betaFeedback

WorksMathematics

Multiplicity of positive solutions for degenerate nonlocal pp-Laplacian problems without monotonicity of the normalized reaction

Abstract

We study degenerate nonlocal Dirichlet problems driven by the pp-Laplacian, with diffusion depending on weighted sums of integral powers of the solution. Under explicit sign, spectral and size conditions, we obtain two positive solutions for each bounded positivity interval of the diffusion coefficient with zero endpoint values. Neither monotonicity of the normalized reaction f(x,t)/tp−1f(x,t)/t^{p-1} nor uniqueness of the frozen problem is required. The proof combines Schauder's theorem on a product space of a scalar parameter and a function with subsolution bounds and an energy estimate. The multiplicity result also persists under uniformly small continuous perturbations of the nonlocal functional, including perturbations that do not preserve order.

Introduction

We consider the nonlocal Dirichlet problem

{−m(H(u))Δpu=f(x,u)in Ω,u>0in Ω,u=0on ∂Ω.\begin{cases} -m(\mathcal{H}(u))\Delta_p u = f(x,u) & \text{in } \Omega,\\ u > 0 & \text{in } \Omega,\\ u = 0 & \text{on } \partial\Omega. \end{cases}

where Ω⊂RN\Omega\subset\mathbb{R}^{N} is a bounded domain with smooth boundary, 1<p<+∞1 < p < +\infty, and Δpu=div⁡(∣∇u∣p−2∇u)\Delta_p u = \operatorname{div}(|\nabla u|^{p-2}\nabla u). The nonlocal functional is

H(u)=∑i=1M∫Ωμi(x)∣u∣qi dx,u∈L∞(Ω),(1)\mathcal{H}(u)=\sum_{i=1}^{M}\int_{\Omega}\mu_i(x)|u|^{q_i}\,dx,\qquad u\in L^{\infty}(\Omega), \tag*{(1)}

with M≥1M \ge1, qi≥1q_i \ge1 and nonnegative bounded measurable weights μi\mu_i. We seek two positive solutions for each bounded interval on which mm is positive and vanishes at both endpoints.

Equations of this type describe stationary states of models in which a spatially uniform diffusion rate depends on an integral of the state. For example, (P) is the stationary equation associated with

∂tu−div⁡(m(H(u))∣∇u∣p−2∇u)=f(x,u).\partial_tu-\operatorname{div}\bigl(m(\mathcal{H}(u))|\nabla u|^{p-2}\nabla u\bigr)=f(x,u).

This interpretation justifies the terms diffusion coefficient for mm and reaction for ff used below. The nonlocal dependence is in the scalar coefficient, while the differential operator is local. Classical motivations include Carrier’s string models [6, 7] and population dispersal models [8, 9]. More recent work studies how spatial heterogeneity and degeneracy of the coefficient affect positive stationary states [11, 14], as well as nonlocal evolution equations [5]. In the present setting, weights such as μi=χΩi\mu_i=\chi_{\Omega_i} allow the coefficient to depend on measurements over prescribed regions. These regions may overlap and need not cover Ω\Omega. Our result concerns the multiplicity of stationary states and does not address their dynamical stability.

Degeneracy of the nonlocal coefficient may change the structure of the positive solution set. For a logistic reaction, Delgado et al. [11] show that the continuum of positive solutions of the local problem is replaced by two disjoint branches in the nonlocal problem. For degenerate coefficients, multiplicity results with two positive solutions per positivity interval were obtained for the Laplacian in [15, 16] and for the pp-Laplacian in [4]. Related extensions include variable exponent, Neumann and double phase problems [3, 10, 13]. In [4], the nonlocal term is ∫Ω∣u∣q dx\int_{\Omega}|u|^q\,dx and the autonomous reaction satisfies a strict decrease condition on f(t)/tp−1f(t)/t^{p-1}. This condition yields uniqueness for the nonlinear problem obtained by fixing the nonlocal argument. Its solution uαu_\alpha then defines a continuous scalar map α↦∫Ωuαq dx\alpha\mapsto\int_{\Omega}u_\alpha^q\,dx, whose fixed points give the required solutions.

Gasiński, Santos Júnior and Siciliano [17] treat the semilinear equation −Δu=F(u,∫Ω∣u∣q dx)-\Delta u = F\left(u,\int_{\Omega}|u|^q\,dx\right), allowing singular dependence of FF on its second argument. Assumption (f2) of that paper requires s↦F(s,α)/ss \mapsto F(s,\alpha)/s to be decreasing. Its Proposition 2.1 gives a unique positive solution for each admissible α\alpha, and its Propositions 3.1 and 3.2 use the resulting continuous scalar map to obtain 2K2K solutions. Here the reaction may depend measurably on xx, the operator is the pp-Laplacian, and no decrease condition is imposed on the normalized reaction. We retain the same multiplicity pattern through a different fixed point construction. The two results have different scopes, since we do not treat the general singular dependence of FF on its second argument allowed in [17].

Our main contribution is to obtain this multiplicity conclusion without assuming monotonicity of

(0,t∗)∋t⟼f(x,t)tp−1.(2)(0,t_*) \ni t \longmapsto\frac{f(x,t)}{t^{p-1}}. \tag*{(2)}

We refer to this quotient as the normalized reaction. Fixing the nonlocal argument at a value α\alpha produces the auxiliary Dirichlet problem

−m(α)Δpu=f(x,u)in Ω,u=0on ∂Ω,(3)-m(\alpha)\Delta_p u=f(x,u) \quad\text{in } \Omega,\qquad u=0 \quad\text{on } \partial\Omega, \tag*{(3)}

which we call the frozen problem. If the normalized reaction is not decreasing, uniqueness of positive solutions of (1.3) does not follow from the principles of Brézis and Oswald [2] or Díaz and Saá [12], and no continuous selection α↦uα\alpha\mapsto u_\alpha of solutions is available from those arguments. We instead use an operator T(α,v)T(\alpha,v) defined by solving the Dirichlet problem obtained from (1.3) when the right-hand side is evaluated at a given function vv and suitably truncated. Schauder’s theorem is applied to a map defined on the product of a parameter interval and a ball in C01(Ω‾)C_0^1(\overline{\Omega}). A fixed point (α,u)(\alpha,u) of this map satisfies u=T(α,u)u=T(\alpha,u), so that uu solves the frozen problem at α\alpha. Bounds at the two ends of the parameter interval then enforce the additional identity α=H(u)\alpha=\mathcal{H}(u), which turns uu into a solution of (P). Lemma 2.2 records this elementary consequence of Schauder’s theorem.

