Abstract
The p-Laplacian problem with the exponent of nonlinearity p depending on the solution u itself is considered in this work. Both situations when p(u) is a local quantity or when p(u) is nonlocal are studied here. For the associated boundary-value local problem, we prove the existence of weak solutions by using a singular perturbation technique. We also prove the existence of weak solutions to the nonlocal version of the associated boundary-value problem. The issue of uniqueness for these problems is addressed in this work as well, in particular by working out the uniqueness for a one dimensional local problem and by showing that the uniqueness is easily lost in the nonlocal problem.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Let \(\Omega \) be a bounded domain of \(\mathbb {R}^d\), \(d\ge 2\), with its boundary denoted by \(\partial \Omega \). We consider the problem
where f is a given data and
is the exponent function of nonlinearity.
Problems of the type (1.1) appear in the applications of some numerical techniques for the total variation image restoration method that have been used in some restoration problems of mathematical image processing and computer vision [3, 4, 14]. Variational models exhibit the solution of these problems as minimizers of appropriately chosen functionals and the minimization technique of such models involves the solution of nonlinear partial differential equations derived as necessary optimality conditions [13]. Several numerical examples suggesting that the consideration of exponents \(p=p(u)\) preserves the edges and reduces the noise of the restored images u are presented in [14, Section 8]. A related, although far more complicated, minimization problem with the exponent of the regularization term depending on the gradient of the reconstructed image u, was considered in the works [3, 4]. In particular, the pioneer work [3, Section 3.3] presents a numerical example suggesting a reduction of noise in the restored images u when the exponent of the regularization term is \(p=p(|\varvec{\nabla }u|)\).
Along with problem (1.1), we consider also in this work its nonlocal version,
where f is again a given data and
are the functions involved in the exponent of nonlinearity, for some constant exponent \(\alpha \) such that \(1<\alpha <\infty \). In this case, suitable examples for the mapping b in (1.5) are for instance
or else
where \(\Vert \cdot \Vert _r\) denotes the usual \(L^r(\Omega )\)-norm. Problem (1.1) is the natural extension of the p(x)-Laplacian problem introduced by Zhikov [15] and for which the revival of interest in the almost last two decades came from modelling applications such as thermo or electro-rheological fluids [2, 11] and image restoration [10]. For the p(x)-Laplacian problem several issues of existence, uniqueness and regularity were already addressed in many works and by different authors (see again [2, 10, 11] and the references cited therein). However for the p(u)-Laplacian problem (1.1), and to our best knowledge, the only work is due to Andreianov et al. [1], where the considered prototype problem is
As pointed out in [1], the main difficulty in the analysis of these p(u)-problems relies in the fact that neither the weak formulation of (1.1) nor (1.6) can be written as equalities in terms of duality in fixed Banach spaces. In particular, sequences of solutions \(u_n\) to these problems correspond to different exponents \(p_n\) and therefore belong to possible distinct Sobolev spaces. In the proof of [1, Theorem 2.8], the authors could not have used the abstract Minty argument on such a sequence of solutions, preferring instead a tool that pulls everything down to the \(L^1\) space. Then they have used the description of weakly \(L^1\) convergent sequences in terms of Young measures and their reduction using the monotonicity of the nonlinearity \(\xi \mapsto |\xi |^{p-2}\xi \). We prove, in Theorem 5.1, the existence of weak solutions to the problem (1.1) by using the Minty trick together with the technique of Zhikov [16] for passing to the limit in our sequence of \(p(u_n)\)-Laplacian problems. For the sake of completeness we give an elementary proof, in Lemma 3.1, of a version of [16, Lemma 3.3] that is suitable to be applied in our problem. The technique we use here to prove the existence result for the problem (1.1) is rather simpler and more general than the one used in [1]. By assuming that \(\partial \Omega \) belongs to some Hölder class \(C^{0,\alpha }\) and the source term f belongs to \(L^\infty (\Omega )\), it is established in [1, Theorem 2.9] the uniqueness of weak solutions that are Lipschitz-continuous. Establishing an uniqueness result without these assumptions seems to be a rather difficult task since there is a priori no guarantee that distinct solutions \(u_1\) and \(u_2\) are in the same test function space. With no further assumptions, we prove, in Theorem 6.1, a uniqueness result for a one dimensional version of our problem (1.1).
