Research Article | Open Access
Dichotomies with No Invariant Unstable Manifolds for Autonomous Equations
We analyze the existence of (no past) exponential dichotomies for a well-posed autonomous differential equation (that generates a C0-semigroup ). The novelty of our approach consists in the fact that we do not assume the T(t)-invariance of the unstable manifolds. Roughly speaking, we prove that if the solution of the corresponding inhomogeneous difference equation belongs to any sequence space (on which the right shift is an isometry) for every inhomogeneity from the same class of sequence spaces, then the continuous-time solutions of the autonomous homogeneous differential equation will exhibit a (no past) exponential dichotomic behavior. This approach has many advantages among which we emphasize on the facts that the aforementioned condition is very general (since the class of sequence spaces that we use includes almost all the known sequence spaces, as the classical spaces, sequence Orlicz spaces, etc.) and that from discrete-time conditions we get information about the continuous-time behavior of the solutions.
The exponential dichotomy is one of the most basic concepts arising in the theory of dynamical systems. For linear differential equations, the notion was introduced by Perron in , who was concerned with the problem of conditional stability of a system and its connection with the existence of bounded solutions of the equation , where the state space is a finite-dimensional Banach space and the operator-valued function is bounded and continuous in the strong operator topology. Relevant results concerning the extension of Perron’s problem in the more general framework of infinite-dimensional Banach spaces were obtained by Daleckij and Krein , Bellman , Massera and Schäffer , and more recently by van Neerven , and van Minh et al. .
For the case of discrete-time systems, analogous results were firstly obtained by Li in . In his paper, we remark the same central concern as in Perron’s work, but in other terms. In fact it was proposed that the inhomogeneous equation is responsible in some sense for the asymptotic behavior of solutions of the homogeneous equation. In this spirit, there were established connections between the condition that the inhomogeneous equation has some bounded solution for every bounded “second member” on the one hand and a certain form of conditional stability of the solutions of the homogeneous equation on the other. This idea was later extensively developed for discrete-time systems in the infinite-dimensional case by Coffman and Schäffer  and Henry . More recently, we have the papers of Ben-Artzi and Gohberg , Pinto , and LaSalle . Applications of this “discrete-time theory” to stability theory of linear continuous-time systems in infinite-dimensional spaces have been presented by Przyluski and Rolewicz in .
The dichotomy of the autonomous equation , (assuming its well-posedness, i.e., generates a -semigroup ) consists in the existence of a bounded projection such that the solutions that start in decay to zero and the solutions that start in are unbounded. In the hypothesis that is -invariant and finite-dimensional, the existence of a dichotomy for implies that solutions starting in exist in backward time (or equivalently, extends to a -group on ). This observation is not necessarily true in the infinite-dimensional setting, but required in many researches (see, e.g., [6, 14, 15]). The novelty of our approach consists in the fact that we do not assume a priori that the operators are invertible (we do not even assume that is -invariant), and subsequently the unstable subspace is allowed to be infinite-dimensional. Roughly speaking, we prove that if the solution of the corresponding inhomogeneous difference equation belongs to any sequence space (on which the right shift is an isometry) for every inhomogeneity from the same class of sequence spaces, then the continuous-time solutions of the autonomous homogeneous differential equation will exhibit a (no past) exponential dichotomic behavior. This approach has many advantages among which we emphasize on the facts that the previous condition is very general (since the class of sequence spaces that we use includes almost all the known sequence spaces, as the classical spaces, sequence Orlicz spaces, etc.), and since we use a discrete-time technique, we are not forced to require any continuity or measurability hypotheses on the trajectories of the one-parameter semigroup generated by the differential system. Also, it is worth to mention that from discrete-time conditions we get information about the continuous-time behavior of the solutions.
