1. Introduction
The study of the double phase problem gained widespread attention in the last century, primarily due to the seminal works of Marcellini [Reference Marcellini62, Reference Marcellini63] and Zhikov [Reference Zhikov77, Reference Zhikov79]. Interest in this type of unbalanced growth problem,
\begin{equation}
\mathcal{G}(u)=\operatorname{div}\left(\mu(x)\vert\nabla u\vert^{p-2}\nabla u
+\vert\nabla u\vert^{q-2}\nabla u\right),\quad\forall u\in W^{1,\mathcal{H}}_{0}(\Omega),
\end{equation}is driven by its broad applicability in physical, biological, and engineering models. Notable examples include transonic flow models studied by Bahrouni–Rădulescu–Repovš [Reference Bahrouni, Radulescu and Repovs4], applications in quantum physics explored by Benci–D’Avenia–Fortunato–Pisani [Reference Benci, D’Avenia, Fortunato and Pisani8], and models in electromagnetism (specifically the Born–Infeld equation) investigated by Bonheure–D’Avenia–Pomponio [Reference Bonheure, D’Avenia and Pomponio9] Further applications include image restoration and denoising (see Charkaoui–Ben Loghfyry–Zeng [Reference Charkaoui, Ben-Loghfyry and Zeng19] and Harjulehto–Hästö [Reference Harjulehto and Hästö55]), reaction-diffusion system models (see Cherfils–Il’yasov [Reference Cherfils and Il’yasov20]), and elasticity theory to describe the behaviour of composite materials under deformation (see Zhikov [Reference Zhikov77]).
In recent years, research on the double phase problem has expanded significantly, resulting in a rich body of literature. For further references, we direct interested readers to Bai–Papageorgiou–Zeng [Reference Bai, Papageorgiou and Zeng5], Crespo Blanco–Papageorgiou–Winkert [Reference Crespo-Blanco, Papageorgiou and Winkert35], D’Aguì–Sciammetta–Tornatore–Winkert [Reference D’Aguì, Sciammetta, Tornatore and Winkert37], Gambera–Guarnotta–Papageorgiou [Reference Gambera, Guarnotta and Papageorgiou43], Gasiński–Papageorgiou [Reference Gasiński and Papageorgiou48, Reference Gasiński and Papageorgiou49], Gasiński–Winkert [Reference Gasiński and Winkert50, Reference Gasiński and Winkert51], Gou–Rădulescu [Reference Gou and Radulescu52], Guarnotta–Livrea–Winkert [Reference Guarnotta, Livrea and Winkert53], Jing–Liu–Papageorgiou [Reference Jing, Liu and Papageorgiou58], Marino–Winkert [Reference Marino and Winkert64], Papageorgiou–Pudełko–Rădulescu [Reference Papageorgiou, Pudełko and Radulescu68], Papageorgiou–Rădulescu–Repovš [Reference Papageorgiou, Radulescu and Repovš69], Papageorgiou–Rădulescu–Zhang [Reference Papageorgiou, Radulescu and Zhang71], Papageorgiou–Rădulescu–Zhang [Reference Papageorgiou, Radulescu and Zhang70], Papageorgiou–Vetro–Vetro [Reference Papageorgiou, Vetro and Vetro72], Sciammetta–Tornatore–Winkert [Reference Sciammetta, Tornatore and Winkert76], and the reference therein.
Moreover, the double phase operator holds significant mathematical interest in the study of Lavrentiev’s phenomenon (see Borowski–Chlebicka–De Filippis–Miasojedow [Reference Borowski, Chlebicka, De Filippis and Miasojedow10], De Filippis–Mingione [Reference De Filippis and Mingione38], and Zhikov [Reference Zhikov78]). For regularity results, we refer to the seminal works of Baroni–Colombo–Mingione [Reference Baroni, Colombo and Mingione7], Colombo–Mingione [Reference Colombo and Mingione28], Ho–Winkert [Reference Ho and Winkert56], Ragusa–Tachikawa [Reference Ragusa and Tachikawa75] and the related references. Recently, Perera–Squassina [Reference Perera and Squassina74] have approached this operator using Morse theory. For a comprehensive overview, we refer to the survey by Papageorgiou [Reference Papageorgiou67] and Chapter 4, Section 9 of the book by Papageorgiou–Winkert [Reference Papageorgiou and Winkert73].
Our problem is nonlocal in the sense introduced by Carrier [Reference Carrier17] in the study of beam deformation in lower dimensions:
\begin{equation*}
\left[1+\frac{\alpha^{-2}}{2\pi}\int_{0}^{\pi}\varphi^{2}
(\xi,\eta)d\xi\right]
\frac{\partial^{2}\varphi}{\partial\xi^{2}}
=\frac{\partial^{2}\varphi}{\partial\eta^{2}},
\end{equation*} where
$\alpha$ is a positive constant. These types of nonlocal models have been extensively used in biological applications, such as population dynamics:
\begin{equation*}
-a\left(\int_{\Omega}\vert u\vert^{p}dx\right)\Delta u=f(u),
\end{equation*}where the dispersion velocity depends on the population density gradient:
\begin{equation}
v=a\left(\int_{\Omega}u^{q}dx\right)\nabla u.
\end{equation}Relevant studies in this context include those by Chang–Chipot [Reference Chang and Chipot18], Chipot–Lovat [Reference Chipot and Lovat21, Reference Chipot and Lovat22], Chipot–Molinet [Reference Chipot and Molinet23], Chipot–Rodrigues [Reference Chipot and Rodrigues24], Chipot–Siegwart [Reference Chipot and Siegwart25], Corrêa [Reference Corrêa29], Corrêa–Menezes–Fereira [Reference Corrêa, Menezes and Ferreira30], Furter–Grinfeld [Reference Furter and Grinfeld42], and the reference therein. Bifurcation arguments for these nonlocal problems can be found in Figueiredo Sousa–Morales Rodrigo–Suárez [Reference Figueiredo-Sousa, Morales-Rodrigo and Suárez41] and Schauder’s fixed point theorem application in Bueno–Ercole–Ferreira–Zumpano [Reference Bueno, Ercole, Ferreira and Zumpano12].
We point out that, unlike Kirchhoff type models,
\begin{equation*}
-a\left(\int_{\Omega}\left(\mu(x)\frac{\vert\nabla u\vert^{p}}{p}
+\frac{\vert\nabla u\vert^{q}}{q}\right)dx\right)\mathcal{G}(u)=f(u),
\end{equation*}which permit a variational formulation (see Arora–Fiscella–Mukherjee–Winkert [Reference Arora, Fiscella, Mukherjee and Winkert2, Reference Arora, Fiscella, Mukherjee and Winkert3], Colasuonno–Perera [Reference Colasuonno and Perera26], Crespo Blanco–Gasński–Winkert [Reference Crespo-Blanco, Gasiński and Winkert34], Santos Júnior–Siciliano [Reference Júnior and Siciliano57]), our nonlocal problem is non-variational due to the presence of the nonlocal weight function on the left-hand side. Consequently, we employ a combination of techniques to obtain our results.
In particular, we study a variational problem with a “frozen” nonlocal weight function. Furthermore, we approach the problem through a one-dimensional fixed-point method, utilizing the sub-super solution technique and set-valued analysis to construct a well-defined fixed-point map. Early works in this direction include Gasiński–Santos Júnior [Reference Gasiński and Júnior44, Reference Gasiński and Júnior45] for the Laplacian case, Gasiński–Santos Júnior–Siciliano [Reference Gasiński, Júnior and Siciliano46] for the Laplacian with possible singularities, Candito–Gasiński–Livrea–Santos Júnior [Reference Candito, Gasiński, Livrea and Júnior15] for the
$p$-Laplacian problem, Crespo Blanco–Failla–Vassallo [Reference Crespo-Blanco, Failla and Vassallo32] for the
$p$-Laplacian with Neumann boundary conditions, and Candito–Failla–Livrea [Reference Candito, Failla and Livrea14] for variable exponent
$p(x)$-Laplacian.
In previous studies, a crucial assumption involved the monotonicity of the map
$t\mapsto \frac{f(t)}{t^{p-1}}$. However, in a recent work by Candito–Failla–Gasiński–Livrea [Reference Candito, Failla, Gasiński and Livrea13], the authors removed this assumption by introducing a novel sub-solution construction combined with a set-valued map. Building on these ideas, our objective is to establish the existence and multiplicity of solutions for the following non-variational nonlocal elliptic problem driven by an unbalanced growth operator, namely, the double phase operator.
To the best of our knowledge, this is the first study addressing the nonlocal double phase problem of the form:
\begin{align}
\left \{
\begin{array}{ll}
-a\left(\displaystyle\int_{\Omega}\left(u^{r}+\nu(x)u^{s}\right)dx\right)
\mathcal{G}(u)
= f(u) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega,
\end{array}
\right.
\end{align} where
$\Omega\subset\mathbb{R}^{N}$ (
$N\geq 2$) has a Lipschitz boundary
$\partial\Omega$ and
$1 \lt q \lt p$,
$r,s\ge 1$.