Subsolution and supersolution methods are already used for nonlocal elliptic equations, for example in [1], where they are combined with pseudomonotone operator theory. In our proof, the lower barrier is a subsolution of the frozen problem which varies continuously with α\alpha. Near the endpoints of each positivity interval it is chosen sufficiently large to make H(u)>α\mathcal{H}(u)>\alpha for every fixed point uu of the truncated operator at such a value of α\alpha, whereas an energy estimate gives H(u)<α\mathcal{H}(u)<\alpha at a suitable interior point α=τk\alpha=\tau_k. Both estimates apply to all fixed points at the specified endpoints, and not only to a selected one. This is what makes the argument insensitive to the possible nonuniqueness for the frozen problem. Applying the construction on each side of the interior point gives two distinct solutions. The same argument acaccommodates the weighted sum in (1.1), with different exponents and weights that need not be bounded away from zero.

Suitably constructed subsolutions allow us to formulate the size condition in terms of the total weighted mass M∗=∑i=1Mt∗qi∫ΩμiM_* = \sum_{i=1}^{M} t_*^{q_i} \int_{\Omega} \mu_i. A continuous family connects these barriers to a small multiple of the first eigenfunction, for every p>1p > 1. Theorem 1.4 further shows that the multiplicity conclusion persists under continuous perturbations of H\mathcal{H} that are uniformly small on 0≤u≤t∗0 \le u \le t_*. The perturbed functional need not preserve order. Order preservation of the reference functional supplies the strict endpoint estimates used in the proof. Example 5.3 gives a perturbation depending on an additional spatial integral and proves that the resulting coefficient cannot be expressed as a function of the reference value H(u)\mathcal{H}(u) alone.

Definition 1.1. A function u∈W01,p(Ω)∩L∞(Ω)u \in W_0^{1,p}(\Omega) \cap L^\infty(\Omega) is a (weak) positive solution of (P) if u>0u > 0 a.e. in Ω\Omega and

m(H(u))∫Ω∣∇u∣p−2∇u⋅∇φ dx=∫Ωf(x,u)φ dxfor all φ∈W01,p(Ω).m(\mathcal{H}(u)) \int_{\Omega} |\nabla u|^{p-2}\nabla u \cdot\nabla\varphi\, dx = \int_{\Omega} f(x,u)\varphi\, dx \qquad\text{for all } \varphi\in W_0^{1,p}(\Omega).

We impose the following assumptions on the weights.

(Hμ)(H_\mu) For every i∈{1,…,M}i \in\{1,\ldots,M\}, μi∈L∞(Ω)\mu_i \in L^\infty(\Omega), μi≥0\mu_i \ge0 a.e. in Ω\Omega, and there exists at least one i∈{1,…,M}i \in\{1,\ldots,M\} such that μi≢0\mu_i \not\equiv0.

The coefficient satisfies the following condition.

(Hm)(H_m) m:[0,+∞)→Rm : [0,+\infty) \to\mathbb{R} is continuous and there exist numbers

0≤α1<α2≤α3<α4≤⋯≤α2K−1<α2K,K≥1,0 \le\alpha_1 < \alpha_2 \le\alpha_3 < \alpha_4 \le\cdots\le\alpha_{2K-1} < \alpha_{2K}, \qquad K \ge1,

such that

m>0in (α2k−1,α2k),m(α2k−1)=m(α2k)=0m > 0 \quad\text{in } (\alpha_{2k-1},\alpha_{2k}), \qquad m(\alpha_{2k-1}) = m(\alpha_{2k}) = 0

for all k∈{1,…,K}k \in\{1,\ldots,K\}.

The positivity intervals in (Hm)(H_m) are pairwise disjoint. No sign condition is imposed on mm outside them. Every solution furnished by Theorem 1.2 satisfies m(H(u))>0m(\mathcal{H}(u)) > 0.

We assume the following properties of the reaction.

(Hf)(H_f) The function f:Ω×R→Rf : \Omega\times\mathbb{R} \to\mathbb{R} is Carathéodory and satisfies the following conditions for some t∗>0t_* > 0.

  1. f(x,t∗)=0f(x,t_*) = 0 for a.e. x∈Ωx \in\Omega.

  1. For every compact interval J⊂(0,t∗)J \subset(0,t_*),

inf⁡t∈Jess inf⁡x∈Ωf(x,t)>0.\inf_{t \in J} \operatorname*{ess\,inf}_{x \in\Omega} f(x,t) > 0.

(iii) There exists T>0T>0 such that

ess sup⁡(x,t)∈Ω×[0,t∗] f(x,t)≤T.\underset{(x,t)\in\Omega\times[0,t_*]}{\operatorname{ess\,sup}}\ f(x,t)\le T.

Let λ1\lambda_1 be the first Dirichlet eigenvalue of −Δp-\Delta_p, and let e1e_1 be its positive eigenfunction normalized by ∥e1∥∞=1\lVert e_1\rVert_\infty=1. We require three compatibility conditions.

(Hλ)(H_\lambda)

lim inf⁡t→0+ess inf⁡x∈Ωf(x,t)tp−1>λ1max⁡k∈{1,…,K} max⁡α∈[α2k−1,α2k]m(α).\liminf_{t\to0^+}\underset{x\in\Omega}{\operatorname{ess\,inf}}\frac{f(x,t)}{t^{p-1}}>\lambda_1\max_{k\in\{1,\ldots,K\}}\ \max_{\alpha\in[\alpha_{2k-1},\alpha_{2k}]}m(\alpha).

(HM)(H_M)

α2K<M∗:=∑i=1Mt∗qi∫Ωμi(x) dx=H(t∗).\alpha_{2K}<M_*:=\sum_{i=1}^{M}t_*^{q_i}\int_{\Omega}\mu_i(x)\,dx=\mathcal{H}(t_*).

(Hτ)(H_\tau) For each k∈{1,…,K}k\in\{1,\ldots,K\} there exists τk∈(α2k−1,α2k)\tau_k\in(\alpha_{2k-1},\alpha_{2k}) such that

τkpm(τk)>∣Ω∣λ1(∑i=1Mt∗qi−1∥μi∥p′)pE,\tau_k^p m(\tau_k)>\frac{|\Omega|}{\lambda_1}\left(\sum_{i=1}^{M}t_*^{q_i-1}\lVert\mu_i\rVert_{p'}\right)^p E,

where

E:=ess sup⁡(x,t)∈Ω×[0,t∗] [tf(x,t)].E:=\underset{(x,t)\in\Omega\times[0,t_*]}{\operatorname{ess\,sup}}\ [tf(x,t)].

Here p′=p/(p−1)p'=p/(p-1), ∥⋅∥p′\lVert\cdot\rVert_{p'} denotes the norm in Lp′(Ω)L^{p'}(\Omega), and ∣Ω∣|\Omega| is the Lebesgue measure of Ω\Omega. Assumptions (Hf)(H_f) and (Hμ)(H_\mu) imply 0<E<+∞0<E<+\infty and ∥μi∥p′<+∞\lVert\mu_i\rVert_{p'}<+\infty.