2. Sequence Schäffer Spaces
Let be the set of all nonnegative integers, , the set of all real numbers, and we will denote by the greatest integer less than or equal with . The linear space of all real-valued sequences is denoted by . Let be a real or complex Banach space and consider the linear space of all sequences . For an -valued sequence , we can associate the sequence defining for all . We also consider two linear operators defined by known as the right shift operator, respectively, the left shift operator. A simple verification gives us and for , , for all . If , the characteristic function of will be denoted by and for the simplicity of notation put for each .
Definition 2.1. A Banach space is said to be a sequence Schäffer space if and the following conditions hold:,if , then and ,if and such that , then and .
Remark 2.2. By and we have that any sequence with finite support is contained in any sequence Schäffer space, hence for any sequence Schäffer space and . The third property is called the ideal property and will play a central role in our investigations.
Example 2.3. Common instances of sequence Schäffer spaces are the spaces of -summable sequences, namely, for ,
The subspace of , (often denoted by ) with the induced norm is another example of sequence Schäffer space.
It is easy to check that (the space of all convergent sequences) is not a sequence Schäffer space.
The spaces , , and occupy particularly important positions in the class of sequence Schäffer spaces. For a sequence Schäffer space, we will define the sequences by which are both nondecreasing and , for all .
Example 2.4. Other remarkable examples of sequence Schäffer spaces are the sequence Orlicz spaces. Let be a left continuous, nondecreasing function and not identically 0 or on . The Young function attached to is defined by . Consider
For the Banach space the conditions , , and are verified; hence is a sequence Schäffer space.
For , taking we have that . Even is a sequence Orlicz space, obtained from for and for [16, Section 3].
Remark 2.5. Let . If , then .
Indeed, if , then , for all , which is equivalent with Let and . Using the fact that is nondecreasing, we have that which implies that for all . Hence and
Example 2.6. Consider :
We claim that , no matter how we choose .
Indeed, and , for all , and using the aforementioned remark, our claim follows easilly. Also, since .
Remark 2.7. By simple computations we obtain that
for (with the convention ) and for the Orlicz sequence spaces,
For two Banach spaces and , we say that is continuously embedded in () if and there exists such that for all . For the following three propositions, proofs can be retrieved from [8, Section 3].
Proposition 2.8. If is a sequence Schäffer space, then with for all and for all .
Proposition 2.9. If and are sequence Schäffer spaces, then if and only if .
Proposition 2.10. Let be a sequence Schäffer space. The following characterizations hold:(i) is bounded if and only if and ; (ii) is bounded if and only if .
For being a sequence Schäffer space and being a Banach space, we consider and . To prove that is a Banach space see, for example, [17, Remark 2.1] or [8, Lemma 3.8]. The following properties of this space are simple verifications.
Proposition 2.11. The space is a Banach space with the following properties:(i)if has finite support, then ;(ii)if , then and ;(iii)if and such that , then and .
To prevent any further confusion, let us fix the notation for the class of bounded linear operators acting on .
3. Semigroups of Bounded Linear Operators, Exponential Dichotomy, and Admissibility
Let be a Banach space, , a linear operator, and consider the (abstract) Cauchy problem : The function is said to be a classical solution for the previous Cauchy problem if it is continuously differentiable on and satisfies . A theorem of E. Hille states that if the resolvent set of is nonempty, then for every the Cauchy problem has a unique classical solution if and only if is the infinitesimal generator of a -semigroup. For the proof of this result we guide the reader to [18, Theorem 6.7, page 150] or [19, Theorem 1.3, page 102].
Definition 3.1. A family of bounded linear operators acting on is called a semigroup if (where is the identity operator on ), for all .If (ii) holds for any , then is called a group. If in addition of (i) and (ii), there exist such that for all ,then is said to be exponentially bounded.
If is a closed subspace of and is a (-)semigroup such that is -invariant for any , then the restrictions form a (-)semigroup called the subspace (-)semigroup on ([18, page 43]). A (-)semigroup can be extended to a (-)group if and only if there exists such that is invertible (see [18, page 80] or [19, Section 1.6]).
