Abstract
We provide new variational settings to study the a.p. (almost periodic) solutions of a class of nonlinear neutral delay equations. We extend Shu and Xu (2006) variational setting for periodic solutions of nonlinear neutral delay equation to the almost periodic settings. We obtain results on the structure of the set of the a.p. solutions, results of existence of a.p. solutions, results of existence of a.p. solutions, and also a density result for the forced equations.
1. Introduction
The aim of this paper is the study of a.p. (almost periodic) solutions of neutral delay equations of the following form: where is a differentiable function, denotes the partial differential with respect to the th vector variable, and is fixed. We will consider the almost periodicity in the sense of Corduneau [1], and in the sense of Besicovitch [2].
A special case of (1.1) is the following forced neutral delay equation: where is a differentiable function, and is an a.p. forcing term. To see (1.2) as a special case of (1.1), it suffices to take where the point denotes the usual inner product in .
Another special case of (1.1) is the following forced second-order neutral delay equation: where and . To see that this last equation is a special case of (1.1), it suffices to take , where the norm is the usual Euclidian norm of . In their work [3], Shu and Xu study the periodic solutions of this last equation by using a variational method. We want to extend such a view point to the study of the a.p. solutions.
And so our approach to the study of the a.p. solutions of (1.1) consists to search critical points of a functional defined on suitable Banach spaces of a.p. functions by At this time, we give some historical elements. Recall that the work [4] of Elsgolc treats the calculus of variations with a retarded argument on a bounded real interval. This work was followed by these ones of Hughes [5] and Sabbagh [6]. Since the variational problems can be seen as optimal control problems, recall also the existence of the theory of the periodic optimal control with retarded argument as developed by Colonius [7]. For instance, we consider a periodic optimal control problem with a criterion of the form and with an equation of motion of the form , where is the state variable and the control variable. In the special case, where , the previous optimal control problem can be transformed into a calculus of variations problem with the criterion , which is a special case of (1.5). Note that the Euler-Lagrange equation of such a variational problem is a special case of (1.1).
On another hand, calculus of variations in mean time was developed to study the a.p. solutions of some (nonretarded) differential equations [8β13]. Here, we extend this approach to treat equation like (1.1).
Now we describe the contents of this paper. In Section 2, we precise the notations about the function spaces used later. In Section 3, we establish a variational formalism suitable to the Bohr a.p. solutions; we give a variational principle and a result on the structure of the set of the a.p. solutions of (1.1) in the convex case. In Section 4, we establish a variational formalism suitable to the Besicovitch a.p. solutions, we give a variational principle, results of existence, and a result of density for the a.p. forced equations.
2. Notations
is the space of the Bohr almost periodic (Bohr a.p.) functions from in ; endowed with the supremum , it is a Banach space [1].
; endowed with the norm , it is a Banach space.
When ,
When , its mean value is
The Fourier-Bohr coefficients of are the complex vectors and
When is the completion of (in ) with respect to the norm . When , is a Hilbert spaces and its norm is associated to the inner product [2]. The elements of these spaces are called Besicovitch almost periodic (Besicovitch a.p.) functions.
Recall the useful following fact: if is a sequence in and if (Lebesgue space), which satisfy then and we have .
We use the generalized derivative of (when it exists) defined by , and we define ; endowed with the inner product , is a Hilbert space [11, 13].
If and are two finite-dimensional-normed spaces, stands for the space of the functions , which are almost periodic in uniformly with respect to in the classical sense given in [14].
To make the writing less heavy, we sometimes use the notations when , and when .
3. A Variational Setting for the Bohr a.p. Functions
We consider the following condition:
Lemma 3.1. Under [9] the functional defined by (1.5) which is of class , and for all then
Proof.
We
introduce the linear operator by setting The four
components of are continuous
linear operators that imply the continuity of , and therefore is of class , and for all we have .
Under (3.1), the Nemytski operator , defined by , is of class (cf.
[15, Lemma 7]) and we have, for all , .
The linear functional is continuous,
therefore it is of class and we have,
for all , .
And so is of class . Furthermore, we have and expressing in terms of , we obtain the announced formula.
Note that in the case without delay, when is autonomous, that is, , in [9], it is established that the functional is of class when is of class . In [16], we can find a proof of the differentiability of the Nemytski operator on which is different to this one of [9].
Theorem 3.2 (variational principle). Under (3.1), for , the following two assertions are equivalent.
