0∈1Z, X.
We need to introduce some basic notations. We denote byW2,pm the Banach space of all sequencesV Vnbelonging to∞Z, X,such thatVn0 if 0≤n≤m, andΔ2V ∈pZ, X equipped with the norm·. For λ > 0, denote byW2,pmλ the ballV ≤λinW2,pm. Our main result in this section is the following local version ofTheorem 3.3.
Theorem 5.2. Suppose that the following conditions are satisfied.
a∗Assumption 5.1holds.
b∗T is anS-bounded operator and it has discrete maximal regularity.
Then, there are a positive constantm∈Nand a unique bounded solutionx xnof 3.1for n≥msuch thatxn0 if 0≤n≤mand the sequenceΔ2xnbelongs topZ, X. Moreover, one has
x∞Δ2xp≤a, 5.2
whereais the constant of condition (ii)∗.
Proof. Letβ∈0,1/3. Usingiii∗andii∗,there aren1andn2inNsuch that M2KT∞
jn1
fj,0,0X≤βa, 5.3 T:β
M2KT∞
jn2
j, a<1
3, 5.4
whereM:supn∈ZSn.
LetV be a sequence in W2,pma/3 , withm max{n1, n2}. A short argument similar to 3.7and involvingAssumption 5.1shows that the sequence
gn:
⎧⎨
⎩
0, if 0≤n≤m,
fn, Vn,ΔVn, ifn > m, 5.5
belongs top. By the discrete maximal regularity, the Cauchy problem3.12withgndefined as in5.5has a unique solutionznsuch thatΔ2zn∈lpZ, X, which is given by
zn KV n
⎧⎪
⎪⎨
⎪⎪
⎩
0, if 0≤n≤m,
n−1−m
k0
Skfn−1−k, Vn−1−k,ΔVn−1−k, ifn≥m1. 5.6
We will prove thatKV belongs toW2,pma/3 . In fact, since
Vj,ΔVjX×X≤3|V|∞≤3|V|< a, 5.7
we have byAssumption 5.1 KV nXMn−2
jm
f
j, Vj,ΔVjX
≤Mn−2
jm
f
j, Vj,ΔVj
−fj,0,0XMn−2
jm
fj,0,0X
≤Mn−2
jm
lj, aVj,ΔVjX×XMn−2
jm
fj,0,0X
≤M∞
jm
lj, aaM∞
jm
fj,0,0X.
5.8
Proceeding in a way similar to3.20, we get forn≥m
Δ2KV nf
n, Vn,ΔVn n−1−m
k1
I−TSkf
n−1−k, Vn−1−k,ΔVn−1−k
. 5.9
Hence,
Δ2KV pf
m, Vm,ΔVmpX ∞
nm1
Δ2KV npX1/p
≤f
m, Vm,ΔVmX
∞
nm1
f
n, Vn,ΔVn n−1−m
k1
I−TSkf
n−1−k, Vn−1−k,ΔVn−1−kpX1/p
≤f
m, Vm,ΔVmX
1KT∞
nm
f
n, Vn,ΔVnpX1/p
≤
2KT∞
nm
f
n, Vn,ΔVnX.
5.10
Therefore, using5.8we get Δ2KV p≤
2KT∞
jm
lj, aa∞
jm
fj,0,0X
. 5.11
Then, inequalities5.8and5.11together with5.3and5.4imply KV≤
M2KT∞
jm
j, aa
M2KT∞
jm
fj,0,0X
≤ 1
3 −β
aβa 1 3a,
5.12
proving that KV belongs to W2,pma/3 . In an essentially similar way to the proof of Theorem 3.3, for allV andWinW2,pma/3 , we prove that
KV −KW ∞≤3M∞
jm
j, aV −W, 5.13
Δ2KV −Δ2KW p≤3
1KT∞
jm
j, aV −W, 5.14
whence
KV −KW ≤3
M1KT∞
jm
j, aV −W3T −βV −W. 5.15
Since 3T −β<1,Kis a 3T −β-contraction. This completes the proof of the theorem.
This enables us to prove, as an application, the following corollary.
Corollary 5.3. LetBi : X ×X → X,i 1,2,be two bounded bilinear operators,y ∈ 1Z, X, andα, β ∈ 1Z,R. In addition, suppose thatT is aS-bounded operator and has discrete maximal regularity. Then, there is a unique bounded solutionxsuch thatΔ2x∈lpZ, Xfor the equation
xn2−2xn1TxnynαnB1
xn, xn βnB2
Δxn,Δxn
. 5.16
Proof. Take ln, R : 2R|αn||βn|B1 B2.Then, ∞
n0n,1 < ∞.Note also that fn,0,0 ynbelongs to1Z, X.Hence,Assumption 5.1is satisfied.
Remark 5.4. We observe that under the hypotheses of the above local theorem and corollary, the same type of conclusions on stability of solutions proved inSection 4remains true.
Acknowledgments
The authors would like to thank the referees for the careful reading of the manuscript and their many useful comments and suggestions. The first author is partially supported by CNPq/Brazil under Grant no. 300068/2005-0. The second author is partially financed by Proyecto Anillo ACT-13 and CNPq/Brazil under Grant no. 300702/2007-08.
References
1 R. P. Agarwal, Difference Equations and Inequalities, vol. 155 of Monographs and Textbooks in Pure and Applied Mathematics, Marcel Dekker, New York, NY, USA, 1992.
2 A. Drozdowicz and J. Popenda, “Asymptotic behavior of the solutions of the second order difference equation,” Proceedings of the American Mathematical Society, vol. 99, no. 1, pp. 135–140, 1987.
3 S. Elaydi, An Introduction to Difference Equations, Undergraduate Texts in Mathematics, Springer, New York, NY, USA, 3rd edition, 2005.
4 V. B. Kolmanovskii, E. Castellanos-Velasco, and J. A. Torres-Mu ˜noz, “A survey: stability and bound- edness of Volterra difference equations,” Nonlinear Analysis: Theory, Methods & Applications, vol. 53, no. 7-8, pp. 861–928, 2003.
5 J. P. LaSalle, The Stability and Control of Discrete Processes, vol. 62 of Applied Mathematical Sciences, Springer, New York, NY, USA, 1986.