The linear operator defined by is the infinitesimal generator of the semigroup . If is a -semigroup, then is closed and densely defined and is exponentially bounded (for details, see [19, Section 1.2]). It is clear that a classical solution for with being the infinitesimal generator of exists only if in which case is the unique classical solution; otherwise it is said to be the mild solution of .
A linear and bounded operator acting on a complex Banach space is said to be hyperbolic if (where denotes the unit circle in the complex plane and is the spectrum of ). The spectral Riesz projection for a hyperbolic operator is given by The projection corresponds to the part of the spectrum of contained in the open unit disk . We note that the projection commutes with . Since we obtain the spectral radius and also the operator is invertible with . Hence, if is hyperbolic, then there exist the constants such that, for all integers , (for the previous exposure we consulted [14, page 28]).
Definition 3.2. A semigroup is hyperbolic if there exists such that is an hyperbolic operator.
Definition 3.3. The semigroup has an exponential dichotomy (or that it is exponentially dichotomic) if there exist a projection (i.e., and ) and the constants such that (i), for all , (ii) is an isomorphism, for each , (iii), for all and , (iv), for all and .
The first condition in the previous definition expresses equivalently that and are both -invariant; the essence of (ii) is that can be extended to a group. The next result establishes the relation between hyperbolicity and exponential dichotomy for -semigroups on complex Banach spaces. For the proof we refer the reader to [14, Lemma 2.15], [20, Theorem 1.1], or alternatively .
Proposition 3.4. For a strongly continuous semigroup acting on a complex Banach space the following statements are equivalent. (i) is hyperbolic. (ii) has an exponential dichotomy. Moreover, if holds, then where , , and is the spectral Riesz projection for that corresponds to . Also, if has an exponential dichotomy, then , for every .
If generates an exponentially dichotomic -semigroup, then the differential equation has the property that the solutions starting from (resp., from ) decay exponentially for (resp., for ) uniformly with respect to the initial data. As it can be seen, the exponential dichotomy concept generalizes strongly the exponential stability concept but it has a serious drawback. It forces the solution that starts from to exist for negative time, or in counterpart it forces the semigroup to be invertible on . We will drop off this requirement here and extend the notion of hyperbolicity by replacing the exponential decay in negative time for the solutions starting in with an exponential blow-up in positive time. We will call the “exponential decay on and exponential blow-up on ” (both on positive time) behavior as no past exponential dichotomy.
Definition 3.5. The semigroup has a no past exponential dichotomy if there exist a projection and the constants such that (i), for all , (ii), for all , , (iii), for all , .
Definition 3.6. The semigroup has an ordinary dichotomy if there exist a projection and the constants such that (i), for all , (ii), for all , , (iii), for all , .
Remark 3.7. Note that if is one-to-one, then and thus the concept of (no past) exponential dichotomy overlaps the concept of exponential stability. Recall that is said to be exponentially stable if one of the following equivalent statements is true: (i)there exist such that for all ; (ii)there exist such that .
It is obvious that the existence of an exponential dichotomy implies the existence of a no past exponential dichotomy, but the converse is not valid as the following example points out.
Example 3.8. The following -semigroup has a no past exponential dichotomy but is not exponentially dichotomic. Let be a matrix with real entries whose spectrum is contained in the open left-half plane and consider the right shift semigroup on , given by Then has a no past exponential dichotomy on with and . However, the restriction of on is not onto, and thus is not exponentially dichotomic.
Lemma 3.9. Let be an exponentially bounded semigroup. If there exist , a vector subspace of , , , and such that (i), for all and , (ii), for all and , then there exist which satisfy , for all and .
Proof. Let , and . Then, there exist and such that . Therefore, . We take to obtain For , take to have . Put in order to write .
Lemma 3.10. Let be an exponentially bounded semigroup. If there exist a vector subspace of , , , and such that (i), for all and and (ii), for all and , then there exist which satisfy , for all and .
Proof. It is analogous with the proof of Lemma 3.9.
