betaFeedback

WorksMathematics

Multiplicity of Positive Solutions for Singular Anisotropic Elliptic Equations with Variable Exponents

Abstract

We study anisotropic elliptic Dirichlet problems with variable exponents, singular nonlinearities, and subcritical superlinear reactions. Our contribution is a direct variational construction that combines Nehari geometry and fibering maps without requiring global differentiability of the singular energy, and the result establishes the existence of two positive weak solutions for sufficiently small λλ, distinguished by their energy signs.

Introduction

In heterogeneous materials, the resistance to diffusion may depend not only on the position at which a process takes place but also on the direction in which it evolves. This feature leads to nonlinear elliptic equations with nonstandard growth. Variable-exponent Lebesgue and Sobolev spaces provide a natural functional framework for these models, allowing the growth of the underlying energy to vary from point to point; see [9, 7].

An extension is obtained when the growth exponent depends simultaneously on the spatial position and the direction of differentiation. In this setting, the classical p(x)p(x)-Laplacian is replaced by an anisotropic variable-exponent operator of the form ∑i=1N∂i(∣∂iu∣pi(x)−2∂iu)\sum_{i=1}^{N} \partial_i\left(\lvert\partial_i u\rvert^{p_i(x)-2}\partial_i u\right), where p⃗(x)=(p1(x),…,pN(x))\vec{p}(x)=(p_1(x),\ldots,p_N(x)) represents the direction-dependent growth exponents. Fundamental contributions to this theory can be found in [18, 20, 15]. More recently, in [4] critical embeddings and a concentration–compactness principle for anisotropic variable-exponent Sobolev spaces are developed. Subsequent investigations in [5, 6] provided further compactness criteria, refined embedding results, and applications to anisotropic elliptic equations with critical growth.

The main difficulty in this work is that the equation contains a singular reaction term of the form u−δ(x)u^{-\delta(x)}, where 0<δ(x)<10<\delta(x)<1, which is unbounded as uu approaches zero. Although such terms have a sublinear primitive, they introduce difficulties associated with differentiability, positivity, and the behavior of solutions near their zero sets. When a singular contribution is coupled with a superlinear reaction, the corresponding variational problem exhibits competing effects at small and large amplitudes. The singular term governs the behavior near the origin, whereas the superlinear contribution may cause the energy functional to be unbounded below on the ambient Sobolev space. Consequently, standard unconstrained minimization is generally insufficient. See [3, 8, 2].

Recent investigations have considerably broadened the scope of existence and multiplicity results for anisotropic equations. We refer the reader to [14, 24]. Although these works are formulated for different operators and nonlinear structures, they highlight the growing interest in the interaction between anisotropy and singular effects.

Recent studies employ Nehari manifolds and fibering maps to establish multiplicity for singular multi-phase and mixed local and nonlocal variable-exponent problems ([1, 19]), while existence and uniqueness under pp-sublinear perturbations have also been investigated in [11]. The classical fibering approach ([10]) and related Nehari manifold techniques ([16, 17]) provide an effective framework for identifying distinct critical points through the decomposition of the energy geometry. In anisotropic variable-exponent settings, this decomposition requires refined estimates to accommodate the interplay of directional growth, singular nonlinearities, and superlinear reactions.

Motivated by these developments, in the present paper we establish the existence of multiple positive weak solutions under suitable assumptions on the exponents and weight functions. In particular, we prove that there exists a constant λ0>0\lambda_0 > 0 such that, for every λ∈(0,λ0)\lambda\in(0,\lambda_0), the problem admits at least two distinct positive weak solutions. The central idea is to study the fibering maps associated with the energy functional and to decompose the Nehari manifold into the subsets Nλ+\mathcal{N}_{\lambda}^{+}, Nλ−\mathcal{N}_{\lambda}^{-}, and Nλ0\mathcal{N}_{\lambda}^{0}. For sufficiently small values of λ\lambda, the degenerate component is excluded, allowing the two nondegenerate components to be treated separately.

Anisotropic Variable Exponent Spaces

Let us recall the properties of variable exponent spaces and anisotropic Sobolev spaces that will be used throughout the paper. Let Ω⊂RN\Omega\subset\mathbb{R}^{N}, N≥2N \ge2, be a bounded domain with C2C^{2} boundary, and put

P(Ω‾)={p∈C(Ω‾):p(x)>1 for every x∈Ω‾}.P(\overline{\Omega})=\{p\in C(\overline{\Omega}):p(x)>1\text{ for every }x\in\overline{\Omega}\}.

For any p∈P(Ω‾)p\in P(\overline{\Omega}), we write

p−:=min⁡x∈Ω‾p(x),p+:=max⁡x∈Ω‾p(x).p^{-}:=\min_{x\in\overline{\Omega}}p(x),\qquad p^{+}:=\max_{x\in\overline{\Omega}}p(x).

The variable exponent Lebesgue space is

Lp(x)(Ω)={u measurable:∫Ω∣u∣p(x) dx<∞},L^{p(x)}(\Omega)=\left\{u\text{ measurable}:\int_{\Omega}|u|^{p(x)}\,dx<\infty\right\},

equipped with the Luxemburg norm

∥u∥p(x)=inf⁡{τ>0:∫Ω∣uτ∣p(x) dx≤1}.\|u\|_{p(x)}=\inf\left\{\tau>0:\int_{\Omega}\left|\frac{u}{\tau}\right|^{p(x)}\,dx\le1\right\}.

This space is separable and reflexive. If ρp(u)=∫Ω∣u∣p(x) dx\rho_{p}(u)=\int_{\Omega}|u|^{p(x)}\,dx, then

∥u∥p(x)≤1⟹∥u∥p(x)p+≤ρp(u)≤∥u∥p(x)p−,\|u\|_{p(x)}\le1\quad\Longrightarrow\quad\|u\|_{p(x)}^{p^{+}}\le\rho_{p}(u)\le\|u\|_{p(x)}^{p^{-}},
∥u∥p(x)≥1⟹∥u∥p(x)p−≤ρp(u)≤∥u∥p(x)p+.\|u\|_{p(x)}\ge1\quad\Longrightarrow\quad\|u\|_{p(x)}^{p^{-}}\le\rho_{p}(u)\le\|u\|_{p(x)}^{p^{+}}.

Moreover, norm convergence to zero is equivalent to modular convergence to zero. If p′(x)=p(x)/(p(x)−1)p'(x)=p(x)/(p(x)-1), the variable exponent Hölder inequality gives

∣∫Ωuv dx∣≤2∥u∥p(x)∥v∥p′(x).\left|\int_{\Omega}uv\,dx\right|\le2\|u\|_{p(x)}\|v\|_{p'(x)}.

We refer to [13, 9] for these properties and the corresponding Sobolev-space theory.

The space W1,p(x)(Ω)W^{1,p(x)}(\Omega) consists of the functions u∈Lp(x)(Ω)u \in L^{p(x)}(\Omega) such that ∣∇u∣∈Lp(x)(Ω)|\nabla u| \in L^{p(x)}(\Omega). It may be equipped with the norm ∥u∥p(x)+∥∇u∥p(x)\|u\|_{p(x)}+\|\nabla u\|_{p(x)}.

Preliminaries

Let p⃗(x)=(p1(x),…,pN(x))\vec{p}(x)=(p_1(x),\ldots,p_N(x)), where pi∈P(Ω‾)p_i\in P(\overline{\Omega}), and define

p−:=min⁡1≤i≤Npi−,p+:=max⁡1≤i≤Npi+,p^{-}:=\min_{1\leq i\leq N}p_i^{-},\qquad p^{+}:=\max_{1\leq i\leq N}p_i^{+},
pm(x):=min⁡1≤i≤Npi(x),pM(x):=max⁡1≤i≤Npi(x).p_m(x):=\min_{1\leq i\leq N}p_i(x),\qquad p_M(x):=\max_{1\leq i\leq N}p_i(x).