Regarding the nonlocal problem (1.3), it should be stressed that many diffusion, or reaction-diffusion, equations with distinct nonlocal terms have been studied in many works and by different authors since the pioneer works by Chipot et al. [8, 9]. However, we could not find in the literature any p-Laplacian problem with a nonlocal exponent of nonlinearity p as we consider here. Usually, the motivation to study nonlocal problems relies in the physical fact that in reality the measurements of some quantities are not made pointwise but through some local averages. In this work, we prove, in Theorem 5.1, the existence of weak solutions to the nonlocal problem (1.3), and we show in the final section how the uniqueness for this problem is easily lost.
This paper is organised as follows. In the next Sect. 2 we introduce the basic properties of generalised Sobolev spaces that we will used later. The Sect. 3 is devoted to two auxiliary lemmas. In the Sect. 4 we give a proof of existence of a solution to the local problem (1.1) using a singular perturbation technique. The Sect. 5 is devoted to the existence of a solution to some nonlocal version of the problems we are interested in, i.e. to (1.3). Finally in the last Sect. 6 we evoke the issue of uniqueness for these problems working out in particular a one dimensional example. The notation used throughout the paper is nowadays rather standard in the analysis of Partial Differential Equations and therefore we address the reader to the monographs [2, 5, 6, 10, 11] for any question related to that matter.
2 Generalised Sobolev spaces
From the statement of the local problem (1.1), we can see that the exponent function p depends on the solution u and therefore it depends ultimately on the space variable x. As a consequence, the power p can be written as a variable exponent q(x) in the following sense,
This motivates us to look for the weak solutions to the problem (1.1) in a Sobolev space with variable exponents. The mathematical theory of these function spaces has been so developed during the last 20 years that we can now analyze the problem (1.1) in the light of this theory. In this section, we recall the properties of Lebesgue and Sobolev spaces with variable exponents which shall be used in the sequel. For this review, we have followed the monograph [2] (see also [10, 11, 17]).
To start with, we denote by \(\mathcal {Q}(\Omega )\) the set of all Lebesgue-measurable functions \(q:\Omega \rightarrow [1,\infty )\) and define
Given \(q\in \mathcal {Q}(\Omega )\), we denote by \(L^{q(\cdot )}(\Omega )\) the space of all Lebesgue-measurable functions \(u:\Omega \longrightarrow \mathbb {R}\) such that the modular is finite, i.e.
Equipped with the Luxembourg norm
\(L^{q(\cdot )}(\Omega )\) becomes a Banach space. Note that the infimum in (2.3) is attained if (2.2) is positive. The space \(L^{q(\cdot )}(\Omega )\) is a sort of Musielak-Orlicz space that we shall denote here by generalised Lebesgue space, because many of its properties are inherited from the classical Lebesgue spaces. If
\(L^{q(\cdot )}(\Omega )\) is separable and, in particular, \(C^\infty _0(\Omega )\) is dense in \(L^{q(\cdot )}(\Omega )\). Moreover, under assumption (2.4), \(L^{\infty }(\Omega )\cap L^{q(\cdot )}(\Omega )\) is also dense in \(L^{q(\cdot )}(\Omega )\). If we restrict (2.4) to
then \(L^{q(\cdot )}(\Omega )\) is reflexive. In this case, the dual space of \(L^{q(\cdot )}(\Omega )\) is identified with \(L^{q'(\cdot )}(\Omega )\), where \(q'(x)\) is the generalised Hölder conjugate of q(x),
Note that from (2.1) and (2.5), we have
One problem in generalised Lebesgue spaces, is the relation between the modular (2.2) and the norm (2.3) that is not so direct as in the classical Lebesgue spaces. However, if (2.5) holds, it can be proved, from its definitions in (2.2) and (2.3), that
When proving some estimates the following consequence of (2.6) is very useful,
In generalised Lebesgue spaces, there holds a version of Young’s inequality,
valid for some positive constant \(C(\delta )\) and any \(\delta >0\), and a version of Hölder’s inequality,
valid for \(u\in L^{q(\cdot )}(\Omega )\) and \(v\in L^{q'(\cdot )}(\Omega )\). As a consequence of (2.8), we have, for a bounded domain \(\Omega \) and q satisfying to (2.4), the following continuous imbedding,