We now present the precise hypotheses required for our main multiplicity result. First on the exponents and weight functions we require that
•
$(H_{1})$
$1 \lt q \lt N$,
$q \lt p$,
$\frac{p}{q} \lt 1+\frac{1}{N}$,
$r,s\geq 1$, and
$\mu\in C^{0,1}(\overline{\Omega})$,
$\mu:\Omega\to[0,+\infty)$,
$\nu\in L^{\infty}(\Omega)$,
$\nu:\Omega\to [0,+\infty)$.
Notice that, the above assumption implies that
$q \lt p \lt q^{*}$, where
$q^{*}=\frac{Nq}{N-q}$ is the Sobolev critical exponent. Next, we impose an “oscillatory” behaviour on
$a$, characterized by multiple zeros and strictly positive values in alternating intervals.
•
$(H_{2})$ The function
$a:[0,+\infty)\to\mathbb{R}$ is continuous and there exist positive numbers
$0\leq\alpha_{0}\leq \alpha_{1} \lt \alpha_{2}\leq \alpha_{3} \lt \alpha_{4}\leq...\leq
\alpha_{2k-1} \lt \alpha_{2K}$
$(K\geq 1)$ such that
$a \gt 0$ in
$(\alpha_{2k-1},\alpha_{2k})$ and
$a(\alpha_{2k-1})=a(\alpha_{2k})=0$ for all
$k\in\{1,...,K\}$.
For the right hand side nonlinearity we assume the following regularity conditions.
• (
$H_{3})$ There exists
$t_* \gt 0$, such that function
$f:\mathbb{R}\to\mathbb{R}$ satisfies
$f\in C([0,t_{*}])$,
$f(t) \gt 0$ for all
$t\in(0,t_{*})$,
$f(t_*)=0$ and
\begin{equation*}
\liminf\limits_{s\to 0^{+}}f(s) \gt \max_{k\in\{1,...,K\}}\max\limits_{\alpha\in[\alpha_{2k-1},\alpha_{2k}]}a(\alpha).
\end{equation*}
In our proofs, we will make use of the solution to the following torsion problem:
\begin{align}
\left \{
\begin{array}{ll}
-\mathcal{G}(u) = 1 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{align} By the Minty–Browder Theorem (see Brézis [Reference Brézis11, Theorem 5.16]), since
$-\mathcal{G}(u)$ is strictly monotone (see Gasiński–Papageorgiou [Reference Gasiński and Papageorgiou49, Proposition 2.4]) and
$1\in\left( W^{1,\mathcal{H}}_{0}(\Omega)\right)^{*}$, there exists a unique positive solution
$e_{1}\in W^{1,\mathcal{H}}_{0}(\Omega)$. Moreover, by results from Ho–Winkert [Reference Ho and Winkert56, Theorem 4.2] and Crespo Blanco–Winkert [Reference Crespo-Blanco and Winkert36, Theorem 3.1], it follows that
$e_{1}\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$.
The following conditions impose further constraints on the domains of
$f$ and
$a$.
•
$(H_{4})$
$t_*\le \Vert e_1\Vert_{\infty}$ and
$\alpha_{2K} \lt t_{*}^{r}\Vert e\Vert^{r}_{r}
+t_{*}^{s}\Vert e\Vert_{s}^{s}\min\limits_{\Omega}\nu $, where
$e=e_{1}/\Vert e_{1}\Vert_{\infty}$.
Finally, we impose some assumptions on the height of each “even bump” of
$a$.
•
$(H_{5})$ For each
$k\in\{1,...,K\}$ there exists
$\tau_{k}\in(\alpha_{2k-1},\alpha_{2k})$ such that
\begin{equation*}
\tau_{k}^{q}a(\tau_{k}) \gt \frac{\vert\Omega\vert^{q}}{\lambda_{1,q}}
\left(t_{*}^{r-1}+t_{*}^{s-1}\max\limits_{\Omega}
\nu\right)^{q}\max\limits_{t\in[0,t_{*}]}[tf(t)],
\end{equation*}where
$\lambda_{1,q}$ is the first eigenvalue of the operator
$\left(-\Delta_{q},W^{1,q}_{0}(\Omega)\right)$.
Our goal is to establish the following multiplicity theorem (see Section 2 for the definition of the spaces).
Theorem 1.1. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$,
$(H_{4})$ and
$(H_{5})$ hold. Then, problem (P) admits at least
$2K$ positive solutions in
$W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$.
The rest of the paper is organized as follows. Section 2 introduces the functional setting for our solutions, specifically the Sobolev–Musielak–Orlicz spaces, and recalls relevant results from set-valued analysis. Section 3 establishes the existence of bounded solutions for a frozen problem using a combination of variational and truncation methods with the sub-super solution approach. Section 4 constructs a set-valued map to build the minimal positive solution of the frozen problem. Section 5 formulates a one-dimensional fixed-point problem and derives geometric constraints on the map to obtain multiple solutions for the original problem.
2. Preliminaries
2.1. Setting spaces
We introduce the natural setting for our problem, see Adams–Fournier [Reference Adams and Fournier1] for a complete overview on this topic. Let us begin by considering the Lebesgue space, set
$1\leq r \lt +\infty$,
\begin{equation*}
L^{r}(\Omega)=\left\{u:\Omega\to\mathbb{R}\
\text{measurable }:\int_{\Omega}
\vert u\vert^{r}dx \lt \infty\right\}
\end{equation*} with the classical
$r$-norm
\begin{equation*}
\Vert u\Vert_{r}=\left(\int_{\Omega}\vert u\vert^{r}
dx\right)^{\frac{1}{r}}.
\end{equation*} For
$r=+\infty$, we consider
with the sup-norm
Moreover, for
$1\leq r \lt +\infty$, we use the Sobolev space
with the norm
In particular, we introduce the Sobolev space
\begin{equation}
W^{1,r}_{0}(\Omega)
=\overline{C_{0}^{\infty}(\Omega)}^{\Vert\cdot\Vert_{1,r}}
\end{equation}endowed with the equivalent norm
For our purpose, we introduce the notion of Musielak–Orlicz space and Musielak–Orlicz–Sobolev space, see Harjulehto–Hästö [Reference Harjulehto and Hästö54] for details. Let us define the nonlinear function
$\mathcal{H}\colon\Omega\times[0,+\infty)\to[0,+\infty)$
which in fact has an unbalance growth, that is,
for some
$b_{1},b_{2} \gt 0$ and let us consider the related modular function
\begin{equation}
\rho_{\mathcal{H}}(u)=\int_{\Omega}\mathcal{H}(x,\vert u\vert)dx.
\end{equation}Then, we introduce the so called Musielak–Orlicz space as follows
endowed with the Luxemburg norm
\begin{equation*}
\Vert u\Vert_{L^{\mathcal{H}}(\Omega)}=\inf\left\{\lambda \gt 0:
\rho_{\mathcal{H}}\left(\frac{u}{\lambda}\right)\leq 1\right\}.
\end{equation*}Consider the following space
\begin{equation}
L^{p}_{\mu}(\Omega)=\left\{u:\Omega\to\mathbb{R}, \text{measurable }:
\int_{\Omega}\mu(x)\vert u\vert^{p}dx \lt +\infty\right\},
\end{equation}endowed with the semi-norm
\begin{equation}
\Vert u\Vert_{p,\mu}=\left(\int_{\Omega}\mu(x)\vert u\vert^{p}dx\right)^{\frac{1}{p}}.
\end{equation}By Colasuonno–Squassina [Reference Colasuonno and Squassina27, Proposition 2.15], it follows that the embeddings
\begin{equation}
L^{p}(\Omega)\hookrightarrow L^{\mathcal{H}}(\Omega)\hookrightarrow L^{p}_{\mu}(\Omega)\cap L^{q}(\Omega),
\end{equation}are continuous. Analogously, we introduce the Musielak–Orlicz–Sobolev space as
\begin{equation*}
W^{1,\mathcal{H}}(\Omega)=\left\{u\in L^{\mathcal{H}}(\Omega):
\vert \nabla u\vert\in L^{\mathcal{H}}(\Omega)\right\},
\end{equation*}with the norm
Starting from Colombo–Mingione [Reference Colombo and Mingione28] and Crespo Blanco–Gasiński–Harjulehto–Winkert [Reference Crespo-Blanco, Gasiński, Harjulehto and Winkert33, Theorem 2.23], assumption
$(H_{1})$ allows us to consider the space
\begin{equation}
W^{1,\mathcal{H}}_{0}(\Omega)
=\overline{C_{0}^{\infty}(\Omega)}^{\Vert\cdot\Vert_{W^{1,\mathcal{H}}(\Omega)}},
\end{equation}and by the Musielak–Orlicz version of the Poincaré inequality (see e.g., Harjulehto–Hästö [Reference Harjulehto and Hästö54, Theorem 6.2.8, pag. 130]) we can equip this space with the equivalent norm
Note that, by Colasuonno–Squassina [Reference Colasuonno and Squassina27, Proposition 2.18] (see also, Crespo Blanco–Gasiński–Harjulehto–Winkert [Reference Crespo-Blanco, Gasiński, Harjulehto and Winkert33, Proposition 2.12] and Gasiński–Winkert [Reference Gasiński and Winkert50, Section 2]), in our assumptions, the norm
$\Vert\cdot\Vert_{L^{\mathcal{H}}(\Omega)}$ is uniformly convex, and then,
$L^{\mathcal{H}}(\Omega)$,
$W^{1,\mathcal{H}}(\Omega)$ and
$W^{1,\mathcal{H}}_{0}(\Omega)$ are reflexive Banach spaces.