The modular associated with our setting is

ρ(u):=∑i=1N∫Ω(∣∂iu∣pi(x)+∣u∣pi(x)) dx.\rho(u):=\sum_{i=1}^{N}\int_{\Omega}\left(|\partial_i u|^{p_i(x)}+|u|^{p_i(x)}\right)\,dx.

and its Luxemburg norm

∥u∥X:=inf⁡{τ>0:ρ(u/τ)≤1}.\|u\|_X:=\inf\{\tau>0:\rho(u/\tau)\leq1\}.

Throughout the paper,

X=W01,p⃗(x)(Ω):=Cc∞(Ω)‾∥⋅∥X.X=W_0^{1,\vec{p}(x)}(\Omega):=\overline{C_c^\infty(\Omega)}^{\|\cdot\|_X}.

Thus, the zero boundary condition is understood through this closure. The space XX is a reflexive Banach space, and X↪W01,p−(Ω)X\hookrightarrow W_0^{1,p^{-}}(\Omega) continuously. It is stable under absolute values, with

ρ(∣u∣)=ρ(u),∥∣u∣∥X=∥u∥X.\rho(|u|)=\rho(u),\qquad\||u|\|_X=\|u\|_X.

These facts follow by viewing XX as a closed subspace of the finite product of the variable exponent spaces containing uu and its coordinate derivatives. The Sobolev chain rule gives the assertion concerning absolute values; see also [12].

Proposition 2.1. For every u∈Xu\in X, the following inequalities hold:

∥u∥X≤1 ⟹ ∥u∥Xp+≤ρ(u)≤∥u∥Xp−,\|u\|_X\leq1\ \Longrightarrow\ \|u\|_X^{p^{+}}\leq\rho(u)\leq\|u\|_X^{p^{-}},
∥u∥X≥1 ⟹ ∥u∥Xp−≤ρ(u)≤∥u∥Xp+.(1)\|u\|_X\geq1\ \Longrightarrow\ \|u\|_X^{p^{-}}\leq\rho(u)\leq\|u\|_X^{p^{+}}. \tag*{(1)}

In particular, ρ(u)=1\rho(u)=1 if and only if ∥u∥X=1\|u\|_X=1.

Proof. For u≠0u\neq0, the definition of the Luxemburg norm gives ρ(u/∥u∥X)=1\rho(u/\|u\|_X)=1. Factoring ∥u∥Xpi(x)\|u\|_X^{p_i(x)} from each integrand and using p−≤pi(x)≤p+p^{-}\leq p_i(x)\leq p^{+} proves (1). The case u=0u=0 is immediate. □

Define

pm∗(x):={Npm(x)N−pm(x),pm(x)<N,+∞,pm(x)≥N.p_m^*(x):= \begin{cases} \dfrac{Np_m(x)}{N-p_m(x)}, & p_m(x)<N,\\ +\infty, & p_m(x)\geq N. \end{cases}

Only strictly subcritical embeddings are needed below.

Proposition 2.2. If r∈C(Ω‾)r\in C(\overline{\Omega}) and

1≤r(x)<pm∗(x)for every x∈Ω‾,1\leq r(x)<p_m^*(x)\qquad\text{for every }x\in\overline{\Omega},

then X↪Lr(x)(Ω)X\hookrightarrow L^{r(x)}(\Omega) continuously and compactly.

Estimates for the nonlinear terms

Let δ,q∈C(Ω‾)\delta,q \in C(\overline{\Omega}) satisfy

0<δ−≤δ(x)≤δ+<1,1<q(x)<pm∗(x),0 < \delta^{-} \le\delta(x) \le\delta^{+} < 1,\qquad1 < q(x) < p_{m}^{*}(x),

and let f,g∈L∞(Ω)f,g \in L^{\infty}(\Omega) be nonnegative. It is convenient to put

s(x):=1−δ(x),s−:=1−δ+,s+:=1−δ−.s(x):=1-\delta(x),\qquad s^{-}:=1-\delta^{+},\qquad s^{+}:=1-\delta^{-}.

Thus 0<s−≤s(x)≤s+<10 < s^{-} \le s(x) \le s^{+} < 1. Define

F(u):=∫Ωf(x)∣u∣s(x) dx,S(u):=∫Ωf(x)s(x)∣u∣s(x) dx,F(u):=\int_{\Omega}f(x)|u|^{s(x)}\,dx,\qquad S(u):=\int_{\Omega}\frac{f(x)}{s(x)}|u|^{s(x)}\,dx,
G(u):=∫Ωg(x)∣u∣q(x) dx,Q(u):=∫Ωg(x)q(x)∣u∣q(x) dx.G(u):=\int_{\Omega}g(x)|u|^{q(x)}\,dx,\qquad Q(u):=\int_{\Omega}\frac{g(x)}{q(x)}|u|^{q(x)}\,dx.

We work directly with these integrals.

Proposition 2.3. There exist constants Cf,Cg>0C_f,C_g>0 such that

F(u)≤Cf{∥u∥Xs−,∥u∥X≤1,∥u∥Xs+,∥u∥X≥1,(2)F(u)\le C_f \begin{cases} \lVert u\rVert_X^{s^{-}}, & \lVert u\rVert_X\le1,\\ \lVert u\rVert_X^{s^{+}}, & \lVert u\rVert_X\ge1, \end{cases} \tag*{(2)}

and

G(u)≤Cg{∥u∥Xq−,∥u∥X≤1,∥u∥Xq+,∥u∥X≥1.(3)G(u)\le C_g \begin{cases} \lVert u\rVert_X^{q^{-}}, & \lVert u\rVert_X\le1,\\ \lVert u\rVert_X^{q^{+}}, & \lVert u\rVert_X\ge1. \end{cases} \tag*{(3)}

The functionals F,S,G,QF,S,G,Q are sequentially weakly continuous on XX. Moreover,

F(u)s+≤S(u)≤F(u)s−,G(u)q+≤Q(u)≤G(u)q−.(4)\frac{F(u)}{s^{+}}\le S(u)\le\frac{F(u)}{s^{-}},\qquad \frac{G(u)}{q^{+}}\le Q(u)\le\frac{G(u)}{q^{-}}. \tag*{(4)}

Proof. The embedding into L1(Ω)L^{1}(\Omega) and the inequality ∣v∣s(x)≤1+∣v∣|v|^{s(x)}\le1+|v| give a uniform bound for F(v)F(v) on ∥v∥X=1\lVert v\rVert_X=1. The embedding into Lq(x)(Ω)L^{q(x)}(\Omega) gives the analogous bound for G(v)G(v). Writing u=∥u∥Xvu=\lVert u\rVert_Xv proves (2) and (3).

If un⇀uu_n\rightharpoonup u in XX, Proposition 2.2 gives strong convergence in L1(Ω)L^{1}(\Omega) and Lq(x)(Ω)L^{q(x)}(\Omega). For 0<s<10<s<1,

∣∣a∣s−∣b∣s∣≤∣a−b∣s.\bigl||a|^s-|b|^s\bigr|\le|a-b|^s.

Consequently,

∣F(un)−F(u)∣≤∥f∥∞∫Ω(∣un−u∣s−+∣un−u∣s+) dx⟶0,|F(u_n)-F(u)|\le\lVert f\rVert_{\infty}\int_{\Omega}\left(|u_n-u|^{s^{-}}+|u_n-u|^{s^{+}}\right)\,dx\longrightarrow0,

where the last assertion follows from strong L1L^1 convergence and Hölder’s inequality with constant exponents. The same argument applies to SS, since s−>0s^{-}>0. Strong Lq(x)L^{q(x)} convergence implies convergence of the q(x)q(x)-power integrals, and hence of GG and QQ. Finally, (4) follows from the pointwise bounds on s(x)s(x) and q(x)q(x). □\square

Statement of the Singular Problem and Variational Setting

In this section, we formulate the singular anisotropic elliptic problem that will be analyzed throughout the paper. For a given parameter λ>0\lambda>0, we consider

{−∑i=1N∂i(∣∂iu∣pi(x)−2∂iu)+∑i=1N∣u∣pi(x)−2u=λf(x)u−δ(x)+g(x)uq(x)−1in Ω,u>0a.e. in Ω,u=0on ∂Ω.(5)\left\{ \begin{aligned} -\sum_{i=1}^{N}\partial_i\left(|\partial_i u|^{p_i(x)-2}\partial_i u\right)+\sum_{i=1}^{N}|u|^{p_i(x)-2}u &=\lambda f(x)u^{-\delta(x)}+g(x)u^{q(x)-1} &&\text{in }\Omega,\\ u&>0 &&\text{a.e. in }\Omega,\\ u&=0 &&\text{on }\partial\Omega. \end{aligned} \right. \tag*{(5)}