Assuming the weak derivatives \(\displaystyle \frac{\partial \,u}{\partial \,x_i}\) exist for any \(i\in \{1,\dots ,d\}\), we define
which is a Banach space for the norm
This space belongs to a special class of Sobolev-Orlicz spaces so called generalised Sobolev spaces. The generalised Sobolev spaces \(W^{1,q(\cdot )}(\Omega )\) inherit many of the properties of the generalised Lebesgue spaces \(L^{q(\cdot )}(\Omega )\). In particular, \(W^{1,q(\cdot )}(\Omega )\) is separable if (2.4) holds, and is reflexive when (2.5) is fulfilled. We have as in (2.9)
We now introduce the following function space
which we endow with the norm
If \(q\in C(\overline{\Omega })\), then an equivalent norm in \(W^{1,q(\cdot )}_0(\Omega )\) is \(\Vert \varvec{\nabla }u\Vert _{q(\cdot )}.\)
Unlike classical Sobolev spaces, smooth functions are not necessarily dense in \(W^{1,q(\cdot )}_0(\Omega )\). So, defining
where \(C^\infty _0(\Omega )\) denotes the space of \(C^\infty \)-functions with compact support in \(\Omega \) we have generally
However, if \(\Omega \) is a bounded domain with \(\partial \Omega \) Lipschitz-continuous and q is log-Hölder continuous, then \(C^\infty _0(\Omega )\) is dense in \(W^{1,q(\cdot )}_0(\Omega )\). Recall that a function q is log-Hölder continuous, if
This means that
for the modulus of continuity \(\omega :\mathbb {R}^+\longrightarrow \mathbb {R}^+\) defined by
which is an increasing and continuous function for \(t< \frac{1}{2}\), and such that \(\lim _{t\rightarrow 0^+}\omega (t)=0\). If (2.13) holds, then we have
Note that in particular,
The Log-Hölder continuity property (2.13) is also very important to establish Sobolev inequalities in the framework of Sobolev spaces with variable exponents. Let us define the pointwise Sobolev conjugate of q(x) by
If q is a measurable function in \(\Omega \) satisfying to \(1\le q_-\le q_+<d\) and (2.13), then
for some positive constant C depending on \(q_+\), d and on the constant of (2.13). On the other hand, if q satisfies (2.13) and \(q_->d\), then
and where C is another positive constant depending on \(q_-\), d and on the constant of (2.13).
3 Auxiliary results
To prove later that \(|\varvec{\nabla }u|^{p(u)}\in L^1(\Omega )\), we shall make use of the following result which is a particular case of a more general one established by Zhikov [16]. We give here an elementary proof of this result which does not require all the assumptions considered in [16, Lemma 3.3].
Lemma 3.1
Assume that
Then \(\varvec{\nabla }u\in L^{q(\cdot )}(\Omega )^d\) and
Proof
By Young’s inequality one has for \(\mathbf {a},\ \mathbf {b}\in \mathbb {R}^d\) and \(1<q<\infty \),
If now \(\mathbf {b}\) is a function in \(L^{\infty }(\Omega )^d\) and we make \(q=q_n\) in (3.6) and use assumption (3.1), one derives
Using assumptions (3.2) and (3.3), we can pass to the limit in (3.7) as \(n\rightarrow \infty \), so that
Then we consider the following function \(\mathbf {b}\),
and where \(u\wedge v:= \min \{u,v\}\). Inserting this function \(\mathbf {b}\) into (3.8), one obtains
which implies
Observing that \(\frac{1}{q'(x)-1}+1=q(x)\), we arrive at
Finally (3.5) follows by letting \(k\rightarrow \infty \) in (3.9), and \(\varvec{\nabla }u\in L^{q(\cdot )}(\Omega )^d\) due to assumption (3.4). \(\square \)
We recall also the following inequalities which are classical in the theory of p-Laplace equations.
Lemma 3.2
For all \(\varvec{\xi },\ \varvec{\eta }\in \mathbb {R}^d\), the following assertions hold true:
Proof
4 Existence for the local problem
We define the set where we are going to look for the solutions to the problem (1.1) as
If \(1< p(u)<\infty \) for all \(u\in \mathbb {R}\), this set is a Banach space for the norm \(\Vert u\Vert _{W^{1,p(\cdot )}_0(\Omega )}\) defined at (2.12) which is equivalent to \(\Vert \varvec{\nabla }u\Vert _{p(u)}\) in the case of \(p(u)\in C(\overline{\Omega })\). If for some constant \(\alpha \), \(p \ge \alpha >1\), p continuous, then, in view of (2.11), \(W_0^{1,p(u)}(\Omega )\) is a closed subspace of \(W^{1,\alpha }_0(\Omega )\) and therefore it is separable and reflexive. In what follows, \(W^{-1,\gamma '}(\Omega )\), with \(1<\gamma <\infty \), denotes, as usual, the dual space of \(W_0^{1,\gamma }(\Omega )\).