(i), that is, is a critical
point of in .(ii) is a Bohr a.p.
solution of (1.1).
Proof. First
we assume (i). Since the mean value is translation invariant, we have and so by using
Lemma 3.1 we obtain, for all ,
Setting , denoting by its coordinates
for , setting , and denoting by its coordinates
for , we deduce from the previous equality that, for all we have . Then by reasoning like
in the proof of Theorem 1 in
[8], we obtain that in the sense of
the a.p. distributions of Schwartz [17], and by using the proposition of the
Fourier-Bohr series, we obtain that is and that in the ordinary
sense. From this, we obtain that is and that which is
exactly (ii).
Conversely by using the formula , for all and , and by translating
the time, we obtain from (ii),
for all the following
relation: and so we have
(i).
Theorem 3.2 is an extension to the nonautonomous case in presence of a delay of [8, Theorem 1]. Now we use Theorem 3.2 to provide some results on the structure of the set of the Bohr a.p. solutions of (1.1) in the case, where is autonomous and convex.
Theorem 3.3 (structure result).
Assume that , and that is convex.
Then the following assertions
hold.
(i) The set of the
Bohr a.p. solutions of (1.1) is a closed convex subset of .(ii) If is a -periodic
nonconstant solution of (1.1), if is a -periodic
nonconstant solution of (1.1), and if is no rational,
then is a Bohr a.p.
nonperiodic solution of (1.1), for all .(iii) If is a Bohr a.p.
solution of (1.1), then is a constant
solution of (1.1).(iv) If is a Bohr a.p.
solution of (1.1), if is such that , then there exists a nonconstant -periodic
solution of (1.1).
Proof. Since is convex, the
functional of (1.5) is
also convex on . Since is autonomous
and of class , satisfies
(3.1), and so is of class . Therefore, we have which is closed
and convex, and (i) becomes a consequence of Theorem 3.2. The assertion (ii) is
a straightforward consequence of (i).
We introduce , when is a Bohr
a.p. solution of (1.1), for all . By using a theorem of Besicovitch
(see [2, page 144]),
there exists a -periodic
continuous function, denoted by , such that
We easily verify that
Since is autonomous, is a Bohr a.p.
solution of (1.1). Since is a convex
combination of Bohr a.p. solutions of (1.1), is a Bohr a.p.
solution of (1.1), and also by using
the closeness of the set of Bohr a.p. solutions. And so is a -periodic
solution of (1.1). By using a straightforward calculation, we see that and
consequently . When , then is not constant
that proves (iv).
To prove (iii) it suffices to choose such that , and then all the Fourier-Bohr coefficients of are zero except
(perhaps) the mean value of which is equal
to .
The assertions (i) and (ii) are extensions of [8, Theorems 3 and 4]; the assertions (iii) and (iv) are extension to neutral delay equations of [10, Theorem 2].
The space does not possess good topological properties like to be a reflexive space. It is why in the following section we extend our variational formalism to the Hilbert space .
4. A Variational Setting for the Besicovitch a.p. Functions
and are Euclidean finite-dimensional spaces.
Lemma 4.1. Let be a function which satisfies the following HΓΆlder condition: Let be such that .
Then the following two assertions hold.
(i)If then .(ii)The
Nemytski operator on , defined by
satisfies ,
for all .
Proof. We set , and so we have and the HΓΆlder assumption implies for all . If then we have for all , and since is continuous, we have (the Lebesgue space), and since , we have Since , there exists a sequence in such that . By using [14, Theorem 2.7 page 16], setting , we have , and a straightforward calculation gives us the following inequality: and consequently we obtain that implies , and so (i) is proven; moreover the last inequality becomes the one of (ii) when we replace by .
This lemma is an extension to the nonautonomous case of [13, Theorem 1].
Lemma 4.2. Let be a function
such that the partial differential exists, for all , such that . We assume the following condition fulfilled.
(C) There exist , such that, for all , and for all ,
Then the Nemytski
operator , defined by , is of class and, for all
Proof. First
step: we show that there exist , , such that, for all , . The following hold:
By using the
mean value theorem (see [18, page 144]), we have, for all , Setting , then . Since , and , we have .
Second step: we show that when .
Let . Then the inequality implies that By using Lemma 4.1 with , and , we have . Let be a sequence
in such that . By using the mean value theorem (see [18, page 144]), we
have, for all , and consequently we obtain Since and since , we have By using
(see [14, Theorem 2.7 page 16] ), we have and so, by setting we have . The last inequality implies and therefore we have .