Remark 2.1. We point out that the following relation holds, see e.g., [Reference Gasiński and Winkert50, eq. (2.2)]
\begin{align}
\min\left\{\Vert u\Vert_{L^{\mathcal{H}}(\Omega)}^{p},
\Vert u\Vert_{L^{\mathcal{H}}(\Omega)}^{q}\right\}\leq \Vert u\Vert_{p,\mu}^{p}
+\Vert u\Vert_{q}^{q}&\leq \max\left\{\Vert u\Vert_{L^{\mathcal{H}}(\Omega)}^{p},
\Vert u\Vert_{L^{\mathcal{H}}(\Omega)}^{q}\right\}\nonumber\\ &\qquad\qquad\qquad\forall u\in L^{\mathcal{H}}(\Omega).
\end{align}For readers convenience, we recall the following properties of double phase operator, see Liu–Dai [Reference Liu and Dai60, Proposition 3.1].
Proposition 2.2. Let
$L:W^{1,\mathcal{H}}_{0}(\Omega)\to \left(W^{1,\mathcal{H}}_{0}(\Omega)\right)^{*}$ be such that
\begin{equation*}
\langle Lu,v\rangle=\int_{\Omega}\left(\mu(x)
\vert\nabla u\vert^{p-2}\nabla u+\vert\nabla u\vert^{q-2}.
\nabla u\right)\nabla vdx.
\end{equation*} Then,
$L$ is a continuous, bounded, strictly monotone operator. Moreover,
$L$ is type
$(S_{+})$, i.e., if
$u_{n}\rightharpoonup u$ in
$W^{1,\mathcal{H}}_{0}(\Omega)$ and
$\limsup\limits_{n\to+\infty}\langle L(u_{n})-L(u),u_{n}-u\rangle\leq 0$, then
$u_{n}\to u$ in
$W^{1,\mathcal{H}}_{0}(\Omega)$.
2.2. Set-valued analysis
Next, let us recall some notions concerning partially ordered sets and set-valued maps.
Definition 2.3. Let
$(P,\leq)$ be a partially ordered set and let
$A\subseteq P$.
An element
$b$ is called lower bound of
$A$
if
$b\leq x$ for each
$x\in A$.
If
$b\in A$, we say that
$b$ is the
smallest element of
$A$.
An element
$x\in A$, is called minimal element,
if there is no
$y\neq x$ in
$A$ for which
$y\leq x$,
and every smallest element of
$A$ is minimal element of
$A$.
A subset
$C$ of
$P$ is said downward direct
if for each pair
$x,y\in C$ there is
$w\in C$
such that
$w\leq x$ and
$w\leq y$.
The following Kuratowski–Zorn Lemma can be found in Carl–Le–Motreanu [Reference Carl, Le and Motreanu16, Theorem 2.62]
Lemma 2.4. If in a partially ordered set
$P$,
every chain has a lower bound,
then
$P$ possesses a minimal element.
Definition 2.5. We say that a function
$G:X\to Y$ is compact if
$G$
is continuous and maps bounded sets into relatively
compact sets (i.e., sets whose closure is compact).
For more information concerning set-valued functions, we refer to Denkowski–Migórski–Papageorgiou [Reference Denkowski, Migórski and Papageorgiou39, Chapter 4], Gasiński–Papageorgiou [Reference Gasiński and Papageorgiou47, Chapters 3 and 7] or Migórski–Ochal–Sofonea [Reference Migórski, Ochal and Sofonea65, Chapter 3].
Definition 2.6. Let
$X$,
$Y$ be topological spaces. Consider a set-valued map
$F:X\to 2^{Y}$. We say that
•
$F$ is upper semi-continuous (briefly u.s.c.), if for every closed subset
$C\subset Y$, the set
\begin{equation*}
F^{-}(C)=\{x\in X:F(x)\cap C\neq\varnothing\}
\end{equation*}is closed in
$X$. Alternatively,
$F$ is u.s.c., if for every open subset
$O\subset Y$, the set
\begin{equation*}
F^{+}(O)=\{x\in X:F(x)\subset O\}
\end{equation*}is open in
$X$.•
$F$ is lower semi-continuous (briefly l.s.c.), if for every closed subset
$C\subset Y$, the set
\begin{equation*}
F^{+}(C)=\{x\in X:F(x)\subset C\}
\end{equation*}is closed in
$X$. Alternatively,
$F$ is l.s.c., for every open subset
$O\subset Y$, the set
\begin{equation*}
F^{-}(O)=\{x\in X:F(x)\cap O\neq\varnothing\}
\end{equation*}is open in
$X$.• We say that
$F$ is continuous if it is u.s.c. and l.s.c.
The following characterization of l.s.c. set-valued function holds (see e.g., Migórski–Ochal–Sofonea [Reference Migórski, Ochal and Sofonea65, Proposition 3.9], Liu–Motreanu–Zeng [Reference Liu, Motreanu and Zeng61, Proposition 5] or Baldelli–Guarnotta [Reference Baldelli and Guarnotta6, Lemma 4.4] and Klein–Thompson [Reference Klein and Thompson59, Theorem 7.1.7] for the proof).
Lemma 2.7. Let
$X$,
$Y$ be two topological spaces. Let
$F:X\to 2^{Y}$ be a set-valued function. Then,
$F$ is l.s.c. if and only if the following hold: if
$u\in X$,
$\{u_{\gamma}\}_{\gamma\in I}\subset X$ is a net such that
$u_{\gamma}\to u$ and
$u^{*}\in F(u)$, then for each
$\gamma\in I$ there is
$u_{\gamma}^{*}\in F(u_{\gamma})$ with
$u_{\gamma}^{*}\to u^{*}$ in
$Y$.
3. Frozen problem
In this section we fix
$k\in\{1,...,K\}$ and
$\alpha\in(\alpha_{2k-1},\alpha_{2k})$. Consider the following auxiliary frozen problem
\begin{align}
\left \{
\begin{array}{ll}
-a(\alpha)\mathcal{G}(u) = f(u) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{align}We recall the following definitions (see e.g., Guarnotta–Livrea–Winkert [Reference Guarnotta, Livrea and Winkert53, Definition 5.2] or Motreanu–Motreanu–Papageorgiou [Reference Motreanu, Motreanu and Papageorgiou66, Definition 11.5]) of sub- and super-solutions of the above problem. Consider
\begin{equation*}
L^{\mathcal{H}}_{+}(\Omega)=\left\{u\in L^{\mathcal{H}}(\Omega):u\geq 0\right\}.
\end{equation*}Definition 3.1. We say that
$\underline{u}\in W^{1,\mathcal{H}}(\Omega)$ is a sub-solution for (F α), if
$\underline{u}\leq 0$ on
$\partial\Omega$ and
\begin{align*}
a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \underline{u}\vert^{p-2}
\nabla \underline{u}+\vert\nabla \underline{u}\vert^{q-2}
\nabla \underline{u}\right)\nabla v dx
&\leq \int_{\Omega}f(x,\underline{u})vdx\\
& \quad\forall v\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L_{+}^{\mathcal{H}}(\Omega).
\end{align*} Analogously,
$\bar{u}\in W^{1,\mathcal{H}}(\Omega)$ is called super-solution for (F α), if
$\bar{u}\geq 0$ on
$\partial\Omega$ and
\begin{align*}
a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \bar{u}\vert^{p-2}
\nabla \bar{u}+\vert\nabla \bar{u}\vert^{q-2}
\nabla \bar{u}\right)\nabla v dx
&\geq \int_{\Omega}f(x,\bar{u})vdx\\
& \quad\forall v\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L_{+}^{\mathcal{H}}(\Omega).
\end{align*} Similarly,
$u\in W^{1,\mathcal{H}}_{0}(\Omega)$ is called (weak) solution for (F α), if
$u=0$ on
$\partial\Omega$ and
\begin{equation*}
a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla u\vert^{p-2}
\nabla u+\vert\nabla u\vert^{q-2}
\nabla u\right)\nabla v dx= \int_{\Omega}f(x,u)vdx
\quad\forall v\in W^{1,\mathcal{H}}_{0}(\Omega).
\end{equation*} In all three cases, we assume that the right-hand side is well defined, i.e.,
$f(\cdot, u(\cdot))\in L^{m'}(\Omega)$ for some
$m\in [1,q^{*})$.
Lemma 3.2. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold. Then (F α) admits a positive bounded sub-solution
$\underline{u}\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$.
Proof. To obtain a positive sub-solution for (F α), consider the torsion problem (T). It admits a unique, positive solution
$e_{1}\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$. Define
and consider
$\underline{u}=t(\alpha)e$ where
$e=e_{1}/\Vert e_{1}\Vert_{\infty}$. By the definition of
$t(\alpha)$, we have
$f(t(\alpha))\leq f(s)$ for all
$s\in(0,t(\alpha))$ and by
$(H_{4})$,
$t(\alpha)\le t_*\le\Vert e_1\Vert_{\infty}$. So
\begin{eqnarray*}
&&-a(\alpha)\operatorname{div}\left(\mu(x)\vert\nabla \underline{u}
\vert^{p-2}\nabla \underline{u}+\vert \nabla \underline{u}
\vert^{q-2}\nabla \underline{u}\right)\\
&\leq &-a(\alpha)\min\left\{\left(\frac{t(\alpha)}{\Vert e_{1}
\Vert_{\infty}}
\right)^{p-1},
\left(\frac{t(\alpha)}{\Vert e_{1}\Vert_{\infty}}
\right)^{q-1}\right\}\\ &&\times\operatorname{div}\left(\mu(x)\vert\nabla e_{1}
\vert^{p-2}\nabla e_{1} +\vert \nabla e_{1}\vert^{q-2}\nabla e_{1}\right)\\
&\leq & a(\alpha)=f(t(\alpha))\leq f(\underline{u}),
\end{eqnarray*} thus
$\underline{u}\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$ is a sub-solution of problem (F α).