We now state the first main result.

Theorem 1.2. Assume (Hμ)(H_\mu), (Hm)(H_m), (Hf)(H_f), (Hλ)(H_\lambda), (HM)(H_M) and (Hτ)(H_\tau). Then problem (P) has at least 2K2K pairwise distinct positive solutions in C01(Ω‾)C^1_0(\overline{\Omega}). More precisely, for each k∈{1,…,K}k\in\{1,\ldots,K\} there are solutions uk,1u_{k,1} and uk,2u_{k,2} such that 0<uk,j≤t∗0<u_{k,j}\le t_* in Ω\Omega for j=1,2j=1,2, and

α2k−1<H(uk,1)<τk<H(uk,2)<α2k.(4)\alpha_{2k-1}<\mathcal{H}(u_{k,1})<\tau_k<\mathcal{H}(u_{k,2})<\alpha_{2k}. \tag*{(4)}

The ordering in (1.4) concerns the values of H\mathcal{H}, and does not assert a pointwise ordering of the solutions. The role of the assumptions is as follows. Assumption (Hλ)(H_\lambda) provides a small positive subsolution θ0e1\theta_0e_1 of the frozen problem (1.3) for every α\alpha in the positivity intervals of mm. Assumptions (Hλ)(H_\lambda), (Hf)(H_f)(ii) and (HM)(H_M) provide larger subsolutions for α\alpha near the zeros of mm. These give the inequality H(u)>α\mathcal{H}(u)>\alpha for the fixed points at the outer endpoints of the parameter intervals used in the proof. Assumption (Hτ)(H_\tau) yields the reverse inequality H(u)<τk\mathcal{H}(u)<\tau_k for every solution uu of the frozen problem at α=τk\alpha=\tau_k with 0≤u≤t∗0\le u\le t_*. No condition on ff outside [0,t∗][0,t_*] is used, and the exponents qiq_i are not restricted by a Sobolev critical exponent. Example 5.1 verifies these assumptions for a normalized reaction that is neither nondecreasing nor nonincreasing.

Remark 1.3. For measurable sets Ωi⊂Ω\Omega_i \subset\Omega of positive measure, the choice μi=χΩi\mu_i=\chi_{\Omega_i} gives

H(u)=∑i=1M∫Ωi∣u∣qi dx=∑i=1M∥u∥qi,Ωiqi.\mathcal{H}(u)=\sum_{i=1}^{M}\int_{\Omega_i}|u|^{q_i}\,dx=\sum_{i=1}^{M}\|u\|_{q_i,\Omega_i}^{q_i}.

No regularity, disjointness or covering condition is required of these sets.

The multiplicity conclusion persists for a class of perturbations that need not preserve order. Let X=C01(Ω‾)X=C^1_0(\overline{\Omega}) and set

D={u∈X:0≤u≤t∗ in Ω}.\mathcal{D}=\{u\in X:0\leq u\leq t_* \text{ in }\Omega\}.

Theorem 1.4. Assume the hypotheses of Theorem 1.2. There exists ε0>0\varepsilon_0>0 such that, if H~:X→[0,∞)\widetilde{\mathcal{H}}:X\to[0,\infty) is continuous and satisfies

sup⁡u∈D∣H~(u)−H(u)∣≤ε0,(5)\sup_{u\in\mathcal{D}}|\widetilde{\mathcal{H}}(u)-\mathcal{H}(u)|\leq\varepsilon_0, \tag*{(5)}

then the problem

{−m(H~(u))Δpu=f(x,u)in Ω,u>0in Ω,u=0on ∂Ω(6)\begin{cases} -m(\widetilde{\mathcal{H}}(u))\Delta_pu=f(x,u)&\text{in }\Omega,\\ u>0&\text{in }\Omega,\\ u=0&\text{on }\partial\Omega \tag*{(6)} \end{cases}

has at least 2K2K pairwise distinct positive solutions uk,j∈Xu_{k,j}\in X, k∈{1,…,K}k\in\{1,\ldots,K\}, j∈{1,2}j\in\{1,2\}, with 0<uk,j≤t∗0<u_{k,j}\leq t_* in Ω\Omega, and

α2k−1<H~(uk,1)<τk<H~(uk,2)<α2k(k=1,…,K).(7)\alpha_{2k-1}<\widetilde{\mathcal{H}}(u_{k,1})<\tau_k<\widetilde{\mathcal{H}}(u_{k,2})<\alpha_{2k}\qquad(k=1,\ldots,K). \tag*{(7)}

Remark 1.5. No order-preserving property is required of H~\widetilde{\mathcal{H}}. The smallness assumption controls only the values of the functional on D\mathcal{D}, not its derivatives or any Lipschitz constant. Although D\mathcal{D} is not bounded in the C1C^1 norm, the reference values satisfy 0≤H(u)≤M∗0\leq\mathcal{H}(u)\leq M_* there. No perturbation bound outside D\mathcal{D} is needed. Order preservation of the reference functional is used to establish the strict endpoint margins, which then persist for the actual functional. In particular, this theorem does not replace the monotonicity of the differential operator −Δp-\Delta_p, and it does not claim the same conclusion for arbitrary continuous nonlocal functionals. The conclusion concerns persistence of multiplicity and does not assert continuous dependence of individual solutions on the perturbation. The proof shows that ε0\varepsilon_0 may be any number in (0,d)(0,d), where d>0d>0 is the explicit margin defined in (4.3).

Section 2 contains the fixed point lemma and the subsolution construction. Section 3 develops the auxiliary problems and endpoint estimates. Section 4 contains the proofs of Theorems 1.2 and 1.4. Section 5 gives three examples illustrating the main results.

Preliminaries and auxiliary tools

For a measurable set D⊆ΩD \subseteq\Omega of positive measure, we use the standard Lebesgue spaces Lr(D)L^r(D) and norms ∥⋅∥r,D\|\cdot\|_{r,D}. When D=ΩD=\Omega, the subscript DD is omitted. The space W01,p(Ω)W_0^{1,p}(\Omega) is endowed with the norm ∥u∥=∥∇u∥p\|u\|=\|\nabla u\|_p, and its dual is denoted by W−1,p′(Ω)W^{-1,p'}(\Omega). We write s+=max⁡{s,0}s^+=\max\{s,0\} and s−=max⁡{−s,0}s^-=\max\{-s,0\}, and denote the characteristic function of a measurable set DD by χD\chi_D. We also use

C01(Ω‾)={u∈C1(Ω‾):u=0 on ∂Ω},∥u∥C1(Ω‾)=∥u∥∞+∥∇u∥∞.C_0^1(\overline{\Omega})=\{u\in C^1(\overline{\Omega}):u=0\text{ on }\partial\Omega\},\qquad\|u\|_{C^1(\overline{\Omega})}=\|u\|_\infty+\|\nabla u\|_\infty.