Here Ω⊂RN\Omega\subset\mathbb{R}^{N}, N≥2N \ge2, is a bounded domain with C2C^{2} boundary. We impose the following assumptions:

(A1) pi,δ,q∈C(Ω‾)p_i,\delta,q \in C(\overline{\Omega}), 1<p−≤p+<∞1 < p^{-} \le p^{+} < \infty, 0<δ−≤δ+<10 < \delta^{-} \le\delta^{+} < 1;

(A2) 0<1−δ+≤1−δ−<p−≤p+<q−≤q+0 < 1-\delta^{+} \le1-\delta^{-} < p^{-} \le p^{+} < q^{-} \le q^{+}, q(x)<pm∗(x)q(x) < p_m^{*}(x) for every x∈Ω‾x \in\overline{\Omega};

(A3) f,g∈C(Ω‾)f,g \in C(\overline{\Omega}), f>0f > 0 a.e. in Ω\Omega, g≥0g \ge0, g≢0g \not\equiv0.

Definition 3.1. A function u∈Xu \in X is a weak solution of (5) if u>0u > 0 a.e. in Ω\Omega,

∫Ωf(x)u−δ(x)∣φ∣ dx<∞for every φ∈X,(6)\int_{\Omega} f(x)u^{-\delta(x)}|\varphi|\,dx < \infty\qquad\text{for every } \varphi\in X, \tag*{(6)}

and

∑i=1N∫Ω∣∂iu∣pi(x)−2∂iu ∂iφ dx+∑i=1N∫Ωupi(x)−1φ dx=λ∫Ωf(x)u−δ(x)φ dx+∫Ωg(x)uq(x)−1φ dxfor every φ∈X.(7)\begin{aligned} \sum_{i=1}^{N}\int_{\Omega}|\partial_i u|^{p_i(x)-2}\partial_i u\,\partial_i\varphi\,dx +\sum_{i=1}^{N}\int_{\Omega}u^{p_i(x)-1}\varphi\,dx &= \lambda\int_{\Omega}f(x)u^{-\delta(x)}\varphi\,dx \\ &\quad+\int_{\Omega}g(x)u^{q(x)-1}\varphi\,dx \qquad\text{for every } \varphi\in X. \tag*{(7)} \end{aligned}

The energy functional associated with (5) is

Iλ(u)=E(u)−λS(u)−Q(u),u∈X,I_{\lambda}(u)=E(u)-\lambda S(u)-Q(u), \qquad u\in X,

where

E(u):=∑i=1N∫Ω∣∂iu∣pi(x)+∣u∣pi(x)pi(x) dx.E(u):=\sum_{i=1}^{N}\int_{\Omega}\frac{|\partial_i u|^{p_i(x)}+|u|^{p_i(x)}}{p_i(x)}\,dx.

Note that the functionals EE and QQ belong to C1(X,R)C^{1}(X,\mathbb{R}), and EE is convex. In particular,

⟨E′(u),φ⟩=∑i=1N∫Ω∣∂iu∣pi(x)−2∂iu ∂iφ dx+∑i=1N∫Ω∣u∣pi(x)−2uφ dx,\langle E'(u),\varphi\rangle =\sum_{i=1}^{N}\int_{\Omega}|\partial_i u|^{p_i(x)-2}\partial_i u\,\partial_i\varphi\,dx +\sum_{i=1}^{N}\int_{\Omega}|u|^{p_i(x)-2}u\varphi\,dx,

and ⟨E′(u),u⟩=ρ(u)\langle E'(u),u\rangle=\rho(u). The functional IλI_{\lambda} is continuous and sequentially weakly lower semicontinuous, by the convexity of EE and Proposition 2.3.

The superlinear term prevents global coercivity. Indeed, choosing ψ∈C0∞(Ω)\psi\in C_{0}^{\infty}(\Omega) with G(ψ)>0G(\psi)>0, we have Iλ(tψ)→−∞I_{\lambda}(t\psi)\to-\infty as t→∞t\to\infty because q−>p+q^{-}>p^{+}. Note that SS is not differentiable on XX. Therefore, we do not assume that Iλ∈C1(X,R)I_{\lambda}\in C^{1}(X,\mathbb{R}) and do not apply a Lagrange multiplier argument to it.

Instead, we use the differentiability of the restrictions t↦Iλ(tu)t\mapsto I_{\lambda}(tu) for t>0t>0, construct minimizers on the two Nehari components, and identify their weak equations by one-sided variations. This direct treatment of the singular energy is consistent with the constrained variational approach used in [22, 21], while the arguments required for the present anisotropic variable-exponent setting are given below. The construction is carried out directly for the singular functional.

Here is the main result of this paper.

Theorem 3.2. Assume (A1)–(A3). There exists λ0>0\lambda_{0}>0 such that, for every λ∈(0,λ0)\lambda\in(0,\lambda_{0}), problem (5) has at least two distinct positive weak solutions uλ,vλ∈Xu_{\lambda},v_{\lambda}\in X. Moreover,

Iλ(uλ)<0<Iλ(vλ).I_{\lambda}(u_{\lambda})<0<I_{\lambda}(v_{\lambda}).

Positivity is understood almost everywhere, as in Definition 3.1.

The constant λ0\lambda_{0} will be chosen explicitly in terms of the embedding estimates in § 4. The proof of the theorem is completed in § 6.

Nehari Manifold and Fibering Maps

For u∈X∖{0}u \in X \setminus\{0\}, define the fibering map

Φu(t):=Iλ(tu),t≥0.\Phi_u(t) := \mathcal{I}_{\lambda}(tu), \qquad t \ge0.

Although Iλ\mathcal{I}_{\lambda} is not assumed to be differentiable on XX, the function Φu\Phi_u belongs to C2((0,∞))C^2((0,\infty)). To simplify notation, put

ai(u,x):=∣∂iu∣pi(x)+∣u∣pi(x),fu(x):=f(x)∣u∣s(x),gu(x):=g(x)∣u∣q(x).a_i(u,x) := |\partial_i u|^{p_i(x)}+|u|^{p_i(x)}, \qquad f_u(x) := f(x)|u|^{s(x)}, \qquad g_u(x) := g(x)|u|^{q(x)}.

Then

Φu′(t)=∑i=1N∫Ωtpi(x)−1ai(u,x) dx−λ∫Ωts(x)−1fu dx−∫Ωtq(x)−1gu dx.\Phi'_u(t)=\sum_{i=1}^{N}\int_{\Omega}t^{p_i(x)-1}a_i(u,x)\,dx-\lambda\int_{\Omega}t^{s(x)-1}f_u\,dx-\int_{\Omega}t^{q(x)-1}g_u\,dx.

and

Φu′′(t)=∑i=1N∫Ω(pi(x)−1)tpi(x)−2ai(u,x) dx+λ∫Ωδ(x)t−δ(x)−1fu dx−∫Ω(q(x)−1)tq(x)−2gu dx.\Phi''_u(t)=\sum_{i=1}^{N}\int_{\Omega}(p_i(x)-1)t^{p_i(x)-2}a_i(u,x)\,dx+\lambda\int_{\Omega}\delta(x)t^{-\delta(x)-1}f_u\,dx-\int_{\Omega}(q(x)-1)t^{q(x)-2}g_u\,dx.

The Nehari set is defined by the radial constraint

Nλ:={u∈X∖{0}:Jλ(u)=0},Jλ(u):=Φu′(1)=ρ(u)−λF(u)−G(u).\mathcal{N}_{\lambda}:=\{u\in X\setminus\{0\}:J_{\lambda}(u)=0\}, \qquad J_{\lambda}(u):=\Phi'_u(1)=\rho(u)-\lambda F(u)-G(u).