Third step:
we show that, for all , the operator , defined by , is linear continuous. We have yet seen that . The linearity of is easy to
verify. By using a Cauchy-Schwartz-Bunyakovsky
inequality, we have that proves the
continuity of .
Fourth step: we show the differentiability of .
Let and . By using the mean value inequality (see [18, page 144]),
we have, for all , and by using the monotonicity of , we obtain that is, that implies that is differentiable
at and that .
Fifth step:
we show that is of class .
Let . By using (C), for all , such that , for all , we have That implies, by using the Cauchy-Schwartz-Bunyakovsky
inequality, the following majorization holds: Therefore we have that implies
the continuity of .
Note that Lemma 4.2 is an extension to the nonautonomous case of [13, Theorem 2].
Theorem 4.3 (variational principle). Let , be a function and let . Assume the following conditions fulfilled: Then the
functional , defined by is of class , and the two following assertions are
equivalent.
(i), that is, is a critical
point of .(ii)
Definition 4.4. When satisfies the equation of (ii) in Theorem 4.3, we say that is a weak Besicovitch a.p. solution of (1.1).
Proof.
We
consider the operator , defined by . is clearly
linear continuous, therefore is of class and we have .
We consider the Nemytski operator By using
Lemma 4.2, is of class and, for all we have The mean value is linear
continuous, therefore it is of class , and , for all .
Consequently is of class as a
composition of three mappings of class .
Let . If (i) is true then, for all , we have and then we
obtain (ii) by using [13, Proposition 10].
Conversely, if (ii) is true, then , and for all we have therefore by using
[13, Proposition 9], we obtain Since is dense in , we have , for all , therefore .
Note that the Theorem 4.3 is an extension to the nonautonomous case of [13, Theorem 4].
Theorem 4.5 (existence, uniqueness). Let be a function
which satisfies (4.21) and also satisfies the following two conditions: Then there
exists a function which is a weak
Besicovitch a.p. solution of (1.1).
Moreover, if in addition the following
condition fulfilled: then the weak
Besicovitch a.p. solution of (1.1) is unique.
Proof.
By
using Theorem 4.3, the functional is of class and, by using
(4.28), is a convex functional.
Assumption (4.29) ensures that, for all , we have Since the mean value is translation invariant, consequently is coercive on , and so (see [19, page 46]) there exists such that . Therefore we have and by using
Theorem 4.3, is a weak
Besicovitch a.p. solution of (1.1). The existence is proven.
To treat the uniqueness, we note that, under (4.30),
the functional , defined by , is convex and since is of class , is also of
class . Note that we have . By using the
Mint monotonicity of the differential of a
convex functional, for all we have Now if and are two weak
Besicovitch a.p. solutions of (1.1),
by using Theorem 4.3 we have , and consequently , therefore .
Theorem 4.6 (existence and density). Let be a function
which satisfies the following conditions: Then the
following conclusions hold.
(i) For all , there exists a unique which is a weak
Besicovitch a.p. solution of (1.2).(ii) The set of the for which there
exists a Bohr a.p. solution of (1.2) is dense in with respect to
the norm
Proof. We
introduce the functionals and from in setting and . They are special cases of the functional of the Theorem 4.3, and consequently they are of class . Note that . By using the F. Riesz isomorphism , , for all , we can define the gradients and . By using the Minty-monotonicity of (due to the
convexity of ) we have, for
all , that implies
that is strongly
monotone and consequently (see [20, page 100]) the following property holds:
From each we define the linear functional by setting
Therefore we have and by using
(4.36), there exists such that , that is, which means
that, for all , and by using
[13, Proposition 10], we obtain that is a weak
Besicovitch a.p. solution of (1.2).
About the uniqueness, note that if is a weak
Besicovitch a.p. solution of (1.2), then we verify that , for all , and consequently , that is, , and by using
(4.36), we have . And so (i) is proven.
Now we introduce the nonlinear unbounded operator defined by And so means that is a weak
Besicovitch a.p. solution of (1.2). By using the assertion (i), is bijective.
We verify that for all , and by using (4.36) we see that is a
homeomorphism from on . Since is dense in , is dense in with respect to
the norm , and since , we have proven (ii).
This result is an extension to the neutral delay equations of [13, Theorem 5].
Acknowledgment
The authors thank an anonymous referee for several remarks which have permitted them to improve Theorems 4.3 and 4.5.