Consider the autonomous inhomogeneous abstract Cauchy problem : If generates the -semigroup , and , then the function given by is said to be the mild solution of the Cauchy problem . For sufficient conditions assuring that the mild solution is also a classical solution (i.e., a continuously differentiable function that verifies the initial value problem ), we refer the reader to [19, Section 4.2].
The “test functions method” (or “Perron's method”) was often used until now (see e.g., [1–4]) to study properties of asymptotic behavior such as exponential dichotomy. According to Massera and Schäffer  by “test functions method” it is meant the relation between certain “test functions” and “nice solutions” of the inhomogeneous equations : . The crudest expression of this method is the notion of admissibility of a pair of classes of functions both in (Massera and Schäffer named these classes of functions as the class of “test functions” and the class of “nice solutions” and defined the pair to be admissible if for every “test function” , the equation has a “nice solution” (see [4, Chapter 5, page 124]).
In this spirit, we set the expression of the mild solution of the equation in discrete-time to give the following definition of admissibility in terms of “test sequences” and “nice discrete-time mild solutions”. In this way, we do not need any assumption of continuity or measurability and we still obtain continuous-time asymptotic properties for the autonomous differential equation : .
Definition 3.11. Let be sequence Schäffer spaces. The pair is said to be admissible to if for each , there exists such that , where for each .
4. Main Results
In this section, for being a semigroup on the Banach space (with assuring the exponential boundedness if it is the case for to be exponentially bounded) and and being two sequence Schäffer spaces, we denote which are obviously vector subspaces of .
Hypothesis 1. The vector subspace is closed and admits a closed complement; that is, there exists , a closed vector subspace, such that
We denote by the projection onto along and set (we will prove that in the case of a no past exponential dichotomy for , coincides always with and thus the -independent notation for projectors is consistent).
Remark 4.1. We have that and , for each and .
Proof. If and , then it is to see that for all , and since , it follows that .
For the second part, assume for a contradiction that there exist and such that . Then, for every , and thus . It follows that , which is not possible since . Thus, is one-to-one, for all .
Proposition 4.2. If the pair is admissible to the semigroup , then for each , there exists unique such that .
Proof. Let and from Definition 3.11. For we have that and . Since and , it follows that .
To prove the uniqueness of , suppose that there exists with the property . Since , we have that and therefore .
The unique vector will be denoted by .
Proposition 4.3. If the pair is admissible to the semigroup , there exists such that for all .
Proof. We define the operator
It is obvious that is a linear operator. Now, we will show that it is also closed.
Let such that and where , , and . For each , we take and . We have that and .
Since we have that , for all . On the other hand, which implies , for all . It follows that and by Proposition 4.2 we have . Therefore .
Hence, is a closed linear operator, and by the Closed-Graph Theorem it is also bounded which means that there exists such that and the proof is complete.
Remark 4.4. If is admissible to the semigroup , then is admissible to and , for all and .
With this intermediate result we are able to prove that the admissibility of to an exponentially bounded semigroup is a sufficient condition for a no past exponential dichotomy of . The restriction over such a pair is that and are not simultaneously the bounds of the chain of sequence Schäffer spaces (in the sense of Proposition 2.8).
Theorem 4.5. Let and be sequence Schäffer spaces such that or and let be an exponentially bounded semigroup. If is admissible to , then it has a no past exponential dichotomy and .
Proof (Part I). The exponential decay of on .
Let and consider the sequence which is in with . Observe that , for all . Then, and from we obtain , for all . For , taking we have Since the constant does not depend on , we can write down For and , we evaluate which implies and therefore By Proposition 4.3 we get that
If , then is not bounded and therefore there exists such that and , for all .
If , then . From Proposition 2.8, it follows that there exists . Consider which is nondecreasing and . For and , the sequence has finite support, and thus . Observing that for every , we are led to the evaluation . Also, and thus we have , for all . Since , , we have that . Taking in , we can deduce that It follows that there exists such that and , for all .