Thus tΦu′(t)=Jλ(tu)t\Phi'_u(t)=J_{\lambda}(tu). We retain the term Nehari manifold, but no global differentiable-manifold structure is required. Every nontrivial weak solution of (5) belongs to Nλ\mathcal{N}_{\lambda}, by testing its equation with uu.

For u∈Nλu\in\mathcal{N}_{\lambda}, set

Dλ(u):=∑i=1N∫Ωpi(x)ai(u,x) dx−λ∫Ωs(x)fu dx−∫Ωq(x)gu dx.D_{\lambda}(u):=\sum_{i=1}^{N}\int_{\Omega}p_i(x)a_i(u,x)\,dx-\lambda\int_{\Omega}s(x)f_u\,dx-\int_{\Omega}q(x)g_u\,dx.

Since Jλ(u)=0J_{\lambda}(u)=0, we have Dλ(u)=Φu′′(1)D_{\lambda}(u)=\Phi''_u(1). Accordingly,

Nλ±:={u∈Nλ:Dλ(u)≶0},Nλ0:={u∈Nλ:Dλ(u)=0}.\mathcal{N}_{\lambda}^{\pm}:=\{u\in\mathcal{N}_{\lambda}:D_{\lambda}(u)\lessgtr0\}, \qquad\mathcal{N}_{\lambda}^{0}:=\{u\in\mathcal{N}_{\lambda}:D_{\lambda}(u)=0\}.

Lemma 4.1. The functional Iλ\mathcal{I}_{\lambda} is coercive and bounded below on Nλ\mathcal{N}_{\lambda}.

Proof. For u∈Nλu\in\mathcal{N}_{\lambda}, we have G(u)=ρ(u)−λF(u)G(u)=\rho(u)-\lambda F(u). Therefore,

Iλ(u)≥ρ(u)p+−λF(u)s−−G(u)q−\mathcal{I}_{\lambda}(u)\ge\frac{\rho(u)}{p^{+}}-\frac{\lambda F(u)}{s^{-}}-\frac{G(u)}{q^{-}}
=(1p+−1q−)ρ(u)−λ(1s−−1q−)F(u).=\left(\frac{1}{p^{+}}-\frac{1}{q^{-}}\right)\rho(u)-\lambda\left(\frac{1}{s^{-}}-\frac{1}{q^{-}}\right)F(u).

If ∥u∥X≥1\|u\|_X\ge1, Propositions 2.1 and 2.3 give

Iλ(u)≥(1p+−1q−)∥u∥Xp−−λCf(1s−−1q−)∥u∥Xs+.\mathcal{I}_{\lambda}(u)\ge\left(\frac{1}{p^{+}}-\frac{1}{q^{-}}\right)\|u\|_X^{p^{-}}-\lambda C_f\left(\frac{1}{s^{-}}-\frac{1}{q^{-}}\right)\|u\|_X^{s^{+}}.

Since q−>p+q^{-}>p^{+} and p−>s+p^{-}>s^{+}, RHS tends to +∞+\infty as ∥u∥X→∞\|u\|_X\to\infty. For ∥u∥X≤1\|u\|_X\le1, the same estimates provide a uniform lower bound. □\square

The following lemma shows that, for sufficiently small λ>0\lambda>0, both the energy functional and the Nehari constraint functional remain strictly positive on a suitable sphere in XX. This property plays an important role in the analysis of fibering maps and the construction of distinct variational solutions.

Lemma 4.2. Choose

r0:=min⁡{12,(14Cg)1/(q−−p+)},λ0:=s−4p+Cfr0p+−s−,η0:=r0p+2p+.(8)r_0 := \min\left\{\frac{1}{2},\left(\frac{1}{4C_g}\right)^{1/(q^{-}-p^{+})}\right\},\qquad\lambda_0 := \frac{s^{-}}{4p^{+}C_f}r_0^{p^{+}-s^{-}},\qquad\eta_0 := \frac{r_0^{p^{+}}}{2p^{+}}. \tag*{(8)}

For every λ∈(0,λ0)\lambda\in(0,\lambda_0) and every u∈Xu \in X with ∥u∥X=r0\lVert u\rVert_X=r_0, one has

Jλ(u)>0,Iλ(u)≥η0>0.J_\lambda(u)>0,\qquad I_\lambda(u)\geq\eta_0>0.

Proof. Since r0<1r_0<1, the preceding estimates give

ρ(u)≥r0p+,F(u)≤Cfr0s−,G(u)≤Cgr0q−.\rho(u)\geq r_0^{p^{+}},\qquad F(u)\leq C_f r_0^{s^{-}},\qquad G(u)\leq C_g r_0^{q^{-}}.

By the choice of r0r_0 and λ0\lambda_0,

Cgr0q−≤14r0p+,λCfr0s−<s−4p+r0p+.C_g r_0^{q^{-}}\leq\frac{1}{4}r_0^{p^{+}},\qquad\lambda C_f r_0^{s^{-}}<\frac{s^{-}}{4p^{+}}r_0^{p^{+}}.

It follows that

Jλ(u)>(1−14−s−4p+)r0p+>0.J_\lambda(u)>\left(1-\frac{1}{4}-\frac{s^{-}}{4p^{+}}\right)r_0^{p^{+}}>0.

Similarly,

Iλ(u)≥(1p+−14p+−14q−)r0p+≥r0p+2p+.I_\lambda(u)\geq\left(\frac{1}{p^{+}}-\frac{1}{4p^{+}}-\frac{1}{4q^{-}}\right)r_0^{p^{+}}\geq\frac{r_0^{p^{+}}}{2p^{+}}.

Lemma 4.3 (Geometry of the fibering maps). Let λ∈(0,λ0)\lambda\in(0,\lambda_0) and u∈X∖{0}u \in X\setminus\{0\}. If G(u)>0G(u)>0, there exist exactly two numbers

0<t+(u)<r0∥u∥X<t−(u)0<t^{+}(u)<\frac{r_0}{\lVert u\rVert_X}<t^{-}(u)

such that t+(u)u∈Nλ+t^{+}(u)u\in\mathcal{N}_{\lambda}^{+} and t−(u)u∈Nλ−t^{-}(u)u\in\mathcal{N}_{\lambda}^{-}. If G(u)=0G(u)=0, there is exactly one such number t+(u)t^{+}(u), and t+(u)u∈Nλ+t^{+}(u)u\in\mathcal{N}_{\lambda}^{+}. All these radial intersections are nondegenerate.

Proof. Assumption (A3) implies F(u)>0F(u)>0 for every u≠0u\neq0. Set Ku(t):=t−s+Jλ(tu)K_u(t):=t^{-s^{+}}J_\lambda(tu), for all t>0t>0. Differentiation gives

Ku′(t)=tp+−s+−1(∑i=1N∫Ω(pi(x)−s+)ai(u,x)tpi(x)−p+ dx+λ∫Ω(s+−s(x))futs(x)−p+ dx−∫Ω(q(x)−s+)gutq(x)−p+ dx).(9)\begin{aligned} K_u'(t)={}&t^{p^{+}-s^{+}-1}\left(\sum_{i=1}^{N}\int_{\Omega}(p_i(x)-s^{+})a_i(u,x)t^{p_i(x)-p^{+}}\,dx\right.\\ &\left.+\lambda\int_{\Omega}(s^{+}-s(x))f_u t^{s(x)-p^{+}}\,dx-\int_{\Omega}(q(x)-s^{+})g_u t^{q(x)-p^{+}}\,dx\right). \tag*{(9)} \end{aligned}

Suppose first that G(u)>0G(u)>0. Since

s(x)≤s+<p−≤pi(x)≤p+<q−≤q(x),s(x)\leq s^{+}<p^{-}\leq p_i(x)\leq p^{+}<q^{-}\leq q(x),

the first two terms inside the parentheses in (9) are nonincreasing in tt, whereas the last term is strictly decreasing. Moreover, the expression inside the parentheses is positive for sufficiently small tt and tends to −∞-\infty as t→∞t\to\infty. Consequently, it vanishes at exactly one point, and hence KuK_u is strictly increasing up to that point and strictly decreasing thereafter.

For 0<t≤10<t\leq1, we have

Ku(t)≤tp−−s+ρ(u)−λF(u).K_u(t)\leq t^{p^{-}-s^{+}}\rho(u)-\lambda F(u).