For ν∈(0,1)\nu\in(0,1), the embedding C1,ν(Ω‾)↪C1(Ω‾)C^{1,\nu}(\overline{\Omega})\hookrightarrow C^1(\overline{\Omega}) is compact.

The operator −Δp:W01,p(Ω)→W−1,p′(Ω)-\Delta_p:W_0^{1,p}(\Omega)\to W^{-1,p'}(\Omega) is defined by

⟨−Δpu,v⟩=∫Ω∣∇u∣p−2∇u⋅∇v dx.\langle-\Delta_pu,v\rangle=\int_\Omega|\nabla u|^{p-2}\nabla u\cdot\nabla v\,dx.

It is continuous, bounded on bounded sets, coercive and strictly monotone [19]. This operator monotonicity is distinct from the decrease condition on the normalized reaction discussed in the Introduction. The first eigenvalue satisfies

λ1=inf⁡u∈W01,p(Ω)∖{0}∫Ω∣∇u∣p dx∫Ω∣u∣p dx>0.(8)\lambda_1=\inf_{u\in W_0^{1,p}(\Omega)\setminus\{0\}}\frac{\int_\Omega|\nabla u|^p\,dx}{\int_\Omega|u|^p\,dx}>0. \tag*{(8)}

The corresponding normalized positive eigenfunction satisfies

−Δpe1=λ1e1p−1,e1∈C01(Ω‾),0<e1≤1 in Ω.-\Delta_p e_1=\lambda_1e_1^{p-1},\qquad e_1\in C_0^1(\overline{\Omega}),\qquad0<e_1\leq1\text{ in }\Omega.

These properties follow from the spectral theory of the Dirichlet pp-Laplacian, together with the regularity theory and the strong maximum principle, see [19 Chapters 8 and 9].

We shall repeatedly use the following consequence of the Carathéodory property of ff. Suppose that a nonstrict inequality between two functions of (x,t)(x,t) which are continuous in tt holds, for each fixed tt in an interval JJ, for a.e. x∈Ωx\in\Omega. Then it holds for every t∈Jt\in J outside a single null set of xx. It suffices to take the union of the exceptional null sets corresponding to a countable dense subset of JJ and to use continuity in tt. If JJ has nonempty interior, a nonstrict inequality valid almost everywhere on Ω×J\Omega\times J has the same property, by Fubini’s theorem and continuity in tt. In this way, (Hf)(iii)(H_f)(iii) and the definition of EE in (Hτ)(H_\tau) give

0≤f(x,t)≤Tfor a.e. x∈Ω and every t∈[0,t∗],(9)0\leq f(x,t)\leq T\quad\text{for a.e. }x\in\Omega\text{ and every }t\in[0,t_*], \tag*{(9)}

and

tf(x,t)≤Efor a.e. x∈Ω and every t∈[0,t∗],(10)tf(x,t)\leq E\quad\text{for a.e. }x\in\Omega\text{ and every }t\in[0,t_*], \tag*{(10)}

where the lower bound in (9) follows from (Hf)(ii)(H_f)(ii), applied on a countable exhaustion of (0,t∗)(0,t_*) by compact intervals, together with (Hf)(i)(H_f)(i) and the continuity of f(x,⋅)f(x,\cdot) at t=0t=0. The same argument gives a uniform positive lower bound for ff on Ω×J\Omega\times J for each compact interval J⊂(0,t∗)J \subset(0,t_*). All these estimates remain valid when tt is replaced by a measurable function with values in the relevant interval.

We use the following form of Schauder’s theorem [19 Theorem 4.7].

Theorem 2.1 (Schauder fixed point theorem). Let XX be a Banach space, let C⊂XC \subset X be nonempty, closed, bounded, and convex, and let K:C→C\mathcal{K}: C \to C be continuous and compact, in the sense that K(C)\mathcal{K}(C) is relatively compact in XX. Then K\mathcal{K} has a fixed point.

Lemma 2.2 (Fixed point principle for a scalar constraint). Let XX be a Banach space, let B⊂XB \subset X be nonempty, closed, bounded and convex, and let Φ:X→R\Phi: X \to\mathbb{R} be continuous. Let a<ba < b and let T:[a,b]×B→B\mathcal{T}: [a,b] \times B \to B be continuous with relatively compact image. Suppose that one of the following two alternatives holds, the implications being required for every u∈Bu \in B.

(a)

u=T(a,u) ⟹ Φ(u)>a,u=T(b,u) ⟹ Φ(u)<b.u = \mathcal{T}(a,u) \ \Longrightarrow\ \Phi(u) > a,\qquad u = \mathcal{T}(b,u) \ \Longrightarrow\ \Phi(u) < b.

(b)

u=T(a,u) ⟹ Φ(u)<a,u=T(b,u) ⟹ Φ(u)>b.u = \mathcal{T}(a,u) \ \Longrightarrow\ \Phi(u) < a,\qquad u = \mathcal{T}(b,u) \ \Longrightarrow\ \Phi(u) > b.

Then there exists (α,u)∈(a,b)×B(\alpha,u) \in(a,b) \times B such that

u=T(α,u),α=Φ(u).u = \mathcal{T}(\alpha,u),\qquad\alpha= \Phi(u).

Proof. Let Π(s)=min⁡{b,max⁡{a,s}}\Pi(s) = \min\{b,\max\{a,s\}\}, and set σ=1\sigma= 1 in case (a) and σ=−1\sigma= -1 in case (b). Define

K(α,v)=(Π(α+σ[Φ(T(α,v))−α]),T(α,v)).\mathcal{K}(\alpha,v) = \left(\Pi\left(\alpha+ \sigma[\Phi(\mathcal{T}(\alpha,v))-\alpha]\right),\mathcal{T}(\alpha,v)\right).

This is a continuous map from [a,b]×B[a,b] \times B, a nonempty, closed, bounded and convex subset of the Banach space R×X\mathbb{R} \times X, into itself. Its image is contained in the compact set [a,b]×T([a,b]×B)[a,b]\times\mathcal{T}([a,b]\times B), so K\mathcal{K} is a compact map. Theorem 2.1 gives a fixed point (α,u)(\alpha,u), so that u=T(α,u)u=\mathcal{T}(\alpha,u) and α=Π(α+σ[Φ(u)−α])\alpha=\Pi(\alpha+\sigma[\Phi(u)-\alpha]). At α=a\alpha=a the endpoint assumption gives σ[Φ(u)−a]>0\sigma[\Phi(u)-a]>0, so the first component of K(a,u)\mathcal{K}(a,u) is greater than aa. At α=b\alpha=b it gives σ[Φ(u)−b]<0\sigma[\Phi(u)-b]<0, so the first component is less than bb. Both possibilities contradict the fixed point identity. Hence α∈(a,b)\alpha\in(a,b). Since Π(s)=α\Pi(s)=\alpha with α∈(a,b)\alpha\in(a,b) forces s=αs=\alpha, the projection identity reduces to

α=α+σ[Φ(u)−α].\alpha= \alpha+ \sigma[\Phi(u)-\alpha].