Definition 1
Let the function p given in (1.2) be continuous and assume that
for some constants \(\alpha \) and \(\beta \). Assume also that
We say a function u is a weak solution to the problem (1.1) if
where \(\langle \cdot ,\cdot \rangle \) denotes the duality pairing between \((W_0^{1,p(u)}(\Omega ))'\) and \(W_0^{1,p(u)}(\Omega )\).
Note that \(q=p(u)\in \mathcal {Q}(\Omega )\) and the essential infimum \(q_{-}\) and the essential supremum \(q_{+}\) satisfy to (2.4) for all \(u\in W_0^{1,p(u)}(\Omega )\) (see (2.1)).
Theorem 4.1
Let \(\Omega \subset \mathbb {R}^d\), \(d\ge 2\), be a bounded domain with \(\partial \Omega \) Lipschitz-continuous. Assume that
and that condition (4.2) holds. If
then there exists, at least, one weak solution to the problem (1.1) in the sense of Definition 1.
Proof
The proof of Theorem 4.1 will be split into two main steps.
1. Approximation: For each \(\varepsilon >0\), we consider the auxiliary problem
where \(\beta \) is the upper bound constant from assumption (4.4).
For an exponent function p satisfying (4.3) and (4.4), we say that a function u is a weak solution to the regularized problem (4.5), if
where here \(\langle \cdot ,\cdot \rangle \) denotes the duality pairing between \(W^{-1,\alpha '}(\Omega )\) and \(W^{1,\alpha }_0(\Omega )\).
Claim 1
For each \(\varepsilon >0\) there exists a weak solution \(u_\varepsilon \) to the problem (4.5).
Proof of Claim 1
Let \(w\in L^2(\Omega )\) be given. From (4.4), we have
Observing that, in view of the assumption (4.2) and of (4.6), we have
and, by the usual theory of monotone operators, there exists a unique \(u=u_w\) solution to the problem
Taking \(v=u\) in the second line of (4.7) we derive using the Hölder inequality
for some positive constant \(C=C(\alpha ,\beta ,\Omega ,f)\), and where \(\Vert \cdot \Vert _{-1,\alpha '}\) is the operator norm associated to the norm \(\Vert \varvec{\nabla }\cdot \Vert _{\alpha }\). Thus
and
for some positive constant \(C=C(\alpha ,\beta ,\Omega ,\varepsilon ,f)\) independent of w. Since \(\beta >d \ge 2\) one has \(W^{1,\beta }_0(\Omega )\hookrightarrow L^2(\Omega )\), compactly and
for some positive constant \(C=C(\alpha ,\beta ,\Omega ,\varepsilon ,f,d)\) independent of w. Let us now consider the mapping
where \(B:=\{v\in L^2(\Omega ): \Vert v\Vert _2\le C\}\). From Schauder’s fixed point theorem, it is clear that this mapping will have a fixed point provided it is continuous. To prove this, let us assume that \(w_n\) is a sequence in \(L^2(\Omega )\) such that
For every \(n\in \mathbb {N}\), let \(u_n\) be the solution to the problem (4.7) associated to \(w=w_n\). By (4.8), one has
for some positive constant C which does not depend on n. Hence, for some subsequence still labelled by n and some u we have
By (4.7) written with \(u_n\) and \(w_n\) in the places of u and w, one has
Then, by monotonicity, one has also
Taking \(v=u_n-v\) in (4.13) and using the resulting equation in (4.14), we derive
In view of (4.10) one can assume that for some subsequence
By virtue of this and by assumption (4.3), we can apply Lebesgue’s theorem so that
for all \(v\in W^{1,\beta }_0(\Omega )\). Using (4.11) and (4.16) we can pass to the limit in (4.15) to get
Taking \(v=u\mp \delta z\), with \(z\in W^{1,\beta }_0(\Omega )\) and \(\delta >0\), we obtain from (4.17)
Letting \(\delta \rightarrow 0\) in (4.18), it comes
Thus \(u=u_w\). Since the limit is uniquely determined we have, in view of (4.12)
which proves the continuity of the mapping (4.9) and thus concludes the proof of the claim. \(\square \)
So far, we have proven that for each \(\varepsilon >0\) there exists \(u_\varepsilon \in W^{1,\beta }_0(\Omega )\) such that
Moreover recall that
2. Passage to the limit as\(\varepsilon \rightarrow 0\): Taking \(v=u_\varepsilon \) in (4.19), we get
Note that the first inequality in (2.7) can be written as
Thus by the Hölder inequality (2.8) one has
for some positive constant \(C=C(\alpha ,\beta ,\Omega )\). Thus we have
One deduces from (4.20), (4.22) and by using Young’s inequality that
for some positive constant C which does not depend on \(\varepsilon \). From (4.21) and (4.22) one then also has