Since p−>s+p^{-}>s^{+} and F(u)>0F(u)>0, it follows that Ku(t)<0K_u(t)<0 for all sufficiently small tt. Moreover, Ku(t)→−∞K_u(t)\to-\infty as t→∞t\to\infty, because q−>p+q^{-}>p^{+}.

Lemma 4.2 gives

Ku(r0∥u∥X)>0.K_u\left(\frac{r_0}{\lVert u\rVert_X}\right)>0.

Thus KuK_u, and hence Jλ(tu)J_\lambda(tu) (as a function of tt), has exactly two zeros, one on each side of r0/∥u∥Xr_0/\lVert u\rVert_X. At the first zero (Jλ(tu))′>0(J_\lambda(tu))' > 0, and at the second (Jλ(tu))′<0(J_\lambda(tu))' < 0. Since Φu′=Jλ(tu)/t\Phi'_u = J_\lambda(tu)/t, these zeros belong to the stated Nehari components and are nondegenerate. If G(u)=0G(u) = 0, the last term in (9) vanishes, so Ku′(t)>0K'_u(t) > 0 for all t>0t > 0. The function KuK_u is negative near zero and tends to +∞+\infty. It therefore has exactly one zero, again before r0/∥u∥Xr_0/\lVert u\rVert_X, and this zero is nondegenerate. □

Corollary 4.4. For every λ∈(0,λ0)\lambda\in(0,\lambda_0),

Nλ0=∅,Nλ+≠∅,Nλ−≠∅.\mathcal{N}^0_\lambda= \varnothing,\qquad\mathcal{N}^+_\lambda\ne\varnothing,\qquad\mathcal{N}^-_\lambda\ne\varnothing.

Furthermore,

∥u∥X<r0(u∈Nλ+),∥v∥X>r0(v∈Nλ−),(10)\lVert u\rVert_X < r_0 \quad(u \in\mathcal{N}^+_\lambda),\qquad\lVert v\rVert_X > r_0 \quad(v \in\mathcal{N}^-_\lambda), \tag*{(10)}

and

Iλ(u)<0(u∈Nλ+),Iλ(v)≥η0(v∈Nλ−).(11)I_\lambda(u) < 0 \quad(u \in\mathcal{N}^+_\lambda),\qquad I_\lambda(v) \ge\eta_0 \quad(v \in\mathcal{N}^-_\lambda). \tag*{(11)}

For every v∈Nλ−v \in\mathcal{N}^-_\lambda,

Iλ(v)=max⁡t≥0Iλ(tv).(12)I_\lambda(v) = \max_{t \ge0} I_\lambda(tv). \tag*{(12)}

Proof. Nondegeneracy and the norm separation follow from Lemma 4.3. Every nonzero direction has a positive-branch intersection. Since g≠0g \ne0, a smooth function can be chosen with G(u)>0G(u) > 0, and its ray also meets the negative branch. The fibering map decreases from Φu(0)=0\Phi_u(0) = 0 to its first critical point; hence the energy on Nλ+\mathcal{N}^+_\lambda is negative. When G(u)>0G(u) > 0, it then increases to its second critical point and subsequently decreases to −∞-\infty. The second critical value is at least the value at r0u/∥u∥Xr_0u/\lVert u\rVert_X, which is bounded below by η0\eta_0. Since this value is positive, it is also the global maximum on the ray, including t=0t = 0. □

Lemma 4.5. For fixed λ∈(0,λ0)\lambda\in(0,\lambda_0), the map u↦t−(u)u \mapsto t^-(u) is continuous on

D:={u∈X:G(u)>0}.\mathcal{D} := \{u \in X : G(u) > 0\}.

Proof. The function (t,u)↦Jλ(tu)(t,u) \mapsto J_\lambda(tu) is continuous for t>0t > 0, and its radial derivative is continuous there. At t=t−(u)t = t^-(u), this radial derivative is strictly negative. Choose two nearby values of tt, one on each side of the zero, at which Jλ(tu)J_\lambda(tu) has opposite signs. For all directions sufficiently close to uu, the same signs persist. The unique descending zero for each such direction lies between these two values. Letting them approach t−(u)t^-(u) proves continuity. This argument uses only radial differentiability, not differentiability with respect to uu. □

Existence of Nontrivial Minimizers

This section presents the constrained minimization arguments on the two nondegenerate components of the Nehari set.

Theorem 5.1. For every λ∈(0,λ0)\lambda\in(0,\lambda_0), there exists a nonnegative uλ∈Nλ+u_\lambda\in\mathcal{N}^+_\lambda such that

Iλ(uλ)=inf⁡u∈Nλ+Iλ(u)<0.I_\lambda(u_\lambda) = \inf_{u \in\mathcal{N}^+_\lambda} I_\lambda(u) < 0.

Moreover, uλu_\lambda is a local minimizer of IλI_\lambda on XX.

Proof. Consider the closed ball

B‾r0:={u∈X:∥u∥X≤r0}.\overline{B}_{r_0} := \{u \in X : \lVert u\rVert_X \le r_0\}.

It is weakly compact, and IλI_\lambda is sequentially weakly lower semicontinuous. The functional therefore attains its minimum on this ball at some uλu_\lambda. Choose a nonzero, nonnegative ψ∈Cc∞(Ω)\psi\in C^\infty_c(\Omega). Then F(ψ)>0F(\psi) > 0. For 0<t<10 < t < 1,

Iλ(tψ)≤ρ(ψ)p−tp−−λF(ψ)s+ts+<0I_\lambda(t\psi) \le\frac{\rho(\psi)}{p^-}t^{p^-} - \frac{\lambda F(\psi)}{s^+}t^{s^+} < 0

for all sufficiently small tt, since s+<p−s^{+}<p^{-}. Hence Iλ(uλ)<0I_\lambda(u_\lambda)<0 and uλ≠0u_\lambda\ne0. On the boundary of the ball the energy is positive by Lemma 4.2, so ∥uλ∥X<r0\lVert u_\lambda\rVert_X<r_0. Thus uλu_\lambda is a local minimizer on XX. The one-dimensional map t↦Iλ(tuλ)t\mapsto I_\lambda(tu_\lambda) has a local minimum at t=1t=1. In particular, Jλ(uλ)=0J_\lambda(u_\lambda)=0. By (10) and the absence of a degenerate component, uλ∈Nλ+u_\lambda\in\mathcal{N}_\lambda^{+}. Every element of Nλ+\mathcal{N}_\lambda^{+} belongs to B‾r0\overline{B}_{r_0}, so the minimizer also attains the infimum on Nλ+\mathcal{N}_\lambda^{+}. Finally, replacing uλu_\lambda by ∣uλ∣\lvert u_\lambda\rvert preserves the norm, the energy, and the Nehari component. We may therefore take uλ≥0u_\lambda\ge0.

Lemma 5.2. For every v∈Nλ−v\in\mathcal{N}_\lambda^{-},

G(v)>p−−s+q+−s+ρ(v)≥p−−s+q+−s+r0p+.G(v)>\frac{p^{-}-s^{+}}{q^{+}-s^{+}}\rho(v)\ge\frac{p^{-}-s^{+}}{q^{+}-s^{+}}r_0^{p^{+}}.

Proof. Using λF(v)=ρ(v)−G(v)\lambda F(v)=\rho(v)-G(v), we obtain

Dλ(v)≥p−ρ(v)−λs+F(v)−q+G(v)=(p−−s+)ρ(v)−(q+−s+)G(v).D_\lambda(v)\ge p^{-}\rho(v)-\lambda s^{+}F(v)-q^{+}G(v)=(p^{-}-s^{+})\rho(v)-(q^{+}-s^{+})G(v).

Since Dλ(v)<0D_\lambda(v)<0, the first inequality follows. The second follows from ∥v∥X>r0\lVert v\rVert_X>r_0, r0<1r_0<1, and (1), regardless of whether ∥v∥X\lVert v\rVert_X is smaller or larger than one.