Since σ≠0\sigma\ne0, we obtain α=Φ(u)\alpha= \Phi(u).

The next two lemmas provide subsolutions for the frozen problem.

Lemma 2.3. Assume that (Hm)(H_m), (Hf)(H_f) and (Hλ)(H_\lambda) hold, and set

m0:=max⁡k∈{1,…,K}max⁡α∈[α2k−1,α2k]m(α).m_0 := \max_{k\in\{1,\ldots,K\}}\max_{\alpha\in[\alpha_{2k-1},\alpha_{2k}]}m(\alpha).

Then m0>0m_0 > 0 and there exist ε1>0\varepsilon_1 > 0 and η0∈(0,t∗)\eta_0 \in(0,t_*) such that

f(x,s)≥(λ1m0+ε1)sp−1>λ1m0sp−1for a.e. x∈Ω and every s∈(0,η0].(11)f(x,s) \ge(\lambda_1m_0+\varepsilon_1)s^{p-1} > \lambda_1m_0s^{p-1}\quad\text{for a.e. }x\in\Omega\text{ and every }s\in(0,\eta_0]. \tag*{(11)}

Consequently, for every θ0∈(0,η0)\theta_0 \in(0,\eta_0) and every α∈⋃k=1K[α2k−1,α2k]\alpha\in\bigcup_{k=1}^{K}[\alpha_{2k-1},\alpha_{2k}], the function θ0e1\theta_0e_1 is a subsolution of

−m(α)Δpu=f(x,u),u∈W01,p(Ω),(12)-m(\alpha)\Delta_pu=f(x,u),\qquad u\in W_0^{1,p}(\Omega), \tag*{(12)}

that is,

m(α)λ1(θ0e1)p−1≤f(x,θ0e1)for a.e. x∈Ω.m(\alpha)\lambda_1(\theta_0e_1)^{p-1}\le f(x,\theta_0e_1)\quad\text{for a.e. }x\in\Omega.

Proof. Assumption (Hm)(H_m) gives m0>0m_0 > 0. By (Hλ)(H_\lambda), there are ε1>0\varepsilon_1 > 0 and η0∈(0,t∗)\eta_0 \in(0,t_*) such that

ess inf⁡x∈Ωf(x,s)sp−1>λ1m0+ε1for every s∈(0,η0].\operatorname*{ess\,inf}_{x\in\Omega}\frac{f(x,s)}{s^{p-1}}>\lambda_1m_0+\varepsilon_1\quad\text{for every }s\in(0,\eta_0].

The null set argument recalled at the beginning of this section gives (11). For 0<θ0<η00<\theta_0<\eta_0, the inequalities 0<e1≤10<e_1\le1 and m(α)≤m0m(\alpha)\le m_0 then yield

−m(α)Δp(θ0e1)=m(α)λ1(θ0e1)p−1≤f(x,θ0e1)a.e. in Ω.-m(\alpha)\Delta_p(\theta_0e_1)=m(\alpha)\lambda_1(\theta_0e_1)^{p-1}\le f(x,\theta_0e_1)\quad\text{a.e. in }\Omega.

Lemma 2.4. Assume that (Hm)(H_m), (Hf)(H_f) and (Hλ)(H_\lambda) hold. Let θ∗∈(0,t∗)\theta_* \in(0,t_*). There exists c0>0c_0>0 such that

f(x,s)≥λ1c0sp−1for a.e. x∈Ω and every s∈(0,θ∗].(13)f(x,s)\ge\lambda_1c_0s^{p-1}\quad\text{for a.e. }x\in\Omega\text{ and every }s\in(0,\theta_*]. \tag*{(13)}

Consequently, if m(α)<c0m(\alpha)<c_0 and θ∈(0,θ∗]\theta\in(0,\theta_*], then θe1\theta e_1 is a subsolution of (12).

Proof. Use m0m_0 and η0\eta_0 from Lemma 2.3. If θ∗≤η0\theta_*\le\eta_0, take c0=m0c_0=m_0. Otherwise, (Hf)(ii)(H_f)(ii) applied to J=[η0,θ∗]J=[\eta_0,\theta_*] and the null set argument give ρ∗>0\rho_*>0 such that

f(x,s)≥ρ∗≥ρ∗θ∗p−1sp−1for a.e. x∈Ω and every s∈[η0,θ∗].f(x,s)\ge\rho_*\ge\frac{\rho_*}{\theta_*^{p-1}}s^{p-1}\quad\text{for a.e. }x\in\Omega\text{ and every }s\in[\eta_0,\theta_*].

Together with (11), this proves (13) with

c0=min⁡{m0,ρ∗λ1θ∗p−1}>0.c_0=\min\left\{m_0,\frac{\rho_*}{\lambda_1\theta_*^{p-1}}\right\}>0.

If m(α)<c0m(\alpha)<c_0 and 0<θ≤θ∗0<\theta\le\theta_*, substituting s=θe1(x)s=\theta e_1(x) in (13) gives the subsolution inequality.

Lemma 2.5 (A continuous path of subsolutions). Assume (Hμ)(H_\mu), (Hm)(H_m), (Hf)(H_f), (Hλ)(H_\lambda) and (HM)(H_M). There exist 0<θ0<θ∗<t∗0 < \theta_0 < \theta_* < t_*, w∈C01(Ω‾)w \in C_0^1(\overline{\Omega}), and a continuous map [0,1]∋r↦Lr∈C01(Ω‾)[0,1] \ni r \mapsto L_r \in C_0^1(\overline{\Omega}) such that

L0=θ0e1,L1=w,0<Lr≤θ∗ in Ω,H(w)>α2K.(14)L_0 = \theta_0 e_1,\qquad L_1 = w,\qquad0 < L_r \leq\theta_* \text{ in } \Omega,\qquad\mathcal{H}(w) > \alpha_2 K. \tag*{(14)}

The function L0L_0 is a subsolution of (2.5) throughout the positivity intervals of mm. There exists c∗>0c_* > 0 such that, for every r∈[0,1]r \in[0,1], LrL_r is a subsolution of (2.5) whenever 0<m(α)<c∗0 < m(\alpha) < c_*.