for some positive constant C independent of \(\varepsilon \). Thus by the compact imbedding \(W^{1,\alpha }_0(\Omega )\hookrightarrow L^2(\Omega )\) we have for some subsequence still labelled with n and some u
It should be noticed that, due to (4.4), u is Hölder-continuous and, in view of this and (4.3), so does p(u). Due to (4.27), one has also
Moreover, recall that
We can then use (4.23), with \(u_{\varepsilon _n}\) in the place of \(u_\varepsilon \), together with (4.23), (4.26), (4.28) and (4.29) so that, by the application of Lemma 3.1, we have
Using the monotonicity, one has
Using the identity (4.19), with \(u_{\varepsilon _n}\) in the place of \(u_\varepsilon \) and \(u_{\varepsilon _n}-v\) in the place of v, we can write the inequality (4.31) as
say for all \(v\in C^\infty _0(\Omega )\). Note that, as in (4.16) but now using (4.28), by the Lebesgue theorem, we have for such a v
Using (4.24), (4.25) and (4.33), we can pass to the limit in (4.32) as \(n\rightarrow \infty \) so that
As observed above, due to assumptions (4.3) and (4.4), p(u) is Hölder-continuous and therefore \(C^\infty _0(\Omega )\) is dense in \(W^{1,p(u)}_0(\Omega )\) due to (2.13)–(2.14). Thus, (4.34) holds true also for all \(v\in W^{1,p(u)}_0(\Omega )\). Hence we can take \(v=u\mp \delta z\), with \(z\in W^{1,p(u)}_0(\Omega )\) and \(\delta >0\), in (4.34) so that
As a consequence,
which, together with (4.30), completes the proof of Theorem 4.1. \(\square \)
5 Nonlocal problems
In this section we consider a real function p such that
for some constants \(\alpha , \beta \). We denote by b a mapping from \(W_0^{1,\alpha }(\Omega )\) into \(\mathbb {R}\) such that
i.e. b sends bounded sets of \(W_0^{1,\alpha }(\Omega )\) into bounded sets of \(\mathbb {R}\).
Definition 2
A function u is a weak solution to the problem (1.3) if
where \(\langle \cdot ,\cdot \rangle \) denotes the duality pairing between \((W_0^{1,p(b(u))}(\Omega ))'\) and \(W_0^{1,p(b(u))}(\Omega )\).
One should notice that p(b(u)) is here a real number and not a function so that the Sobolev spaces involved are the classical ones. We refer to [5, 7,8,9] for more on nonlocal problems.
Then one has:
Theorem 5.1
Let \(\Omega \subset \mathbb {R}^d\), \(d\ge 2\), be a bounded domain and assume that (5.1) and (5.2) hold together with
Then there exists at least one weak solution to the problem (1.3) in the sense of Definition 2.
The proof of Theorem 5.1 is based on the following result.
Lemma 5.1
For \(n\in \mathbb {N}\), let \(u_n\) be the solution to the problem
where \(\langle \cdot ,\cdot \rangle \) denotes here the duality pairing between \((W_0^{1,p_n}(\Omega ))'\) and \(W_0^{1,p_n}(\Omega )\). Suppose that
Then
where u is the solution to the problem
Proof of Lemma 5.1
We shall split this proof into two steps.
1. Weak convergence: We first observe that, in view of \(p_n\rightarrow p\), as \(n\rightarrow \infty \), and \(q<p\), we may assume that
Taking \(v=u_n\) in the equation of (5.4) we get
Recall that \( \Vert f\Vert _{-1,q'}\) denotes the strong dual norm of f associated to the norm \(\Vert \varvec{\nabla }\cdot \Vert _{q}\). On the other hand, by using Hölder’s inequality and (5.9), we have
for some positive constant \(C=C(p,q,\Omega )\). Plugging (5.11) into (5.10) it comes
for some other positive constant \(C=C(p,q,\Omega , f)\). Combining (5.11) with (5.12), it follows that
for some positive constant C independent of n. From (5.13) we deduce then that for some subsequence still labelled by n and for some \(u \in W^{1,q}_0(\Omega )\)
Due to (5.5), (5.9), (5.12) and (5.14), we can also apply Lemma 3.1 so that
As a consequence we have
Clearly the equation in (5.4) is equivalent to
and by the Minty lemma to
Taking \(v\in C_0^\infty (\Omega )\), one can use (5.5) and (5.14) to pass to the limit in (5.16), as \(n\rightarrow \infty \), so that
Using the density of \(C_0^\infty (\Omega )\) in \(W_0^{1,p}(\Omega )\), we see that (5.17) also holds for all \(v\in W_0^{1,p}(\Omega )\). In this case, taking \(v=u\pm \delta z\), with \(z\in W_0^{1,p}(\Omega )\) and \(\delta >0\), and letting \(\delta \rightarrow 0\) after simplifying the resulting inequality, one obtains
Thus u is the solution to the problem (5.8).