Theorem 5.3. For every λ∈(0,λ0)\lambda\in(0,\lambda_0), there exists a nonnegative vλ∈Nλ−v_\lambda\in\mathcal{N}_\lambda^{-} such that

Iλ(vλ)=inf⁡v∈Nλ−Iλ(v)≥η0>0.I_\lambda(v_\lambda)=\inf_{v\in\mathcal{N}_\lambda^{-}}I_\lambda(v)\ge\eta_0>0.

Proof. Put

cλ−:=inf⁡v∈Nλ−Iλ(v).c_\lambda^{-}:=\inf_{v\in\mathcal{N}_\lambda^{-}}I_\lambda(v).

This number is finite because the set is nonempty, and it is at least η0\eta_0 by (11). Let {vn}⊂Nλ−\{v_n\}\subset\mathcal{N}_\lambda^{-} be a minimizing sequence. Replacing its elements by their absolute values, we assume vn≥0v_n\ge0. Lemma 4.1 shows that the sequence is bounded in XX. After passing to a subsequence,

vn⇀v in X,vn→v in Lq(x)(Ω) and L1(Ω),v_n\rightharpoonup v\text{ in }X,\qquad v_n\to v\text{ in }L^{q(x)}(\Omega)\text{ and }L^1(\Omega),

with v≥0v\ge0. Proposition 2.3 and Lemma 5.2 imply

G(v)=lim⁡n→∞G(vn)≥p−−s+q+−s+r0p+>0.G(v)=\lim_{n\to\infty}G(v_n)\ge\frac{p^{-}-s^{+}}{q^{+}-s^{+}}r_0^{p^{+}}>0.

In particular, v≠0v\ne0 and the descending radial parameter t∗=t−(v)t_*=t^{-}(v) is well defined. By weak lower semicontinuity and the global maximum property on each ray,

cλ−≤Iλ(t∗v)≤lim inf⁡n→∞Iλ(t∗vn)≤lim sup⁡n→∞Iλ(t∗vn)≤lim⁡n→∞Iλ(vn)=cλ−.c_\lambda^{-}\le I_\lambda(t_*v)\le\liminf_{n\to\infty}I_\lambda(t_*v_n)\le\limsup_{n\to\infty}I_\lambda(t_*v_n)\le\lim_{n\to\infty}I_\lambda(v_n)=c_\lambda^{-}.

Thus vλ:=t∗v∈Nλ−v_\lambda:=t_*v\in\mathcal{N}_\lambda^{-} attains the infimum.

Proof of the Main Result

In this section, we show that the minimizers constructed above are positive weak solutions. The essential point is to establish the integrability of the singular term before differentiating it in an arbitrary direction.

Lemma 6.1. Assume (A1)–(A3) and let λ>0\lambda>0, w∈X∖{0}w\in X\setminus\{0\} with w≥0w\ge0. Suppose that for every φ∈X\varphi\in X and all sufficiently small ε>0\varepsilon>0 there is tε>0t_\varepsilon>0, possibly depending on φ\varphi, such that tε→1t_\varepsilon\to1 and

Iλ(tε(w+εφ))≥Iλ(tεw).(13)I_\lambda\bigl(t_\varepsilon(w+\varepsilon\varphi)\bigr)\ge I_\lambda(t_\varepsilon w). \tag*{(13)}

Then ww is a positive weak solution of (5).

Proof. Take h∈Xh \in X, h≥0h \ge0. By (13),

λ∫Ωf(x)tεs(x)s(x)(w+εh)s(x)−ws(x)ε dx≤(E−Q)(tε(w+εh))−(E−Q)(tεw)ε(14)\lambda\int_{\Omega}\frac{f(x)t_{\varepsilon}^{s(x)}}{s(x)}\frac{(w+\varepsilon h)^{s(x)}-w^{s(x)}}{\varepsilon}\,dx\le\frac{(E-Q)(t_{\varepsilon}(w+\varepsilon h))-(E-Q)(t_{\varepsilon}w)}{\varepsilon} \tag*{(14)}

Since tε→1t_{\varepsilon}\to1 and E−Q∈C1(X)E-Q\in C^{1}(X), the mean value formula gives

(E−Q)(tε(w+εh))−(E−Q)(tεw)ε=∫01⟨(E−Q)′(tεw+θεtεh),tεh⟩ dθ⟶⟨(E−Q)′(w),h⟩.\frac{(E-Q)(t_{\varepsilon}(w+\varepsilon h))-(E-Q)(t_{\varepsilon}w)}{\varepsilon}=\int_{0}^{1}\langle(E-Q)'(t_{\varepsilon}w+\theta\varepsilon t_{\varepsilon}h),t_{\varepsilon}h\rangle\,d\theta\longrightarrow\langle(E-Q)'(w),h\rangle.

The integrand on the left is nonnegative. At any point where w=0w=0, f>0f>0, and h>0h>0, it tends to +∞+\infty, because s(x)<1s(x)<1. Fatou’s lemma therefore implies that {w=0,f>0,h>0}\{w=0,f>0,h>0\} has measure zero. Taking nonnegative smooth test functions which are positive on the members of a countable covering of Ω\Omega, and using f>0f>0 a.e., we obtain w>0w>0 a.e. in Ω\Omega.

Another application of Fatou’s lemma to (14) gives

λ∫Ωf(x)w−δ(x)h dx≤⟨(E−Q)′(w),h⟩<∞(h∈X, h≥0).(15)\lambda\int_{\Omega}f(x)w^{-\delta(x)}h\,dx\le\langle(E-Q)'(w),h\rangle<\infty\qquad(h\in X,\ h\ge0). \tag*{(15)}

In particular, choosing h=∣φ∣h=|\varphi| proves (6) for every φ∈X\varphi\in X. It also shows that the singular density defines a bounded linear functional on XX.

We may now return to arbitrary signed directions. For a>0a>0, b∈Rb\in\mathbb{R}, and 0<s<10<s<1, the elementary inequality

∣∣a+b∣s−as∣≤as−1∣b∣(16)\left||a+b|^{s}-a^{s}\right|\le a^{s-1}|b| \tag*{(16)}

holds. Indeed, after dividing by asa^{s}, it follows by considering whether ∣1+b/a∣|1+b/a| is smaller or larger than one and using the concavity of t↦tst\mapsto t^{s}. Applying (16) with a=w(x)a=w(x) and b=εφ(x)b=\varepsilon\varphi(x) gives

f(x)s(x)∣∣w+εφ∣s(x)−ws(x)∣ε≤1sf(x)w−δ(x)∣φ∣.\frac{f(x)}{s(x)}\frac{\left||w+\varepsilon\varphi|^{s(x)}-w^{s(x)}\right|}{\varepsilon}\le\frac{1}{s}f(x)w^{-\delta(x)}|\varphi|.

The right-hand side is integrable by (15). The factors tεs(x)t_{\varepsilon}^{s(x)} are uniformly bounded and converge to one. Dominated convergence consequently yields

lim⁡ε↓0S(tε(w+εφ))−S(tεw)ε=∫Ωf(x)w−δ(x)φ dx.\lim_{\varepsilon\downarrow0}\frac{S(t_{\varepsilon}(w+\varepsilon\varphi))-S(t_{\varepsilon}w)}{\varepsilon}=\int_{\Omega}f(x)w^{-\delta(x)}\varphi\,dx.

Dividing (13) by ε\varepsilon after subtracting the right-hand side, and passing to the limit, one obtains

⟨E′(w),φ⟩−∫Ωg(x)wq(x)−1φ dx−λ∫Ωf(x)w−δ(x)φ dx≥0.\langle E'(w),\varphi\rangle-\int_{\Omega}g(x)w^{q(x)-1}\varphi\,dx-\lambda\int_{\Omega}f(x)w^{-\delta(x)}\varphi\,dx\ge0.

Applying the same argument to −φ-\varphi gives the reverse inequality. Thus equality holds for every φ∈X\varphi\in X, which is precisely (7). □

Proof of Theorem 3.2. Let $0<\lambda<\lambda_{0}$, with λ0\lambda_{0} given by (8). By Theorem 5.1, there is a nonnegative local minimizer uλ∈Nλ+u_{\lambda}\in\mathcal{N}_{\lambda}^{+}. For any φ∈X\varphi\in X and all sufficiently small $\varepsilon>0,

Iλ(uλ+εφ)≥Iλ(uλ).I_{\lambda}(u_{\lambda}+\varepsilon\varphi)\ge I_{\lambda}(u_{\lambda}).