Proof. By (HM)(H_M) and the continuity of [0,t∗]∋θ↦H(θ)[0,t_*] \ni\theta\mapsto\mathcal{H}(\theta), choose θ∗∈(0,t∗)\theta_* \in(0,t_*) with H(θ∗)>α2K\mathcal{H}(\theta_*) > \alpha_2 K. Fix an integer n>max⁡{1,1/(p−1)}n > \max\{1,1/(p-1)\} and put

ζ(s)={1,0≤s≤1,(2−s)n,1<s<2,0,s≥2,A=∫02ζ(s) ds=n+2n+1.\zeta(s) = \begin{cases} 1, & 0 \leq s \leq1,\\ (2-s)^n, & 1 < s < 2,\\ 0, & s \geq2, \end{cases} \qquad A = \int_0^2 \zeta(s)\,\mathrm{d}s = \frac{n+2}{n+1}.

For 0<δ<1/20 < \delta< 1/2, let

hδ(s)=1A∫0s/δζ(t) dt(s≥0).h_\delta(s) = \frac{1}{A}\int_0^{s/\delta} \zeta(t)\,\mathrm{d}t \qquad(s \geq0).

This is a concave, nondecreasing C1C^1 function, linear with slope (Aδ)−1(A\delta)^{-1} on [0,δ][0,\delta], and equal to 11 on [2δ,∞)[2\delta,\infty). In particular hδ(0)=0h_\delta(0) = 0, hδ(1)=1h_\delta(1) = 1, and s≤hδ(s)≤1s \leq h_\delta(s) \leq1 for s∈[0,1]s \in[0,1]. As e1>0e_1 > 0 in Ω\Omega, hδ(e1(x))→1h_\delta(e_1(x)) \to1 for every x∈Ωx \in\Omega as δ→0+\delta\to0^+. Dominated convergence, applied to the weighted integrals, yields

H(θ∗hδ(e1))⟶H(θ∗)>α2K.\mathcal{H}(\theta_* h_\delta(e_1)) \longrightarrow\mathcal{H}(\theta_*) > \alpha_2 K.

Fix δ\delta so small that w=θ∗hδ(e1)w = \theta_* h_\delta(e_1) satisfies H(w)>α2K\mathcal{H}(w) > \alpha_2 K. Only convergence of these integrals is used. Uniform convergence up to the boundary is neither asserted nor needed.

For r∈[0,1]r \in[0,1] define

ψr(s)=(1−r)s+rhδ(s),vr=ψr(e1),ar(s)=(ψr′(s))p−1.\psi_r(s) = (1-r)s + r h_\delta(s),\qquad v_r = \psi_r(e_1),\qquad a_r(s) = (\psi_r'(s))^{p-1}.

Then e1≤vr≤1e_1 \leq v_r \leq1 and r↦vrr \mapsto v_r is continuous in C01(Ω‾)C_0^1(\overline{\Omega}). We verify a bound on ar′a_r' uniform in rr. This also addresses the singular case 1<p<21 < p < 2. Set B=max⁡{1,(Aδ)−1}B = \max\{1,(A\delta)^{-1}\}. Clearly 0≤ar≤Bp−10 \leq a_r \leq B^{p-1}. On δ<s<2δ\delta< s < 2\delta,

hδ′(s)=(2−s/δ)nAδ,hδ′′(s)=−n(2−s/δ)n−1Aδ2.h_\delta'(s) = \frac{(2-s/\delta)^n}{A\delta},\qquad h_\delta''(s) = -\frac{n(2-s/\delta)^{n-1}}{A\delta^2}.

For p≥2p \geq2 it follows that ∣ar′∣≤(p−1)Bp−2n/(Aδ2)|a_r'| \leq(p-1)B^{p-2}n/(A\delta^2). If 1<p<21 < p < 2 and r>0r > 0, the negative exponent p−2p-2 and 1−r+rhδ′≥rhδ′1-r+rh_\delta' \geq rh_\delta' give

∣ar′(s)∣=(p−1)r(1−r+rhδ′(s))p−2(−hδ′′(s))≤(p−1)nrp−1Ap−1δp(2−s/δ)n(p−1)−1≤(p−1)nAp−1δp.|a_r'(s)| = (p-1)r(1-r+rh_\delta'(s))^{p-2}(-h_\delta''(s)) \leq\frac{(p-1)nr^{p-1}}{A^{p-1}\delta^p}(2-s/\delta)^{n(p-1)-1} \leq\frac{(p-1)n}{A^{p-1}\delta^p}.

For r=0r=0 the derivative vanishes, as it does outside the transition interval. Since n(p−1)>1n(p-1)>1, ara_r is Lipschitz across both joining points, also for r=1r=1. Thus ar′≤0a'_r\leq0 and ∣ar′∣≤D\lvert a'_r\rvert\leq D, uniformly, where

D={(p−1)Bp−2n/(Aδ2),p≥2,(p−1)n/(Ap−1δp),1<p<2.D=\begin{cases} (p-1)B^{p-2}n/(A\delta^2), & p\geq2,\\ (p-1)n/(Ap^{-1}\delta^p), & 1<p<2. \end{cases}

Testing the weak eigenfunction equation with ar(e1)φa_r(e_1)\varphi gives

−Δpvr=λ1ar(e1)e1p−1−ar′(e1)∣∇e1∣p-\Delta_p v_r=\lambda_1a_r(e_1)e_1^{p-1}-a'_r(e_1)\lvert\nabla e_1\rvert^p

in the weak sense. {#eq:2.8}

Indeed, ∇vr=ψr′(e1)∇e1\nabla v_r=\psi'_r(e_1)\nabla e_1, and the product test is admissible since ara_r is Lipschitz and e1∈C1(Ω‾)e_1\in C^1(\overline{\Omega}). The chain rule is understood a.e. The values of ar′a'_r at joining points are immaterial since ∇e1=0\nabla e_1=0 a.e. on each corresponding level set. No second derivatives of e1e_1 are required. On {e1≤δ}\{e_1\leq\delta\}, vrv_r is a constant multiple of e1e_1, so the right-hand side is λ1vrp−1\lambda_1v_r^{p-1}. On {e1>δ}\{e_1>\delta\}, vr≥e1>δv_r\geq e_1>\delta. Consequently,

0≤−Δpvr≤Cδvrp−1,Cδ=max⁡{λ1,λ1Bp−1+D∥∇e1∥∞pδp−1}.(15)0\leq-\Delta_pv_r\leq C_\delta v_r^{p-1},\qquad C_\delta=\max\left\{\lambda_1,\frac{\lambda_1B^{p-1}+D\lVert\nabla e_1\rVert_\infty^p}{\delta^{p-1}}\right\}. \tag*{(15)}

Choose 0<θ0<min⁡{θ∗,η0}0<\theta_0<\min\{\theta_*,\eta_0\}, with η0\eta_0 from Lemma 2.3, and set

Lr=((1−r)θ0+rθ∗)vr.L_r=((1-r)\theta_0+r\theta_*)v_r.