In both cases, we obtained the existence of some and some constant such that for all . Therefore, the semigroup is exponentially stable (see Remark 3.7). Subsequently, there exist such that
Proof (Part II). The exponential blow-up of on .
Let , and , and consider the sequence We have that with . On the one hand, for all (where ), while on the other hand which implies that . Then, and Taking now we can write down For , we put to evaluate . Denoting we have that Also, from (4.23), taking we have that Since , and are randomly taken (see the beginning of Part II), we have that and therefore . It follows that or equivalently, Using the fact that (4.29), we deduce that
If , then is not bounded and therefore there exists such that and for all and .
If , then and therefore there exists . Consider as in (4.15), and for and we define Since for each , we have that with . Note that , for all . On the one hand, for all (where ), while on the other hand From (4.32) and (4.33), it follows that has finite support, and thus it belongs to . Therefore, and using that we obtain that Thus, , for all and . Then, there exist and such that
In both cases, we obtained the existence of some and some constant satisfying the condition from Lemma 3.9, while the first condition is assured by (4.25). Subsequently, there exist such that
If we take , all the conditions guaranteeing the existence of a no past exponential dichotomy are met.
Proof (Part III). We prove that , no matter how we choose the sequence Schäffer space . If , then , for all , which implies .
Conversely, let , , and such that . For every , we have that If we suppose that , then contradicting the fact that . It follows immediately that and .
Remark 4.6. As we pointed out in the introduction, there is an extensive literature on the connection between admissibility and hyperbolicity (or equivalently, exponential dichotomy). Latest there is known the equivalence between the admissibility of the pair and and the hyperbolicity of a -semigroup , when we assume a priori that the kernel of the splitting projection is -invariant and are invertible. For details we refer the reader to . We try to extend this line of results in two directions. First, we do not assume a priori that is invertible (we do not even assume that is -invariant) and still we succeed to prove that the admissibility of any pair of sequence Schäffer spaces implies the existence of a no past exponential dichotomy. Secondly, it is worth to note that the class of sequence Schäffer spaces is extremely reachable (see e.g., Examples 2.3 and 2.4) and this fact allows the reader to choose the test sequences” in various ways and in the same time it does not force the “output” or “nice discrete-time mild solutions” (i.e., the solution of the inhomogeneous difference equation problem) to stay in , as before. Moreover, this approach can provide interesting input spaces (i.e., the spaces consisting in "test sequences") which are different from the classical spaces (we refer the reader to Example 2.6). Also, it is worth to note that if there exists a pair of vector-valued sequence Schäffer spaces , which is admissible to , and with the property that or , then the subspace (which induces the no past exponential dichotomy) is actually the regular stable subspace . If we would impose in addition that the complement of (denoted by ) is also -invariant, then the aforementioned admissibility condition would imply that extends automatically to a -group on , and thus we would get hyperbolicity for (see Theorem 4.12 below). Therefore, we can conclude that “admissibility” converts to “no past exponential dichotomy”“admissibility” and “-invariance of ” converts to “hyperbolicity”.
The next example shows that the condition “ or ” in the statement of Theorem 4.5 is essential.
Example 4.7. Let and consider the semigroup by for all . If , then there exists (the series being absolutely convergent) such that . Therefore, the pair is admissible to , but one can easily check that does not posses a no past exponential dichotomy.
The above theorem shows that the space induces a no past exponential dichotomy for —an exponentially bounded semigroup—and that is actually . Concerning the hypothesis over and from the Theorem 4.5 we make one last remark.
Remark 4.8. In  there was introduced the order relation between pairs of Banach spaces: is said to be stronger than (or the later pair is weaker than the former) if and . In the case of sequence Schäffer spaces the two inclusions with continuous injection are equivalent with the corresponding algebraic inclusion (Proposition 2.9); so the pair is stronger than if and , or equivalently (considering Proposition 2.8), and . Therefore, the hypothesis “ or ” can be replaced by, “the pair is not stronger than the pair ”.
Dropping off the restriction “