Remark 3.3. Note that
$\overline{u}=t_{*}$ is a super-solution for (F α) and
$\underline{u}\le\overline{u}$.
Consider the following truncation on the reaction term
\begin{equation}
\hat{f}(x,s)
=\left \{\begin{array}{lll}
f(\underline{u}(x))&\text{if}& s \lt \underline{u}(x),\\
f(s)&\text{if}& \underline{u}(x)
\leq s\leq \overline{u}(x),\\
f(\overline{u}(x))&\text{if}& \overline{u}(x) \lt s,
\end{array}
\right.
\end{equation}and consider the associated truncated problem
\begin{align}
\left \{\begin{array}{ll}
-a(\alpha)\mathcal{G}(u) = \hat{f}(x,u) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{align}The following existence result for problem (FT α) holds.
Proposition 3.4. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold. Then (FT α) admits at least one positive solution
$u\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$. Moreover, each solution of (FT α) is such that
$\underline{u} \lt u \lt \overline{u}$.
Proof. The associated energy function is
\begin{equation}
J(u)=\frac{a(\alpha)}{p}\Vert \nabla u\Vert^{p}_{p,\mu}
+\frac{a(\alpha)}{q}\Vert \nabla u\Vert^{q}_{q}-\int_{\Omega}\hat{F}(u)dx,
\end{equation} where
$\hat{F}(t)=\displaystyle\int_{0}^{t}\hat{f}(s)ds$. Clearly,
$J$ is Gâteaux differentiable, coercive and l.s.c., then, by Tonelli–Weiestrass Theorem (see e.g., Costa [Reference Costa31, Theorem 1.2])
$J$ has a global minimum
$u$, which is obviously a weak solution of (FT α). As in Lemma 3.2,
$u\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$. Now, we prove that every solution
$u$ of (FT α) lies in
$[\underline{u},\overline{u}]$, that is,
$u$ is a solution of (F α) by (11). Since
$\underline{u}$ is a sub-solution, then
\begin{align}a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla\underline{u}
\vert^{p-2}\nabla \underline{u}+\vert \nabla
\underline{u}\vert^{q-2}\nabla \underline{u}\right)
\nabla vdx&\leq \int_{\Omega}\hat{f}(x,\underline{u})vdx \nonumber\\ &
\quad\forall v\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\mathcal{H}}_{+}(\Omega).\end{align} Moreover,
$\overline{u}$ is a super-solution, then
\begin{align}a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla\overline{u}\vert^{p-2}
\nabla \overline{u}+\vert \nabla\overline{u}\vert^{q-2}
\nabla \overline{u}\right)\nabla vdx &\geq \int_{\Omega}
\hat{f}(x,\overline{u})vdx \nonumber\\ &
\quad\forall v\in W^{1,\mathcal{H}}_{0}(\Omega)
\cap L^{\mathcal{H}}_{+}(\Omega).\end{align} Finally,
$u$ is a solution of (FT α), then
\begin{equation}a(\alpha)\int_\Omega\left(\mu(x)\vert\nabla u\vert^{p-2}\nabla u+\vert\nabla u\vert^{q-2}\nabla u\right)\nabla vdx=\int_\Omega\widehat f(x,u)vdx\quad\forall v\in W_0^{1,\mathcal H}(\Omega).\end{equation} Next, subtratting (13) and (15) and choosing
$v=(\underline{u}-u)^{+}$, we get
\begin{align}
& a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \underline{u}\vert^{p-2}
\nabla \underline{u}+\vert \nabla \underline{u}\vert^{q-2}
\nabla \underline{u}-\mu(x)\vert\nabla u\vert^{p-2}\nabla u
+\vert \nabla u\vert^{q-2}\nabla u\right)\nonumber\\&\quad\times\nabla (\underline{u}-u)^{+}dx\leq 0.
\end{align}Recalling (9) and strictly monotonicity of double phase operator (Proposition 2.2), we infer
Then, we obtain that the sub-level
$\left\{\underline{u} \gt u\right\}$ has null Lebesgue measure, that is
$u\geq \underline{u}$. Similarly, subtracting (15) and (14) and choosing
$v=(u-\overline{u})^{+}$, we deduce that
$u\leq \overline{u}$.
4. Minimal solution
In this section we fix
$k\in\{1,\ldots,K\}$. Set
$X=W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\infty}(\Omega)$. Following the ideas of Baldelli–Guarnotta [Reference Baldelli and Guarnotta6], Faraci–Motreanu–Puglisi [Reference Faraci, Motreanu and Puglisi40], and Liu–Motreanu–Zeng [Reference Liu, Motreanu and Zeng61], we consider the following set-valued map
$\mathscr{S}_k:(\alpha_{2k-1},\alpha_{2k})\to 2^{X}$,
Note that, by Proposition 3.4,
$\mathscr{S}_k$ is well defined, i.e., it has nonempty values.
Proposition 4.1. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold. Then,
$\mathscr{S}_k$ maps convergent sequences in
$(\alpha_{2k-1},\alpha_{2k})$ into sequences with convergent subsequences in
$X$.
Proof. Let
$\{\alpha_{n}\}\subset(\alpha_{2k-1},\alpha_{2k})$ be a sequence such that
$\alpha_{n}\to\alpha\in(\alpha_{2k-1},\alpha_{2k})$ and
$u_{n}\in\mathscr{S}_k(\alpha_{n})$. We prove that, up to a subsequence
$u_{n}\to u\in \mathscr{S}_k(\alpha)$. First, by definition of
$\mathscr{S}_k$, we have that, for all
$n\in\mathbb{N}$,
$u_{n}$ is a solution of
\begin{equation}
\left \{\begin{array}{ll}
-a(\alpha_{n})\mathcal{G}(u_{n})
= \hat{f}(x,u_{n}) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega,
\end{array}
\right.
\end{equation} such that
$u_{n}\geq \underline{u}$. Note that,
$\{u_{n}\}_{n\in\mathbb{N}}$ is a bounded sequence. Indeed
\begin{align*}
& a(\alpha_{n})\int_{\Omega}\left(\mu(x)\vert\nabla u_{n}\vert^{p-2}
\nabla u_{n}+\vert \nabla u_{n}\vert^{q-2}\nabla u_{n}\right)\nabla vdx
\\ &\quad=\int_{\Omega}\hat{f}(x,u_{n})vdx
\quad\forall v\in W^{1,\mathcal{H}}_{0}(\Omega).
\end{align*} That is, by (9) and
$(H_{3})$,
\begin{equation}
a(\alpha_{n})\int_{\Omega}\left(\mu(x)\vert\nabla u_{n}
\vert^{p-2}\nabla u_{n}
+\vert \nabla u_{n}\vert^{q-2}\nabla u_{n}\right)\nabla vdx
\leq \vert\Omega\vert\max\limits_{t\in[0,t_{*}]}[tf(t)]
\end{equation} and
$\Vert u_{n}\Vert \lt +\infty$
$\forall n\in\mathbb{N}$. Then, up to a subsequence, we have
$u_{n}\rightharpoonup u$ in
$W^{1,\mathcal{H}}_{0}(\Omega)$ and
$u_{n}\to u$ in
$L^{\mathcal{H}}(\Omega)$ (remember that
$W^{1,\mathcal{H}}_{0}(\Omega)\hookrightarrow L^{q}(\Omega)$ compactly for
$q \lt p^{*}$ and
$L^{q}(\Omega)\hookrightarrow L^{\mathcal{H}}(\Omega)$ continuously, see (9) and Colasuonno–Squassina [Reference Colasuonno and Squassina27, Proposition 2.18]). Then, we have
\begin{equation}
\lim\limits_{n\to+\infty}\langle \mathcal{G}(u_{n})
-\mathcal{G}(u),u_{n}-u\rangle=0,
\end{equation} and by the
$(S_{+})$-property of
$\mathcal{G}$ (see Lemma 2.2),
$u_{n}\to u$ in
$W^{1,\mathcal{H}}_{0}(\Omega)$. Moreover, by the Lebesgue Dominate Convergence Theorem (see e.g., Brézis [Reference Brézis11, Theorem 4.2]), passing to the limit in (19),
$u$ is a solution of
\begin{equation}
\left \{\begin{array}{ll}
-a(\alpha)\mathcal{G}(u) = \hat{f}(x,u) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{equation} such that
$u\geq \underline{u}$, i.e.,
$u\in\mathscr{S}_k(\alpha)$.
Proposition 4.2. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold.
Then,
$\mathscr{S}_k$ is l.s.c.