By homogeneity, (15) holds also with LrL_r in place of vrv_r. Lemma 2.4 gives f(x,s)≥λ1c0sp−1f(x,s)\geq\lambda_1c_0s^{p-1} for 0<s≤θ∗0<s\leq\theta_*. Hence the choice

c∗=λ1c0Cδ>0(16)c_*=\frac{\lambda_1c_0}{C_\delta}>0 \tag*{(16)}

ensures −m(α)ΔpLr≤f(x,Lr)-m(\alpha)\Delta_pL_r\leq f(x,L_r) whenever 0<m(α)<c∗0<m(\alpha)<c_*. The remaining assertions follow from the construction and Lemma 2.3.

Auxiliary problems and endpoint estimates

Throughout this section the assumptions of Theorem 1.2 are in force. We first record the properties of the functional H\mathcal{H} defined in (1.1).

Lemma 3.1. Under (Hμ)(H_\mu), the following properties hold.

(a) H\mathcal{H} is continuous on C(Ω‾)C(\overline{\Omega}), in particular on C01(Ω‾)C^1_0(\overline{\Omega}).

(b) H\mathcal{H} preserves order on nonnegative functions, in the sense that 0≤v≤u0\leq v\leq u a.e. in Ω\Omega implies H(v)≤H(u)\mathcal{H}(v)\leq\mathcal{H}(u).

:::`

Proof. (a) Let un→uu_n \to u in C(Ω‾)C(\overline{\Omega}) and set c:=sup⁡n∥un∥∞+∥u∥∞<+∞c := \sup_n \|u_n\|_\infty+ \|u\|_\infty< +\infty. Since qi≥1q_i \ge1, the function t↦∣t∣qit \mapsto|t|^{q_i} is Lipschitz on [−c,c][-c,c], say with constant LiL_i, whence

∣H(un)−H(u)∣≤∑i=1MLi∥μi∥∞∣Ω∣ ∥un−u∥∞⟶0.|\mathcal{H}(u_n)-\mathcal{H}(u)| \le\sum_{i=1}^{M} L_i\|\mu_i\|_\infty|\Omega|\,\|u_n-u\|_\infty\longrightarrow0.

(b) Immediate from μi≥0\mu_i \ge0 a.e. in Ω\Omega and the monotonicity of t↦tqit \mapsto t^{q_i} on [0,+∞)[0,+\infty).

Fix θ0\theta_0, θ∗\theta_*, ww, c∗c_* and the family LrL_r given by Lemma 2.5. In particular,

H(w)>α2K.(17)\mathcal{H}(w) > \alpha_{2K}. \tag*{(17)}

We next arrange these barriers continuously along each parameter interval.

Lemma 3.2. For each k∈{1,…,K}k \in\{1,\ldots,K\} there exist

βk∈(α2k−1,τk),γk∈(τk,α2k)(18)\beta_k \in(\alpha_{2k-1},\tau_k), \qquad\gamma_k \in(\tau_k,\alpha_{2k}) \tag*{(18)}

and continuous maps ℓk−:[βk,τk]→C01(Ω‾)\ell_k^-:[\beta_k,\tau_k]\to C^1_0(\overline{\Omega}), ℓk+:[τk,γk]→C01(Ω‾)\ell_k^+:[\tau_k,\gamma_k]\to C^1_0(\overline{\Omega}) such that 0<ℓk±(α)≤θ∗<t∗0<\ell_k^\pm(\alpha)\le\theta_*<t_* in Ω\Omega and

−m(α)Δpℓk±(α)≤f(x,ℓk±(α))in the weak sense.(19)-m(\alpha)\Delta_p\ell_k^\pm(\alpha)\le f(x,\ell_k^\pm(\alpha)) \quad\text{in the weak sense}. \tag*{(19)}

Moreover,

ℓk−(βk)=ℓk+(γk)=w,ℓk−(τk)=ℓk+(τk)=θ0e1.(20)\ell_k^-(\beta_k)=\ell_k^+(\gamma_k)=w,\qquad\ell_k^-(\tau_k)=\ell_k^+(\tau_k)=\theta_0e_1. \tag*{(20)}

Proof. Continuity of mm and its zero endpoint values allow us to choose

α2k−1<βk<ρk−<τk<ρk+<γk<α2k\alpha_{2k-1}<\beta_k<\rho_k^-<\tau_k<\rho_k^+<\gamma_k<\alpha_{2k}

such that

m(α)<c∗for α∈[α2k−1,ρk−]∪[ρk+,α2k].(21)m(\alpha)<c_* \quad\text{for }\alpha\in[\alpha_{2k-1},\rho_k^-]\cup[\rho_k^+,\alpha_{2k}]. \tag*{(21)}

Define

rk−(α)={(ρk−−α)/(ρk−−βk),βk≤α≤ρk−,0,ρk−≤α≤τk,r_k^-(\alpha)= \begin{cases} (\rho_k^--\alpha)/(\rho_k^--\beta_k), & \beta_k\le\alpha\le\rho_k^-,\\ 0, & \rho_k^-\le\alpha\le\tau_k, \end{cases}
rk+(α)={0,τk≤α≤ρk+,(α−ρk+)/(γk−ρk+),ρk+≤α≤γk,r_k^+(\alpha)= \begin{cases} 0, & \tau_k\le\alpha\le\rho_k^+,\\ (\alpha-\rho_k^+)/(\gamma_k-\rho_k^+), & \rho_k^+\le\alpha\le\gamma_k, \end{cases}

and put ℓk±(α)=Lrk±(α)\ell_k^\pm(\alpha)=L_{r_k^\pm(\alpha)}. On the outer strips, (3.5) and Lemma 2.5 give the subsolution inequality for all values of rr. On the remaining parts, rk±=0r_k^\pm=0, so ℓk±=θ0e1\ell_k^\pm=\theta_0e_1, which is a subsolution throughout the positivity intervals by Lemma 2.3. Continuity and the endpoint values follow directly. For the illustration below, write the amplitude factors as

θk±(α)=(1−rk±(α))θ0+rk±(α)θ∗.(22)\theta_k^\pm(\alpha)=(1-r_k^\pm(\alpha))\theta_0+r_k^\pm(\alpha)\theta_*. \tag*{(22)}

Thus ℓk±(α)=θk±(α)ψrk±(α)(e1)\ell_k^\pm(\alpha)=\theta_k^\pm(\alpha)\psi_{r_k^\pm(\alpha)}(e_1). Both the amplitude and the spatial profile vary on the outer strips.

Figure 1 illustrates the construction.