2. Strong convergence: We want to show that the convergence (5.14) is in fact strong. To prove this, we first note that, taking \(v=u_n\) in the equation of (5.4) and using (5.14) to pass to the limit, we obtain
Consider the case of the \(p_n\)’s such that
One has by Hölder’s inequality
where \(|\Omega |\) denotes the d-Lebesgue measure of \(\Omega \). Thus by (5.18) for such a sequence
which shows (since \( \Vert \varvec{\nabla }u_n\Vert _{p} \rightarrow \Vert \varvec{\nabla }u\Vert _{p}\), as \(n\rightarrow \infty \))
Since \(W^{1,p}_0(\Omega )\subset W^{1,q}_0(\Omega )\), (5.19) implies (5.7).
Next, consider the \(p_n\)’s such that
and set
Due to the monotonicity, \(A_n\ge 0\) and, because of (5.4), one has
From (5.6) and (5.14), we have
Moreover, from (5.15) one easily gets
Hence, (5.20), (5.22) and (5.23) ensure that
Assume first that
This allows us to use property (3.10) of Lemma 3.2 in (5.21) so that
Since, by (5.20), \(p_n>q\), we have by Hölder’s inequality, (5.20), (5.24) and (5.25)
when \(n \rightarrow \infty \). This proves (5.7) in this case.
Consider now the case when
Here, we use Hölder’s inequality as follows
Using property (3.11) of Lemma 3.2 we have
for some positive constant \(C=C(p_n)\). Now, by using (5.26), (5.27) together with (5.12) we deduce that
Thus, as above, (5.7) holds true also in this case. \(\square \)
Let us now show how Lemma 5.1 can be applied to prove the existence of weak solutions to the nonlocal problem (1.3).
Proof of Theorem 5.1
Note that \(f \in (W_0^{1,\alpha }(\Omega ))' \subset (W_0^{1,\delta }(\Omega ))'\) for any \(\delta > \alpha \). Thus for each \(\lambda \in \mathbb {R}\), there exists a unique solution \(u=u_\lambda \) to the \(p(\lambda )\)-Laplacian problem
Taking \(v=u=u_\lambda \) in (5.28) one derives
By Hölder’s inequality one has
Thus by (5.29) it comes
Gathering (5.30) and (5.31), and using (5.1) we obtain
for some positive constant \(C=C(\alpha ,\beta ,\Omega ,f)\). Due to the boundedness of b, see (5.2), and to (5.32), there exists \(L\in \mathbb {R}\) such that
Let us now consider the map
from \([-L,L]\) into itself. This map is continuous. Indeed, if \(\lambda _n\rightarrow \lambda \) as \(n\rightarrow \infty \), due to (5.1), we have \(p(\lambda _n)\rightarrow p(\lambda )\). Applying now Lemma 5.1 with \(p_n=p(\lambda _n)\), it follows that
Now, b being continuous (see (5.2)), it follows that \(b(u_{\lambda _n})\longrightarrow b(u_\lambda )\), as \(n\rightarrow \infty \), and thus the map (5.33) is also continuous. It has then a fixed point \(\lambda _0\) and \(u_{\lambda _0}\) is then solution to (5.3). \(\square \)
6 Uniqueness
The proof of uniqueness of the solution to (1.1) in all generality does not seem to be straightforward. In fact, if we have two weak solutions \(u_1\) and \(u_2\) to the problem (1.1), in the sense of the Definition 1, there is a priori no guarantee that both functions \(u_1\) and \(u_2\) are in the same test function space, \(W_0^{1,p(u_1)}(\Omega )\) or \(W_0^{1,p(u_2)}(\Omega )\). Hence, we cannot use \(u_1-u_2\) as test function as usual. However, if we restrict ourselves to the 1-dimensional problem (1.1), it is possible to prove some uniqueness result. This is what we would like to do now.
Let us set, for instance,
and consider the problem
where \(u'\) denotes the derivative of u.