Proof. We use the characterization of l.s.c. contained in Lemma 2.7. Let
$\{\alpha_{n}\}\subset (\alpha_{2k-1},\alpha_{2k})$ be a sequence such that
$\alpha_{n}\to\alpha\in(\alpha_{2k-1},\alpha_{2k})$. Let
$v\in\mathscr{S}_k(\alpha)$ and for all
$n\in\mathbb{N}$, consider the problem
\begin{align}
\left \{\begin{array}{ll}
-a(\alpha_n)\mathcal{G}(u) = \hat{f}(x,v) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{align} Clearly
$v\geq \underline{u}$ by definition of
$\mathscr{S}_k(\alpha)$. Moreover, by Minty–Browder’s Theorem (see e.g., [Reference Brézis11, Theorem 5.16]), (FT αnv) has a unique positive solution
$u_{n}^{0}$ for all
$n\in\mathbb{N}$. Note that, as in Proposition 4.1, since
$\alpha_{n}\to \alpha\in(\alpha_{2k-1},\alpha_{2k})$, then
$u_{n}^{0}$ is relatively compact in
$X$ and, up to a sub-sequence,
$u_{n_{k}}^{0}\to u$ in
$X$ as
$k\to+\infty$, being
$u$ the unique solution of
\begin{equation*}
\left \{\begin{array}{ll}
-a(\alpha)\mathcal{G}(u) = \hat{f}(x,v) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{equation*} Since
$v\in\mathscr{S}_k(\alpha)$, by the definition of
$\mathscr{S}_k$,
$v$ is a solution of
\begin{equation*}
\left \{\begin{array}{ll}
-a(\alpha)\mathcal{G}(v) = \hat{f}(x,v) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{equation*} Thus, it follows that
$u=v$.
Since each subsequence
$\{u_{n_{k}}^{0}\}$
of
$\{u_{n}^{0}\}$ converges to the same limit
$v$, then
in fact
\begin{equation*}
\lim\limits_{n\to+\infty}u_{n}^{0}=v.
\end{equation*} Again, for each
$n\in\mathbb{N}$ consider the problem
\begin{align}
\left \{\begin{array}{ll}
-a(\alpha_{n})\mathcal{G}(u)
= \hat{f}(x,u_{n}^{0}) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{align} As above, there exists a unique solution
$u_{n}^{1}$
of (FT αnu n0), for each
$n\in\mathbb{N}$, such that
\begin{equation*}
\lim\limits_{n\to+\infty}u_{n}^{1}=v.
\end{equation*} By iterating this process, for any
$k\geq 1$,
we get a sequence
$\{u_{n}^{k}\}$,
such that
$u_{n}^{k}$ is a solution of
\begin{align}
\left \{\begin{array}{ll}
-a(\alpha_{n})\mathcal{G}(u_{n}^{k})
= \hat{f}(x,u_{n}^{k-1}) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega,
\end{array}
\right.
\end{align}and
\begin{equation}
\lim\limits_{n\to+\infty}u_{n}^{k}=v\quad\forall k\in\mathbb{N}.
\end{equation} Fix
$n\geq 1$, since
$\{u_{n}^{k}\}$
is relatively compact in
$X$,
by Proposition 4.1,
we have that
that is, it solves
\begin{align}
\left \{\begin{array}{ll}
-a(\alpha_{n})\mathcal{G}(u_{n})
= \hat{f}(x,u_{n}) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega,
\end{array}
\right.
\end{align} and
$u_{n}\geq\underline{u}$,
i.e.,
$u_{n}\in\mathscr{S}_k(\alpha_{n})$.
By compact embedding (as in Proposition 4.1),
(23),
and double limit lemma
(see e.g., Gasiński–Papageorgiou
[Reference Gasiński and Papageorgiou47, Proposition A.2.35]),
we have that
$u_{n}\to v$
in
$X$.
Thus, by Lemma 2.7 we conclude that
$\mathscr{S}_k$
is l.s.c.
The following is a technical lemma that will allow us to obtain the downward directness of
$\mathscr{S}_k$.
Lemma 4.3. If
$\overline{u}_{1},\overline{u}_{2}\in W^{1,\mathcal{H}}(\Omega)$ are two super-solutions of (F α), then
$\min\{\overline{u}_{1},\overline{u}_{2}\}\in W^{1,\mathcal{H}}(\Omega)$ is also a super-solution for (F α).
Proof. Let
$\overline{u}_{1},\overline{u}_{2}\in W^{1,\mathcal{H}}(\Omega)$
be two super-solutions for (F α).
Let
$\varepsilon \gt 0$ and consider the following Lipschitz continuous truncation function
$\eta_{\varepsilon}:\mathbb{R}\to\mathbb{R}$, defined by
\begin{equation}
\eta_{\varepsilon}(t)
=\left \{
\begin{array}{lll}
0 & \textrm{if} & t\leq 0;\\
\frac{t}{\varepsilon} & \textrm{if} & 0 \lt t \lt \varepsilon;\\
1 & \textrm{if} & t\geq\varepsilon.
\end{array}
\right.
\end{equation} Note that, by Adams–Fournier [Reference Adams and Fournier1, Lemma 8.34]
or Ziemer [Reference Ziemer80, Theorem 2.1.11],
a direct computation and having in mind that
$W^{1,\mathcal{H}}(\Omega)\hookrightarrow W^{1,q}(\Omega)$,
we can shows that,
$\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1})
\in W^{1,\mathcal{H}}(\Omega)$,
and, by the Lebesgue Dominate Convergence Theorem
(see e.g., Brézis [Reference Brézis11, Theorem 4.2]), we get
\begin{equation}
\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1})(x)
\to \chi_{\{\overline{u}_{1} \lt \overline{u}_{2}\}}(x)\text{ for a.a. }x\in\Omega
\text{ as }\varepsilon\to 0^{+},
\end{equation} where
$\chi_{E}$ is the characteristic function of the set
$E$.
Moreover, by chain rule, we observe that
Now, by Definition 3.1, we have
\begin{align*}
a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \overline{u}_{1}
\vert^{p-2}\nabla \overline{u}_{1}+\vert\nabla
\overline{u}_{1}\vert^{q-2}\nabla \overline{u}_{1}
\right)\nabla\varphi_{1}dx
&\geq \int_{\Omega}f(x,\overline{u}_{1})
\varphi_{1}dx\\ &\quad\forall\varphi_{1}
\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\mathcal{H}}_{+}(\Omega),
\end{align*}and
\begin{align*}
a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \overline{u}_{2}
\vert^{p-2}\nabla \overline{u}_{2}+\vert\nabla
\overline{u}_{2}\vert^{q-2}\nabla \overline{u}_{2}
\right)\nabla\varphi_{2}dx
&\geq \int_{\Omega}f(x,\overline{u}_{2})
\varphi_{2}dx\\ &\quad\forall\varphi_{2}
\in W^{1,\mathcal{H}}_{0}(\Omega)\cap L^{\mathcal{H}}_{+}(\Omega),
\end{align*} Choosing
$\varphi_{1}
=\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1})v$
and
$\varphi_{2}=(1-\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1}))v$
for
$v\in C_{0}^{\infty}(\Omega)_{+}$,
and adding the two previous inequalities, we get
\begin{eqnarray*}
&&a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \overline{u}_{1}
\vert^{p-2}\nabla \overline{u}_{1}+\vert\nabla
\overline{u}_{1}\vert^{q-2}\nabla \overline{u}_{1}\right)
\nabla\big[\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1})v\big]dx\cr
&&+a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \overline{u}_{2}
\vert^{p-2}\nabla \overline{u}_{2}+\vert\nabla
\overline{u}_{2}\vert^{q-2}\nabla \overline{u}_{2}\right)
\nabla\big[(1-\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1}))v\big]dx\cr
&\geq &\int_{\Omega}f(x,\overline{u}_{1})
\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1})vdx
+\int_{\Omega}f(x,\overline{u}_{2})
(1-\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1}))vdx.