Thus Lemma 6.1 applies with tε=1t_{\varepsilon}=1, and uλu_{\lambda} is a positive weak solution.

By Theorem 5.3, there is also a nonnegative minimizer vλ∈Nλ−v_{\lambda}\in\mathcal{N}_{\lambda}^{-}. Fix φ∈X\varphi\in X. Since G(vλ)>0G(v_{\lambda})>0, we have G(vλ+εφ)>0G(v_{\lambda}+\varepsilon\varphi)>0 for sufficiently small ε>0\varepsilon>0. Set

tε:=t−(vλ+εφ).t_{\varepsilon}:=t^{-}(v_{\lambda}+\varepsilon\varphi).

Lemma 4.5 gives tε→t−(vλ)=1t_{\varepsilon}\to t^{-}(v_{\lambda})=1. Since (vλ+εφ)tε∈Nλ−(v_{\lambda}+\varepsilon\varphi)t_{\varepsilon}\in\mathcal{N}_{\lambda}^{-}, minimality on this component and (12) give

Iλ(tε(vλ+εφ))≥Iλ(vλ)≥Iλ(tεvλ).I_{\lambda}\bigl(t_{\varepsilon}(v_{\lambda}+\varepsilon\varphi)\bigr)\ge I_{\lambda}(v_{\lambda})\ge I_{\lambda}(t_{\varepsilon}v_{\lambda}).

Hence Lemma 6.1 applies to vλv_\lambda as well. Finally,

Iλ(uλ)<0<η0≤Iλ(vλ),\mathcal{I}_\lambda(u_\lambda)<0<\eta_0\leq\mathcal{I}_\lambda(v_\lambda),

so the two weak solutions are distinct. Their positivity and the integrability of the singular term have been established in Lemma 6.1. This completes the proof. □

Corollary 6.2. The first solution satisfies

∥uλ∥X≤(4Cfλ3)1/(p+−s−).(17)\lVert u_\lambda\rVert_X\leq\left(\frac{4C_f\lambda}{3}\right)^{1/(p^+-s^-)}. \tag*{(17)}

In particular, uλ→0u_\lambda\to0 strongly in XX as λ↓0\lambda\downarrow0.

Proof. Since ∥uλ∥X<r0\lVert u_\lambda\rVert_X<r_0,

G(uλ)≤Cg∥uλ∥Xq−≤14∥uλ∥Xp+≤14ρ(uλ).G(u_\lambda)\leq C_g\lVert u_\lambda\rVert_X^{q^-}\leq\frac{1}{4}\lVert u_\lambda\rVert_X^{p^+}\leq\frac{1}{4}\rho(u_\lambda).

The Nehari identity therefore implies

34∥uλ∥Xp+≤34ρ(uλ)≤λF(uλ)≤λCf∥uλ∥Xs−.\frac{3}{4}\lVert u_\lambda\rVert_X^{p^+}\leq\frac{3}{4}\rho(u_\lambda)\leq\lambda F(u_\lambda)\leq\lambda C_f\lVert u_\lambda\rVert_X^{s^-}.

Dividing by the positive factor ∥uλ∥Xs−\lVert u_\lambda\rVert_X^{s^-} proves (17). □

The preceding result relies on the positivity assumption imposed on the singular weight ff. We now consider the more general case where the weight functions are nonnegative and may vanish on subsets of Ω\Omega. The following proposition shows that the multiplicity result remains valid in this setting, although positivity almost everywhere of the solutions is guaranteed only where ff is positive.

Proposition 6.3. Assume (A1),(A2) and replace (A3) by

f,g∈C(Ω‾),f,g≥0,f≢0,g≢0.f,g\in C(\overline{\Omega}),\qquad f,g\geq0,\qquad f\not\equiv0,\qquad g\not\equiv0.

For every sufficiently small λ>0\lambda>0, the same construction gives two distinct nontrivial nonnegative solutions of the equation and boundary condition in (5), with negative and positive energies. Both solutions are positive almost everywhere on {f>0}\{f>0\}. Here the singular term is interpreted as

Tu(x):={f(x)u(x)−δ(x),f(x)>0,0,f(x)=0,T_u(x):= \begin{cases} f(x)u(x)^{-\delta(x)}, & f(x)>0,\\ 0, & f(x)=0, \end{cases}

and the weak formulation uses TuT_u in place of fu−δ(x)fu^{-\delta(x)}. No assertion of positivity throughout Ω\Omega is made under these weaker weight assumptions.

Proof. The embedding estimates and Lemma 4.2 are unchanged. For directions with F(u)>0F(u)>0, Lemma 4.3 applies without modification. If F(u)=0<G(u)F(u)=0<G(u), the function

t−p+Jλ(tu)=∑i=1N∫Ωai(u,x)tpi(x)−p+ dx−∫Ωgutq(x)−p+ dxt^{-p^+}J_\lambda(tu)=\sum_{i=1}^{N}\int_\Omega a_i(u,x)t^{p_i(x)-p^+}\,dx-\int_\Omega g_u t^{q(x)-p^+}\,dx

is strictly decreasing from a positive value or +∞+\infty to −∞-\infty. It has exactly one zero, on the descending branch and beyond r0/∥u∥Xr_0/\lVert u\rVert_X. The corresponding fibering map has a positive global maximum there. If F(u)=G(u)=0F(u)=G(u)=0, there is no Nehari intersection on that ray.

The first minimization argument uses a smooth test function with F(ψ)>0F(\psi)>0, which exists because f≢0f\not\equiv0. The second minimization argument and the continuity of the descending radial parameter remain valid on {G>0}\{G>0\}. Finally, the Fatou argument in Lemma 6.1 yields positivity on {f>0}\{f>0\} and integrability of Tu∣φ∣T_u|\varphi|. On {f=0}\{f=0\} the singular energy and all its directional increments vanish. Thus the dominated convergence argument still proves the weak equation. □

Remark 6.4. Proposition 6.3 applies, in particular, when both weights have compact support in Ω\Omega. It separates the existence argument from the additional problem of propagating positivity beyond the support of the singular weight. Such propagation cannot be inferred from an isotropic p(x)p(x)-Laplacian Harnack inequality. Strong maximum principles in anisotropic settings depend on the operator and on its directional structure; see [23]. Likewise, Theorem 3.2 asserts positivity almost everywhere, not continuity, a pointwise positive representative, or a positive lower bound on every compact subset. Those stronger properties require additional regularity arguments.

Conclusion and Perspectives

We have established the existence of two distinct positive weak solutions with opposite energy signs for a class of singular anisotropic elliptic problems with variable exponents, for sufficiently small λ>0\lambda>0. A key feature of our approach is the explicit and uniform construction of the constants r0r_0, λ0\lambda_0, and η0\eta_0 in Lemma 4.2, which simultaneously ensures Jλ(u)>0J_\lambda(u)>0 and Iλ(u)≥η0>0I_\lambda(u)\geq\eta_0>0 on the sphere ∥u∥X=r0\|u\|_X=r_0. This provides a quantitative energy barrier and the separation needed for the Nehari geometry despite the absence of a common homogeneity and the presence of a singular term. Combined with constrained minimization and one-sided variational arguments, this yields multiplicity without requiring global C1C^1 differentiability of the energy or regularization of the singularity. The extension to nonnegative weights further distinguishes existence from positivity throughout the domain.

Several questions naturally arise from these results.

  1. Optimal parameter range and continuation of solutions. The number λ0\lambda_0 obtained here is a sufficient smallness threshold, not an optimal existence threshold. A natural problem is to study

Λ∗:=sup⁡{λ>0: problem (5) has a positive weak solution},\Lambda^*:=\sup\{\lambda>0:\text{ problem (5) has a positive weak solution}\},

including whether Λ∗\Lambda^* is finite, whether a solution exists at Λ∗\Lambda^*, and how the two variational families behave as the parameter increases. This project would require a priori bounds and an analysis of the solution set; the present proof does not establish continuous or ordered branches.