Theorem 6.1
Assume (4.1) and (4.3) and suppose that \(f >0\) is a continuous function on \([-1,1]\). Then there exists a unique solution to (6.1) in the distributional sense.
Remark 6.1
One could weaken the assumptions on f. Note that from (6.1) it results that \(|u'|^{p(u)-2}u'\) and also \(u'\) are bounded. u being a Lipschitz continuous function the boundary conditions are well defined.
Proof of Theorem 6.1
In view of (6.1), one has
and therefore the function \(|u'|^{p(u)-2}u'\) is decreasing in \(\Omega \). This function cannot have a constant sign, otherwise \(u'\) would have a constant sign too, i.e. \(u'\) would be always positive or negative, which in turn would render \(u(-1)=u(1)\) impossible. Thus, \(|u'|^{p(u)-2}u'\) and consequently \(u'\) vanish at some point \(\xi \in (-1,1)\). Since \(|u'|^{p(u)-2}u'\) is decreasing in \(\Omega \), one has
Let us set
Note that, due to the positivity of f, F is an increasing function. On the other hand, in view of (6.2), one derives from (6.1) that
for either \(-1<x<\xi \) or \(\xi<x<1\). Thus, (6.1) can be decoupled into two problems namely,
and
Claim 1
For a fixed \(\xi \), the functions
are uniformly Lipschitz-continuous with respect to u in the intervals \((-1,\xi )\) and \((\xi ,1)\), respectively.
Proof of Claim 1
We shall only prove the claim for the function G(x, u), the proof for the function H(x, u) being analogous. Let us first write
By the mean value theorem, one gets for some \(\theta \in (0,1)\)
On the other hand, by the monotony of the \(\ln \) function and due to assumption (4.1), one has
and thus the product of the first two terms in (6.6) is uniformly bounded by a constant C, that does not depend on x. Observing this and using the assumption (4.3), one has
for another positive constant \(C'\), which proves the claim. \(\square \)
As a consequence of Claim 1, the solutions to the problems (6.4) and (6.5) are unique.
Claim 2
The mapping \(\xi \mapsto u(\xi )\) is an increasing function of \(\xi \) when \(\xi \) runs in the interval \((-1,1)\).
Proof of Claim 2
Let us denote by \(u(\xi ,x)\) the unique solution to the problem (6.4). We would like to show that, whenever \(\xi ,\ \xi '\in (-1,1)\), we have
Note again that, in view of the positivity of f the function F defined at (6.3) is increasing and therefore \(F(\xi ')>F(\xi )\). From (6.4) we derive then
which shows that \(u(\xi ,x)<u(\xi ',x)\) near \(x=-1\). If there is a first point \(x_0\in (-1,\xi )\) where \(u(\xi ',x_0)=u(\xi ,x_0)\), we argue as we did for (6.8) so that one will have
which is impossible. Thus, \(u(\xi ,x)<u(\xi ',x)\) on \((-1,\xi )\) and one has (6.7), since \(u(\xi ,\cdot )\) continues to grow after \(\xi \).
With the same arguments, if \(v(x,\xi )\) denotes the unique solution to the problem (6.5), one would show that, whenever \(\xi ,\ \xi '\in (-1,1)\),
which completes the proof of the claim. \(\square \)
Let us now show that the conclusion of Theorem 6.1 follows from Claim 2. In fact, since we know the solution to the problem (6.1) exists, there exists a \(\xi \in (-1,1)\) such that
i.e. the solutions to the problems (6.4) and (6.5) merge at \(\xi \). Then, for \(\xi '>\xi \), Claim 2 ensures that
and thus the two solutions could merge only at a point. The same argument applies for \(\xi '<\xi \) and therefore the uniqueness for the problem (6.1) holds, which completes the proof of Theorem 6.1. \(\square \)
The nonlocal case is completely different and one can see that uniqueness to the problem (1.3) is easily lost. To see this, let \(p_1,\ p_2\in (1,\infty )\), with \(p_1\not =p_2\), and consider \(u_1\) and \(u_2\) respectively the solutions to the problems
for \(i=1\) and \(i=2\). Choose a function b such that
Let us now consider a function p as in (1.4) and such that
Then \(u_1\) and \(u_2\) are both solutions to the problem (1.3). One can this way construct problems with infinitely many solutions.
Remark 6.2
It would be interesting in some cases (nonlocal case for instance) to relax our assumptions on f allowing it for example to be an integrable function. This involves some more work of approximation. One can also think of extending our results to the parabolic case revisiting our estimates and to address the issue of the asymptotic behaviour of the solution at least in the simple case of a single steady state (Theorem 6.1).