\end{eqnarray*}By direct computation, we have
\begin{eqnarray*}
&&a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \overline{u}_{1}
\vert^{p-2}\nabla \overline{u}_{1}+\vert\nabla \overline{u}_{1}
\vert^{q-2}\nabla \overline{u}_{1}\right)
\nabla\big[\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1})v\big]dx\cr
&=&a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla \overline{u}_{1}
\vert^{p-2}\nabla \overline{u}_{1}+\vert\nabla \overline{u}_{1}
\vert^{q-2}\nabla \overline{u}_{1}\right)
\nabla\eta_{\varepsilon}
(\overline{u}_{2}-\overline{u}_{1})\cdot vdx\cr
&&+a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla
\overline{u}_{1}\vert^{p-2}
\nabla \overline{u}_{1}+\vert\nabla \overline{u}_{1}\vert^{q-2}
\nabla \overline{u}_{1}\right)\nabla v \cdot
\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1})dx\cr
&=&\frac{a(\alpha)}{\varepsilon}\int
\limits_{\{0 \lt \overline{u}_{2}
-\overline{u}_{1} \lt \varepsilon\}}\left(\mu(x)
\vert\nabla \overline{u}_{1}\vert^{p-2}
\nabla \overline{u}_{1}+\vert\nabla \overline{u}_{1}\vert^{q-2}
\nabla \overline{u}_{1}\right)\nabla(\overline{u}_{2}
-\overline{u}_{1})\cdot vdx\cr
&&+a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla
\overline{u}_{1}\vert^{p-2}\nabla \overline{u}_{1}
+\vert\nabla \overline{u}_{1}\vert^{q-2}
\nabla \overline{u}_{1}\right)
\nabla v \cdot\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1})dx,
\end{eqnarray*}and
\begin{eqnarray*}
&&a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla
\overline{u}_{2}\vert^{p-2}
\nabla \overline{u}_{2}+\vert\nabla \overline{u}_{2}
\vert^{q-2}\nabla \overline{u}_{2}\right)\nabla
\big[(1-\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1}))v\big]dx\cr
&=&a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla
\overline{u}_{2}\vert^{p-2}\nabla \overline{u}_{2}
+\vert\nabla \overline{u}_{2}\vert^{q-2}
\nabla \overline{u}_{2}\right)
\nabla(1-\eta_{\varepsilon}
(\overline{u}_{2}-\overline{u}_{1}))\cdot vdx\cr
&&+a(\alpha)\int_{\Omega}\left(\mu(x)
\vert\nabla \overline{u}_{2}\vert^{p-2}
\nabla \overline{u}_{2}+\vert\nabla
\overline{u}_{2}\vert^{q-2}\nabla \overline{u}_{2}\right)\nabla v
\cdot(1-\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1}))dx\cr
&=&-\frac{a(\alpha)}{\varepsilon}
\int\limits_{\{0 \lt \overline{u}_{2}-\overline{u}_{1} \lt \varepsilon\}}
\left(\mu(x)\vert\nabla \overline{u}_{2}\vert^{p-2}
\nabla \overline{u}_{2}+\vert\nabla \overline{u}_{2}\vert^{q-2}
\nabla \overline{u}_{2}\right)
\nabla(\overline{u}_{2}-\overline{u}_{1})
\cdot vdx\cr
&&+a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla
\overline{u}_{2}\vert^{p-2}
\nabla \overline{u}_{2}+\vert\nabla
\overline{u}_{2}\vert^{q-2}\nabla \overline{u}_{2}\right)\nabla v
\cdot(1-\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1}))dx.
\end{eqnarray*}Then, we infer
\begin{eqnarray*}
&&a(\alpha)\int_{\Omega}\left(\mu(x)\vert \nabla
\overline{u}_{1}\vert^{p-2}\nabla \overline{u}_{1}
+\vert\nabla \overline{u}_{1}\vert^{q-2}
\nabla \overline{u}_{1}\right)
\nabla v \cdot \eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1})dx\cr
&&+a(\alpha)\int_{\Omega}\left(\mu(x)\vert\nabla
\overline{u}_{2}\vert^{p-2}
\nabla \overline{u}_{2}+
\vert\nabla \overline{u}_{2}\vert^{q-2}
\nabla \overline{u}_{2}\right)\nabla v
\cdot (1-\eta_{\varepsilon}(\overline{u}_{2}
-\overline{u}_{1}))dx\cr
&\geq &\int_{\Omega}f(x,\overline{u}_{1})
\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1})vdx
+\int_{\Omega}f(x,\overline{u}_{2})
(1-\eta_{\varepsilon}(\overline{u}_{2}-\overline{u}_{1}))vdx.
\end{eqnarray*} By (25),
passing to the limit as
$\varepsilon\to 0^{+}$, we get
\begin{eqnarray*}
&&a(\alpha)\int\limits_{\{\overline{u}_{1} \lt \overline{u}_{2}\}}\left(\mu(x)\vert
\nabla \overline{u}_{1}\vert^{p-2}
\nabla \overline{u}_{1}+\vert\nabla \overline{u}_{1}
\vert^{q-2}\nabla \overline{u}_{1}\right)\nabla vdx\cr
&&+a(\alpha)\int\limits_{\{\overline{u}_{1}
\geq\overline{u}_{2}\}}\left(\mu(x)\vert
\nabla \overline{u}_{2}\vert^{p-2}
\nabla \overline{u}_{2}+\vert\nabla \overline{u}_{2}
\vert^{q-2}\nabla \overline{u}_{2}\right)\nabla vdx\cr
&\geq &\int\limits_{\{\overline{u}_{1} \lt \overline{u}_{2}\}}f(x,\overline{u}_{1})vdx
+\int\limits_{\{\overline{u}_{1}
\geq\overline{u}_{2}\}}f(x,\overline{u}_{2})vdx
\end{eqnarray*} for all
$v\in C^{\infty}_{0}(\Omega)_{+}$.
Recall that, since
$\overline{u}_{1},\overline{u}_{2}\in W^{1,\mathcal{H}}(\Omega)$,
then
$\overline{u}=\min\{\overline{u}_{1},
\overline{u}_{2}\}\in W^{1,\mathcal{H}}(\Omega)$, indeed,
since if
$\overline{u}_{1},\overline{u}_{2}
\in W^{1,\mathcal{H}}(\Omega)$,
then,
$\overline{u}_{1},\overline{u}_{2}\in W^{1,q}(\Omega)$,
so, bearing in mind relation (7), we have
\begin{equation}
\rho_{\mathcal{H}}(u)=\int_{\Omega}\left(\mu(x)\vert
\min\{\overline{u}_{1},
\overline{u}_{2}\}\vert^{p}
+\vert \min\{\overline{u}_{1},\overline{u}_{2}\}
\vert^{q}\right)dx \lt +\infty,
\end{equation}and
\begin{equation*}
\rho_{\mathcal{H}}(\nabla u)=\int_{\Omega}\left(\mu(x)\vert\nabla
\min\{\overline{u}_{1},\overline{u}_{2}\}\vert^{p}
+\vert \nabla \min\{\overline{u}_{1},
\overline{u}_{2}\}\vert^{q}\right)dx \lt +\infty.
\end{equation*} Moreover,
$\overline{u}\geq 0$ on
$\partial\Omega$, and
\begin{equation}
\nabla\overline{u}(x)=
\left \{\begin{array}{ll}
\nabla \overline{u}_{1}(x) &\text{for a.a. }
x\in \{\overline{u}_{1} \lt \overline{u}_{2}\},\\
\nabla \overline{u}_{2}(x) &\text{for a.a. }
x\in \{\overline{u}_{1}\geq\overline{u}_{2}\}.
\end{array}
\right.
\end{equation}Thus, we obtain that
\begin{eqnarray*}
a(\alpha)\int_{\Omega}\left(\mu(x)\vert \nabla \overline{u}
\vert^{p-2}\nabla \overline{u}+\vert\nabla\overline{u}\vert^{q-2}
\nabla \overline{u}\right)\nabla vdx
\geq \int_{\Omega}f(x,\overline{u})vdx
\quad\forall v\in C_{0}^{\infty}(\Omega)_{+}.
\end{eqnarray*} Finally, since
$C_{0}^{\infty}(\Omega)_{+}$
is dense in
$W^{1,\mathcal{H}}_{0}(\Omega)_{+}$
(see (8)),
then the last inequality holds for any
$v\in W^{1,\mathcal{H}}_{0}(\Omega)
\cap L^{\mathcal{H}}_{+}(\Omega)$
and we conclude the proof.
Proposition 4.4. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold. Then, for any
$\alpha\in(\alpha_{2k-1},\alpha_{2k})$, the set
$\mathscr{S}_k(\alpha)$ is downward direct.
Proof. Fix
$\alpha\in(\alpha_{2k-1},\alpha_{2k})$
and let
$u_{1},u_{2}\in\mathscr{S}_k(\alpha)$.
Consider
$u=\min\{u_{1},u_{2}\}$.
Since
$u_{1},u_{2}$ are solution of (FT α)
they are also super-solutions of (FT α).
Then, by Lemma 4.3,
$u$ is a super-solution for (FT α).
Let us consider the following truncation
\begin{equation}
f_{*}(x,s)=
\left
\{\begin{array}{lll}
\hat{f}(x,\underline{u})
& \text{if} & s \lt \underline{u}(x);\\
\hat{f}(x,s)
& \text{if} & \underline{u}(x)\leq s\leq u(x);\\
\hat{f}(x,u)
& \text{if} & s \gt u(x),
\end{array}
\right.
\end{equation}and the associated problem
\begin{align}
\left \{\begin{array}{ll}
-a(\alpha)\mathcal{G}(u) = f_{*}(x,u) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega.
\end{array}
\right.
\end{align} Now, as in Proposition 3.4,
there exists
$\tilde{u}$ solution of (FTT α),
such that
$\underline{u}\leq\tilde{u}\leq u$
(choose
$v=(\tilde{u}-u)^{+}$ in the weak formulation).
Then,
$\tilde{u}$ is obviously
a solution of (FT α)
and so
$\tilde{u}\in\mathscr{S}_k(\alpha)$.
Theorem 4.5. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold. Then, for any
$\alpha\in(\alpha_{2k-1},\alpha_{2k})$, the set
$\mathscr{S}_k(\alpha)$ has the smallest element
$u_{\alpha}^{*}$.
Proof. By Proposition 4.4,
for each
$\alpha\in(\alpha_{2k-1},\alpha_{2k})$,
we have that
$\mathscr{S}_k(\alpha)$
is partially ordered and downward direct.
Let
$\mathcal{C}\subset \mathscr{S}_k(\alpha)$ a chain.