  1. Small-parameter asymptotics and limiting profiles. Estimate (17) already shows that the negative-energy solution tends to zero in XX. It would be useful to determine whether this rate is sharp and whether an appropriate normalization produces a nontrivial limiting profile. For the positive-energy solutions, a related problem is to establish convergence, along suitable sequences as λ↓0\lambda\downarrow0, to a nontrivial solution of

−∑i=1N∂i(∣∂iv∣pi(x)−2∂iv)+∑i=1N∣v∣pi(x)−2v=g(x)vq(x)−1.-\sum_{i=1}^{N}\partial_i\left(|\partial_i v|^{p_i(x)-2}\partial_i v\right)+\sum_{i=1}^{N}|v|^{p_i(x)-2}v=g(x)v^{q(x)-1}.

The absence of a common homogeneous scaling makes the identification of such profiles a separate question.

  1. Regularity and boundary behavior. The present results are formulated in the weak sense. Under additional regularity and growth assumptions on the exponents, one may study local boundedness, Hölder continuity, and quantitative lower bounds on compact subsets. A further project is to determine how the boundary decay depends on the singular exponent, the directional growth, and the vanishing rate of ff near ∂Ω\partial\Omega. These questions would clarify when positivity almost everywhere can be strengthened to pointwise positivity of a continuous representative.

  1. Sharper embeddings and critical growth. The proof uses the conservative subcritical scale associated with pm(x)p_m(x). It is natural to investigate the larger growth ranges available from genuinely anisotropic embeddings, whose formulation involves the harmonic mean of the directional exponents. The critical case would additionally require a concentration–compactness

analysis and an energy threshold preventing loss of compactness. Extending the argument to coupled systems would raise related questions concerning the interaction of several singular components.

These directions extend the existence theory developed here while keeping separate the questions of multiplicity, positivity, regularity, and parameter dependence.

Acknowledgment: Part of this work was carried out while the second author was visiting KTH Royal Institute of Technology in Stockholm, Sweden. SK gratefully acknowledges the hospitality received during the visit.

AI declaration: The initial draft of this manuscript was prepared without artificial intelligence assistance. During the revision, the authors used ChatGPT (GPT-6) for proofreading, language editing and suggesting directions for future research. The tool suggested items (2) and (4) in § 7, and also some relevant references like [4, 5, 6] which were previously unknown to the authors.

Data availability statement: No datasets were generated or analyzed during this study.

Funding and/or Conflicts of interests/Competing interests: The authors declare that there are no financial, competing or conflict of interests.

Email address: [email protected]

Email address: [email protected]

Email address: [email protected]

{sup}1`Department of Basic Sciences, Babol Noshirvani University of Technology, Babol, Iran

References

  1. [1]M. Avci, Nehari manifold approach for a singular multi-phase variable exponent problem, Quaest. Math. 49 (2026), 569–589.arxiv.org/abs/2501.17344
  2. [2]Y. Bai, N. S. Papageorgiou, S. Zeng, Anisotropic eigenvalue problems with singular and sign-changing terms, Commun. Nonlinear Sci. Numer. Simul. 138 (2024), 108170.DOI
  3. [3]B. Brandolini, F. C. Cîrstea, Singular anisotropic elliptic equations with gradient-dependent lower order terms, Nonlinear Differential Equations and Applications NoDEA 30(5) (2023), 58.arxiv.org/abs/2001.02887
  4. [4]N. Chems Eddine, M. A. Ragusa, D. D. Repovš, On the concentration-compactness principle for anisotropic variable exponent Sobolev spaces and its applications, Fract. Calc. Appl. Anal. 27 (2024), 725–756.DOI
  5. [5]N. Chems Eddine, M. A. Ragusa, The compact embeddings and the concentration-compactness principles for anisotropic variable exponent Sobolev spaces and applications, J. Geom. Anal. 35 (2025), Article 338.DOI
  6. [6]N. Chems Eddine, M. A. Ragusa, Anisotropic variable exponent Sobolev spaces and applications to elliptic equations with critical growth, Fract. Calc. Appl. Anal. (2026), published online 15 October 2026.DOI
  7. [7]Y. Chen, S. Levine, M. Rao, Variable exponent, linear growth functionals in image restoration, SIAM J. Appl. Math. 66 (4) (2006), 1383-1406.DOI
  8. [8]G. Di Blasio, F. G. Zecca, Existence and uniqueness of solutions to some anisotropic elliptic equations with a singular convection term, Journal d’Analyse Mathématique 155(2) (2025), 501-528.arxiv.org/abs/2307.13564
  9. [9]L. Diening, P. Harjulehto, P. Hästö, M. Růžička, Lebesgue and Sobolev Spaces with Variable Exponents, Lecture Notes in Mathematics, Vol. 2017, Springer, Heidelberg, 2011.
  10. [10]P. Drábek, S. I. Pohozaev, Positive solutions for the p-Laplacian: application of the fibering method, Proceedings of the Royal Society of Edinburgh Section A: Mathematics 127(4) (1997), 703-726.DOI
  11. [11]F. Esposito, F. Oliva, E. Vecchi, Existence and uniqueness of weak solutions to singular anisotropic elliptic problems, arXiv preprint arXiv:2604.08733 (2026).arxiv.org/abs/2604.08733
  12. [12]X. Fan, Anisotropic variable exponent Sobolev spaces and \vec p(x)-Laplacian equations, Complex Var. Elliptic Equ. 56(7–9) (2011), 623–642.
  13. [13]X. Fan, D. Zhao, On the spaces L^{p(x)}(Ω) and W^{1,p(x)}(Ω), J. Math. Anal. Appl. 263(2) (2001), 424-446.
  14. [14]R. Guefaifia, S. Boulaaras, R. Jan, Some new existence results of positive solutions for anisotropic variable exponent operator, Boundary Value Problems 2025 (2025), Article 13.DOI
  15. [15]S. Khademloo, G. A. Afr’ouzi, T. Norouzi Ghara, Infinitely many solutions for anisotropic variable exponent problems, Complex Var. Elliptic Equ. 63 (2018), 1353-1369.DOI
  16. [16]S. Khademloo, G. A. Afr’ouzi, The Nehari manifold for a class of semilinear elliptic equations with indefinite weight, Bull. Iranian Math. Soc. 33(2) (2007), 49-59.
  17. [17]S. Khademloo, S. Khanjani, Existence and multiplicity of nontrivial solutions for p-Laplacian systems with concave-convex nonlinearities and sign-changing weight functions, Bull. Iranian Math. Soc. 40(5) (2014), 1301-1326.
  18. [18]O. Kovacik, J. Rakosnik, On spaces L^{p(x)}(Ω) and W^{m,p(x)}(Ω), Czechoslovak Math. J. 41 (1991), 592-618.
  19. [19]S. Malhotra, A. K. Pandey, K. Sreenadh, Multiplicity results for mixed local-nonlocal variable exponent problem, Diff. Integ. Equ. 39 (9-10) (2026), 817-866.arxiv.org/abs/2506.19793
  20. [20]M. Mihăilescu, P. Pucci, V. Rădulescu, Eigenvalue problems for anisotropic quasilinear elliptic equations with variable exponent, J. Math. Anal. Appl. 340 (2008), 687-698.DOI
  21. [21]N. S. Papageorgiou, D. D. Repovš, C. Vetro, Positive solutions for singular double phase problems, J. Math. Anal. Appl. 501(1) (2021), Article 123896.DOI
  22. [22]D. D. Repovš, K. Saoudi, The Nehari manifold approach for singular equations involving the p(x)-Laplace operator, Complex Variables and Elliptic Equations 68(1) (2023), 135-149.
  23. [23]J. Vétois, Strong maximum principles for anisotropic elliptic and parabolic equations, Adv. Nonlinear Stud. 12(1) (2012), 101–114.DOI
  24. [24]J. Zuo, M. Bahadi, Y. Akdim, Y. Wu, On anisotropic singular elliptic equations with a Finsler p(·)-Laplacian and variable exponents, Netw. Heterog. Media 21(6) (2026), 2196–2222.DOI

Paper details

vlnc:W_yeu4r2un
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

ChatGPT (GPT-6)

View statement
Looking for related papers…
Contents