Change history
04 July 2019
In the Original Publication of the article, few errors have been identified in section 5 and acknowledgements section.
References
Andreianov, B., Bendahmane, M., Ouaro, S.: Structural stability for variable exponent elliptic problems. II. The \(p(u)\)-Laplacian and coupled problems. Nonlinear Anal. 72(12), 4649–4660 (2010)
Antontsev, S., Shmarev, S.: Evolution PDEs with Nonstandard Growth Conditions. Existence, Uniqueness, Localization, Blow-up. Atlantis Press, Paris (2015)
Blomgren, P., Chan, T., Mulet, P., Wong, C.: Total variation image restoration: Numerical methods and extensions. In: Proceedings of the IEEE International Conference on Image Processing, vol. 3, 384–387. IEEE Computer Society Press, Piscataway (1997)
Bollt, E., Chartrand, R., Esedoglu, S., Schultz, P., Vixie, K.: Graduated, adaptive image denoising: local compromise between total-variation and isotropic diffusion. Adv. Comput. Math. 31, 61–85 (2007)
Chipot, M.: Elements of Nonlinear Analysis. Birkhäuser, Basel (2000)
Chipot, M.: Elliptic Equations: An Introductory Course. Birkhäuser, Basel (2009)
Chipot, M., Chang, N.-H.: On some mixed boundary value problems with nonlocal diffusion. Adv. Math. Sci. Appl. 14(1), 1–24 (2004)
Chipot, M., Lovat, B.: Existence and uniqueness results for a class of nonlocal elliptic and parabolic problems. Dyn. Contin. Discr. Impuls. Syst. Ser. A Math. Anal. 8(1), 35–51 (2001)
Chipot, M., Rodrigues, J.F.: On a class of nonlocal nonlinear elliptic problems. RAIRO: Modél. Math. Anal. Numér 26(3), 447–467 (1992)
Cruz-Uribe, D., Fiorenza, A.: Variable Lebesgue Spaces. Foundations and Harmonic Analysis. Birkhäuser/Springer, Heidelberg (2013)
Diening, L., Harjulehto, P., Hästo, P., Ru̇žička, M.: Lebesgue and Sobolev Spaces with Variable Exponents. Springer, Heidelberg (2011)
Glowinski, R., Marrocco, R.: Sur l’approximation, par éléments finis d’ordre un, et la résolution, par pénalisation-dualité, d’une classe de problèmes de Dirichlet non linéaires. RAIRO Ana. Numer. 9(R–2), 41–76 (1975)
Chan, T., Esedoglu, S., Park, F., Yip, A.: Total Variation Image Restoration: Overview and Recent Developments. In: Paragios, N., Chen, Y., Faugeras, O. (eds.) Handbook of Mathematical Models in Computer Vision, 17–32. Springer, New York (2006)
Türola, J.: Image denoising using directional adaptive variable exponents model. J. Math. Imaging. Vis. 57, 56–74 (2017)
Zhikov, V.V.: Averaging of functionals of the calculus of variations and elasticity theory (Russian). Izv. Akad. Nauk SSSR Ser. Mat. 504(4), 675–710 (1986). English translation in Mah. USSR Izvestiya 29 (1987) 33–66
Zhikov, V.V.: On the technique for passing to the limit in nonlinear elliptic equations. Funct. Anal. Appl. 43(2), 96–112 (2009)
Zhikov, V.V.: On variational problems and nonlinear elliptic equations with nonstandard growth conditions. J. Math. Sci. (N.Y.) 173(5), 463–570 (2011)
Acknowledgements
We are very grateful to the referees for their constructive remarks. This work was performed when the first author was visiting the USTC in Hefei and during a part time employment at the S. M. Nikolskii Mathematical Institute of RUDN University, 6 Miklukho-Maklay St, Moscow, 117198, supported by the Ministry of Education and Science of the Russian Federation. He is grateful to these institutions for their support. Main part of this work was carried out also during the visit of the second author to the University of Zurich during the first quarter of 2018. Besides the Grant SFRH/BSAB/135242/2017 of the Portuguese Foundation for Science and Technology (FCT), Portugal, which made this visit possible, the second author also wishes to thank to Prof. Michel Chipot who kindly welcomed him at the University of Zurich.
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Y. Giga.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
About this article
Cite this article
Chipot, M., de Oliveira, H.B. Some results on the p(u)-Laplacian problem. Math. Ann. 375, 283–306 (2019). https://doi.org/10.1007/s00208-019-01803-w
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s00208-019-01803-w