We can find a sequence
$\{u_{n}\}\subset\mathcal{C}$ such that
\begin{equation*}
\lim\limits_{n\to+\infty}u_{n}=\inf\mathcal{C}.
\end{equation*} Moreover, since
$u_{n}\geq\underline{u}$
and it is a solution of (FT α),
arguing as in
Proposition 4.1,
up to a sub-sequence, there exists
$v\in X$ such that
and
$v\geq \underline{u}$.
Therefore,
$v=\inf\mathcal{C}$,
so by Kuratowski–Zorn Lemma (Lemma 2.4),
there exists a minimal element
$u_{\alpha}^{*}$ for
$\mathscr{S}_k(\alpha)$.
Now, we show that
$u_{\alpha}^{*}$ is the smallest solution
of (FT α) such that
$u_{\alpha}^{*}\geq \underline{u}$.
Let
$u\in\mathscr{S}_k(\alpha)$.
By Proposition 4.4,
there exists
$\tilde{u}\in\mathscr{S}_k(\alpha)$ such that
$\tilde{u}\leq\min\{u_{\alpha}^{*},u\}$.
From the minimality of
$u_{\alpha}^{*}\in\mathscr{S}_k(\alpha)$,
we get
$\underline{u}\leq u_{\alpha}^{*}\leq \tilde{u}\leq u$,
that is,
$u_{\alpha}^{*}$ is the smallest solution
of (FT α) such that
$u_{\alpha}^{*}\geq \underline{u}$.
By the use of Theorem 4.5, we can define
the function
$\Gamma:(\alpha_{2k-1},\alpha_{2k})
\subset\mathbb{R}\to X$,
by
where
$u_{\alpha}^{*}$ is the
smallest solution of (FT α)
such that
$u_{\alpha}^{*}\geq \underline{u}$
(its uniqueness comes from the definition of the smallest element
in a partially ordered set).
Proposition 4.6. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold.
Then, the function
$\Gamma$ given by (29) is continuous.
Proof. Consider a sequence
$\{\alpha_{n}\}\subset(\alpha_{2k-1},\alpha_{2k})$
such that
$\alpha_{n}\to\alpha\in (\alpha_{2k-1},\alpha_{2k})$.
Then, we have
\begin{align}
\left \{\begin{array}{cl}
-a(\alpha_{n})\mathcal{G}(u_{n})
= \hat{f}(x,u_{n})& \textrm{in} \ \Omega,\\
u_{n} \gt 0 & \textrm{in}\ \Omega,\\
u_{n}=0 & \textrm{on}\ \partial\Omega,
\end{array}
\right.
\end{align} where
$u_{n}=\Gamma(\alpha_{n})$.
By compactness of
$\mathscr{S}_k$ (see Proposition 4.1),
we have that the sequence
$\{u_{n}\}$
is relatively compact in
$X$.
Thus, passing to a subsequence,
$u_{n}\to u$ in
$X$
for some
$u\in X$,
and, since
$u_{n}\geq \underline{u}$ for all
$n\in\mathbb{N}$,
we get
$u\geq \underline{u}$.
Passing to the limit in ((FT αnu n), we get
\begin{equation}
\left \{\begin{array}{cl}
-a(\alpha)\mathcal{G}(u) = \hat{f}(x,u) & \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in} \ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega,
\end{array}
\right.
\end{equation} so
$u\in\mathscr{S}_k(\alpha)$.
Now, by Lemma 2.7, there exists a sequence
$\{v_{n}\}\subset X$
such that
$v_{n}\in\mathscr{S}_k(\alpha_{n})$ for all
$n\in\mathbb{N}$
and
$v_{n}\to \Gamma(\alpha)\in\mathscr{S}_k(\alpha)$.
Then, by the minimality of
$\Gamma(\alpha)$,
we conclude that
\begin{equation*}
\Gamma(\alpha)
=u_{\alpha}^{*}\leq u
=\lim\limits_{n\to +\infty}u_{n}
=\lim\limits_{n\to +\infty}\Gamma(\alpha_{n})
\leq \lim\limits_{n\to +\infty}v_{n}
=\Gamma(\alpha).
\end{equation*} Thus
$\Gamma(\alpha)=u_{\alpha}^{*}
=\lim\limits_{n\to+\infty}\Gamma(\alpha_{n})$.
5. Fixed Point
We associate with (P)
an auxiliary one dimension fixed-point problem.
By the results of the previous section,
for each
$k\in\{1,...,K\}$ and
$\alpha\in (\alpha_{2k-1},\alpha_{2k})$
there exists
$u_{\alpha}^{*}$,
the (unique) smallest solution of (F α)
in
$[\underline{u},\overline{u}]$.
Thus, we can define the map
$\mathcal{P}_{k}:(\alpha_{2k-1},\alpha_{2k})\to\mathbb{R}$ by
\begin{eqnarray*}
\mathcal{P}_{k}(\alpha)=\int_{\Omega}\left((u_{\alpha}^{*})^{r}
+\nu(x)(u_{\alpha}^{*})^{s}\right) dx
\quad\forall \alpha\in (\alpha_{2k-1},\alpha_{2k}).
\end{eqnarray*}It plays a key role in our approach since we have that
where
$\operatorname{Fix}(\mathcal{P}_{k})
=\left\{\alpha\in (\alpha_{2k-1},
\alpha_{2k}):\mathcal{P}_{k}(\alpha)=\alpha\right\}$.
Indeed, if
$\alpha\in (\alpha_{2k-1},\alpha_{2k})$
is such that
$\mathcal{P}_{k}(\alpha)=\alpha$, then
\begin{equation*}
-a\left(\int_{\Omega}\left((u_{\alpha}^{*})^{r}+\nu(x)(u_{\alpha}^{*})^{s}\right)dx
\right)\mathcal{G}(u_{\alpha}^{*})
=-a(\alpha)\mathcal{G}(u_{\alpha}^{*})=f(u_{\alpha}^{*})
\quad\operatorname{in}\quad\Omega.
\end{equation*}Proposition 5.1. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$ and
$(H_{4})$ hold.
Then
$\mathcal{P}_{k}$ is a continuous function.
Proof. Note that
$\mathcal{P}_{k}(\alpha)
=\Vert \Gamma(\alpha)\Vert^{r}_{r}
+\Vert\Gamma(\alpha)\Vert_{s,\nu}^{s}$.
Let
$\{\alpha_{n}\}\subset (\alpha_{2k-1},\alpha_{2k})$
be a sequence such that
$\alpha_{n}\to \alpha\in (\alpha_{2k-1},\alpha_{2k})$.
Since
$\Gamma(\alpha_{n})=u_{n}$ is a solution of
(FT α),
then
$u_{n}\leq t_{*}$ for each
$n\in\mathbb{N}$.
Now, combining the Lebesgue Dominate Convergence Theorem
(see e.g., Brézis [Reference Brézis11, Theorem 4.2])
and the continuity of
$\Gamma$
(see Proposition 4.6),
we get
\begin{align}
\lim\limits_{n\to+\infty}\mathcal{P}_{k}(\alpha_{n})
& =\lim\limits_{n\to+\infty}\int_{\Omega}\left(
\vert\Gamma(\alpha_{n})\vert^{r}+\nu(x)
\vert\Gamma(\alpha_{n})\vert^{s}\right)dx \nonumber\\
& =\int_{\Omega}\left(
\vert\Gamma(\alpha)\vert^{r}
+\nu(x)\vert\Gamma(\alpha)\vert^{s}\right)dx
=\mathcal{P}_{k}(\alpha).
\end{align}Proposition 5.2. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_{3})$, and
$(H_{4})$ hold. Then,
\begin{equation}
\lim\limits_{\alpha\to\alpha_{2k-1}^{+}}\mathcal{P}_{k}(\alpha) \gt \alpha_{2k-1}\text{ and }
\lim\limits_{\alpha\to\alpha_{2k}^{-}}
\mathcal{P}_{k}(\alpha) \gt \alpha_{2k}.
\end{equation}Proof. We show the first estimate (the second being analogous). First, by Lemma 3.2,
for all
$k\in\{1,...,K\}$ and
$\alpha\in(\alpha_{2k-1},\alpha_{2k})$,
we have
\begin{align}
\mathcal{P}_{k}(\alpha)&=\int_{\Omega}\left((u_{\alpha}^{*})^{r}
+\nu(x)(u_{\alpha}^{*})^{s}\right) dx
\geq \int_{\Omega}\underline{u}^{r}dx+\min\limits_{\Omega}\nu
\int_{\Omega}\underline{u}^{s}dx\nonumber\\ &=t(\alpha)^{r}\Vert e\Vert_{r}^{r}
+t(\alpha)^{s}\min\limits_{\Omega}\nu \Vert e\Vert_{s}^{s}.
\end{align} By the definition of
$t(\alpha)$ (see (10))
and by hypothesis
$(H_{f})$,
we have that
\begin{equation}
\lim\limits_{\alpha\to\alpha_{2k-1}^{+}}t(\alpha)
=\lim\limits_{\alpha\to\alpha_{2k}^{-}}t(\alpha)
=t_{*}.
\end{equation} Indeed, by hypothesis
$(H_{f})$,
the map
$t\mapsto f(t)$
is strictly positive in
$(0,t_{*})$,
it possesses a zero value at
$t=t_{*}$
and
\begin{equation*}
\liminf_{t\to 0^{+}}f(t) \gt \max\limits_{t\in[\alpha_{2k-1},\alpha_{2k}]}a(t)
\geq a(\alpha).
\end{equation*} Arguing by contradiction, assume that
$t(\alpha)\nrightarrow t_{*}$ as
$\alpha\to\alpha_{2k-1}^{+}$.
Recall that
$a(\alpha)\to 0$ as
$\alpha\to\alpha_{2k-1}^{+}$
and so, there exists an accumulation point
$\bar{t} \lt t_{*}$ of
$t(\alpha)$,
i.e., there exists a sequence
$\{\alpha_n\}\subseteq [\alpha_{2k-1},\alpha_{2k}]$,
such that
$\alpha_n\to \alpha_{2k-1}^{+}$,
$t(\alpha_n)\to \bar{t}$ (obviously
$\bar{t} \gt 0$).
But, by the continuity of
$f$ and
$a$,
this means that
$f(t)=\lim\limits_{n\to +\infty}
a(\alpha_n)=0$ (see (10)),
so
$f(\bar{t})=0$, a contradiction
to hypotheses
$(H_{f})$. This prove (35).
Finally, passing to the limit in (34)
as
$\alpha\to\alpha_{2k-1}^{+}$, by
$(H_{4})$, we get
\begin{equation*}
\lim\limits_{\alpha\to\alpha_{2k-1}^{+}}\mathcal{P}_{k}(\alpha)
\geq t_{*}^{r}\Vert e\Vert_{r}^{r}+t_{*}^{s}\min\limits_{\Omega}\nu
\Vert e\Vert_{s}^{s} \gt \alpha_{2K}.
\end{equation*}Proposition 5.3. Assume that
$(H_{1})$,
$(H_{2})$,
$(H_3)$,
$(H_{4})$ and
$(H_{5})$ hold. Then, for each
$k\in\{1,...,K\}$, there exists
$\tilde{\alpha}\in(\alpha_{2k-1},\alpha_{2k})$ such that
Proof. Let us start with the following auxiliary problem
\begin{equation}
\left \{
\begin{array}{ll}
- \Delta_{q}w= (u_{\alpha}^*)^{r-1}+\nu(x)
(u_{\alpha}^*)^{s-1}& \textrm{in}\ \Omega,\\
u \gt 0 & \textrm{in}\ \Omega,\\
u=0 & \textrm{on}\ \partial\Omega,
\end{array}
\right.
\end{equation} where
$u_{\alpha}^{*}\in W^{1,\mathcal{H}}_{0}(\Omega)
\cap L^{\infty}(\Omega)$ the unique minimal solution of (F α). By Minty–Browder’s Theorem (see e.g., Brézis [Reference Brézis11, Theorem 5.16]), there exists a unique positive solution
$w\in W^{1,q}_{0}(\Omega)\cap L^{\infty}(\Omega)$ of (36). Recalling that the embedding
$W^{1,\mathcal{H}}_{0}(\Omega)\hookrightarrow W^{1,q}_{0}(\Omega)$ is continuous, choosing
$v=u_{\alpha}$ as a test function in weak formulation of (36) and applying Hölder’s inequality, we have
\begin{align}\mathcal{P}_{k}(\alpha) &= \int_{\Omega} \left((u_{\alpha}^{*})^{r} + \nu(x)(u_{\alpha}^{*})^{s}\right) \, dx \nonumber \\
&= \int_{\Omega} |\nabla w|^{q-2} \nabla w \nabla u_{\alpha}^{*} \, dx \nonumber \\
&\leq \int_{\Omega} |\nabla w|^{q-1} |\nabla u_{\alpha}^{*}| \, dx \nonumber \\
&\leq \left( \int_{\Omega} |\nabla w|^{(q-1)q'} \, dx \right)^{\frac{1}{q'}} \left( \int_{\Omega} |\nabla u_{\alpha}^{*}|^{q} \, dx \right)^{\frac{1}{q}} \nonumber \\
&= \Vert \nabla w \Vert_{q}^{q-1} \Vert \nabla u_{\alpha}^{*} \Vert_{q}
\end{align} Note that, since
$u_{\alpha}^{*}$ is solution of (F α), then, for
$v=u_{\alpha}^{*}$,
\begin{equation*}
a(\alpha)\int_{\Omega}\left(\mu(x)\vert \nabla u_{\alpha}^{*}\vert^{p}
+\vert\nabla u_{\alpha}^{*}\vert^{q}\right)dx
=\int_{\Omega}f(u_{\alpha}^{*})u_{\alpha}^{*}dx.
\end{equation*}Then,
\begin{equation}
\Vert \nabla u_{\alpha}^{*}\Vert_{q}^{q}
\leq \Vert \nabla u_{\alpha}^{*}\Vert_{p,\mu}^{p}
+\Vert \nabla u_{\alpha}^{*}\Vert_{q}^{q}
\leq \frac{\vert\Omega\vert}{a(\alpha)}
\max\limits_{t\in[0,t_{*}]}[tf(t)]
\end{equation}and so
\begin{equation}
\Vert \nabla u_{\alpha}^{*}\Vert_{q}
\leq\left(\frac{\vert\Omega\vert}{a(\alpha)}
\max\limits_{t\in[0,t_{*}]}[tf(t)]\right)^{\frac{1}{q}}.
\end{equation} On the other hand, since
$w$ is a weak solution of (36), choosing
$v=w$ as test function and by Hölder’s inequality, we have
\begin{eqnarray*}
\Vert \nabla w\Vert_{q}^{q}&=&
\int_{\Omega}\vert\nabla w\vert^{q}dx\\
&=&\int_{\Omega}\left((u_{\alpha}^{*})^{r-1}
+\nu(x)(u_{\alpha}^{*})^{s-1}\right)wdx\cr
&\leq &\left(\int_{\Omega}\left((u_{\alpha}^{*})^{r-1}
+\nu(x)(u_{\alpha}^{*})^{s-1}\right)^{q'}dx
\right)^{\frac{1}{q'}}\left(\int_{\Omega}
\vert w\vert^{q}dx\right)^{\frac{1}{q}}\cr
&\leq & \left(t_{*}^{r-1}+t_{*}^{s-1}\max\limits_{\Omega}\nu \right)
\vert\Omega\vert^{\frac{1}{q'}}\Vert w\Vert_{q}\cr
&\leq &
\frac{\vert\Omega\vert^{1/q'}}{\lambda_{1,q}^{1/q}}
(t_{*}^{r-1}
+t_{*}^{s-1}\max\limits_{\Omega}\nu)\Vert \nabla w\Vert_{q},
\end{eqnarray*}that is,
\begin{equation}
\Vert \nabla w\Vert_{q}^{q-1}
\leq \frac{\vert\Omega\vert^{1/q'}}{\lambda_{1,q}^{1/q}}
\left(t_{*}^{r-1}+t_{*}^{s-1}\max\limits_{\Omega}\nu\right).
\end{equation}Finally, putting together (39) and (40) in (37), we have
\begin{eqnarray*}
&&\mathcal{P}_{k}(\alpha)\leq \Vert\nabla w\Vert_{q}^{q-1}
\Vert\nabla u_{\alpha}^{*}\Vert_{q}\cr
&\leq &\vert\Omega\vert^{\frac{1}{q'}}\left(t_{*}^{r-1}+t_{*}^{s-1}
\max\limits_{\Omega}\nu\right)\left(
\frac{\vert\Omega\vert}{a(\alpha)\lambda_{1,q}}
\max\limits_{t\in[0,t_{*}]}[tf(t)]\right)^{\frac{1}{q}}\cr
&=&\vert\Omega\vert\left(t_{*}^{r-1}+t_{*}^{s-1}
\max\limits_{\Omega}\nu\right)
\left(\frac{\max\limits_{t\in[0,t_{*}]}[tf(t)]}{a(\alpha)
\lambda_{1,q}}\right)^{\frac{1}{q}}.
\end{eqnarray*} By assumption
$(H_{5})$, we conclude the proof.
Now we are in position to conclude the proof of the main result.
Proof of Theorem 1.1
By (31), it is enough to prove that the map
$\mathcal{P}_{k}$ possesses at least two fixed points in the interval
$(\alpha_{k-1},\alpha_k)$ (for each
$k\in \{1,\ldots,K\}$). Propositions 5.2 and 5.3 combined with the continuity of
$\mathcal{P}_{k}$ (Proposition 5.1) permit us to apply the Intermediate Value Theorem and archive the proof.
Acknowledgements
The authors are grateful to the anonymous reviewers for their valuable suggestions, which helped us improve our manuscript.
The first author is member of the Gruppo Nazionale per l’Analisi Matematica, la Probabilità e le loro Applicazioni (GNAMPA) of the Istituto Nazionale di Alta Matematica (INdAM).
The paper is partially supported by INdAM—GNAMPA Project “Problemi differenziali non lineari: esistenza e molteplicità di soluzioni”—CUP E5324001950001.
This work was completed while the first-name author was visiting the University of the National Education Commission of Krakow, he is grateful for the kind hospitality of the host university.
Author contribution
All authors contributed equally to this work. All authors have read and agreed to the published version of the manuscript.
Competing interests
The authors declare no competing interests.
Ethical approval
Not applicable.
Availability of data and materials
All data are available in the paper.