Figure 1. Construction of the parameter-dependent barriers in Lemma 3.2. The upper panel shows a schematic positivity interval of mm. The lower panel shows the amplitude factors θk−\theta_k^{-} and θk+\theta_k^{+}. The spatial profile also varies through ψρk±(e1)\psi_{\rho_k^{\pm}}(e_1). No unimodality assumption on mm is used.

From now on we fix, for every k∈{1,…,K}k \in\{1,\ldots,K\}, a choice of βk\beta_k, γk\gamma_k and of the barrier functions ℓk±\ell_k^{\pm} provided by Lemma 3.2. Let II be one of the two intervals

Ik−:=[βk,τk],Ik+:=[τk,γk],I_k^{-}:=[\beta_k,\tau_k], \qquad I_k^{+}:=[\tau_k,\gamma_k],

and let ℓI(α)\ell_I(\alpha) denote the corresponding lower barrier, namely

ℓIk−(α)=ℓk−(α),ℓIk+(α)=ℓk+(α).\ell_{I_k^{-}}(\alpha)=\ell_k^{-}(\alpha), \qquad\ell_{I_k^{+}}(\alpha)=\ell_k^{+}(\alpha).

For α∈I\alpha\in I define the truncation

f^I(α,x,s)={f(x,ℓI(α)(x)),s<ℓI(α)(x),f(x,s),ℓI(α)(x)≤s≤t∗,0,s>t∗.(23)\widehat{f}_I(\alpha,x,s)= \begin{cases} f(x,\ell_I(\alpha)(x)), & s<\ell_I(\alpha)(x),\\ f(x,s), & \ell_I(\alpha)(x)\le s\le t_*,\\ 0, & s>t_*. \end{cases} \tag*{(23)}

Remark 3.3. For every (α,s)∈I×R(\alpha,s)\in I\times\mathbb{R} the function f^I(α,⋅,s)\widehat{f}_I(\alpha,\cdot,s) is measurable, and for almost every x∈Ωx\in\Omega the function f^I(⋅,x,⋅)\widehat{f}_I(\cdot,x,\cdot) is continuous on I×RI\times\mathbb{R}. Indeed, α↦ℓI(α)\alpha\mapsto\ell_I(\alpha) is continuous with values in C1(Ω‾)C^1(\overline{\Omega}), and the pieces in (3.7) agree at both joining points, at s=t∗s=t_* because of (Hf)(i)(H_f)(i). Moreover, (2.2) gives

0≤f^I(α,x,s)≤Tfor a.e. x∈Ω and every (α,s)∈I×R.(24)0\le\widehat{f}_I(\alpha,x,s)\le T \qquad\text{for a.e. }x\in\Omega\text{ and every }(\alpha,s)\in I\times\mathbb{R}. \tag*{(24)}

Lemma 3.4. Fix k∈{1,…,K}k \in\{1,\ldots,K\} and let I∈{Ik−,Ik+}I \in\{I_k^-,I_k^+\}. For every (α,v)∈I×C01(Ω‾)(\alpha,v) \in I \times C_0^1(\overline{\Omega}), the problem

{−m(α)Δpz=f^I(α,x,v(x))in Ω,z=0on ∂Ω(25)\begin{cases} -m(\alpha)\Delta_p z=\widehat{f}_I(\alpha,x,v(x)) & \text{in }\Omega,\\ z=0 & \text{on }\partial\Omega \tag*{(25)} \end{cases}

has a unique weak solution zz, and z≥0z \ge0 a.e. in Ω\Omega. We denote it by

TI(α,v):=z.\mathcal{T}_I(\alpha,v):=z.

Moreover, TI:I×C01(Ω‾)→C01(Ω‾)\mathcal{T}_I:I\times C_0^1(\overline{\Omega})\to C_0^1(\overline{\Omega}) is continuous, and TI(I×C01(Ω‾))\mathcal{T}_I(I\times C_0^1(\overline{\Omega})) is a relatively compact subset of C01(Ω‾)C_0^1(\overline{\Omega}).

Proof. The interval II is a compact subset of a positivity interval of mm, so

mI:=min⁡α∈Im(α)>0.(26)m_I:=\min_{\alpha\in I}m(\alpha)>0. \tag*{(26)}

For fixed (α,v)(\alpha,v) the right-hand side of (25) is a prescribed function in L∞(Ω)L^\infty(\Omega), bounded between 00 and TT by (24). The theorem of Minty and Browder therefore gives a unique weak solution zz. Testing with −z−-z^- gives

m(α)∫Ω∣∇z−∣p dx=−∫Ωf^I(α,x,v)z− dx≤0,m(\alpha)\int_{\Omega}|\nabla z^-|^p\,dx=-\int_{\Omega}\widehat{f}_I(\alpha,x,v)z^-\,dx\le0,

and hence z≥0z\ge0.

Next, let w0∈W01,p(Ω)∩L∞(Ω)w_0\in W_0^{1,p}(\Omega)\cap L^\infty(\Omega) be the solution of −Δpw0=1-\Delta_p w_0=1 in Ω\Omega. Since

−Δpz=f^I(α,x,v)m(α)≤TmI=−Δp((TmI)1/(p−1)w0)in Ω,-\Delta_p z=\frac{\widehat{f}_I(\alpha,x,v)}{m(\alpha)}\le\frac{T}{m_I}=-\Delta_p\left(\left(\frac{T}{m_I}\right)^{1/(p-1)}w_0\right)\quad\text{in }\Omega,

the weak comparison principle gives 0≤z≤(T/mI)1/(p−1)w00\le z\le(T/m_I)^{1/(p-1)}w_0, a bound in L∞(Ω)L^\infty(\Omega) independent of (α,v)(\alpha,v). The global C1,νC^{1,\nu} regularity estimate for bounded data [18] therefore gives ν∈(0,1)\nu\in(0,1) and CI>0C_I>0, independent of (α,v)(\alpha,v), such that

∥TI(α,v)∥C1,ν(Ω‾)≤CI.(27)\|\mathcal{T}_I(\alpha,v)\|_{C^{1,\nu}(\overline{\Omega})}\le C_I. \tag*{(27)}

The compact embedding into C1(Ω‾)C^1(\overline{\Omega}) proves relative compactness of the image.

To prove continuity, let (αn,vn)→(α,v)(\alpha_n,v_n)\to(\alpha,v) in I×C01(Ω‾)I\times C_0^1(\overline{\Omega}) and set zn=TI(αn,vn)z_n=\mathcal{T}_I(\alpha_n,v_n). Every subsequence has a further subsequence converging in C01(Ω‾)C_0^1(\overline{\Omega}) to some zz. By Remark 3.3 and dominated convergence,

f^I(αn,x,vn(x))⟶f^I(α,x,v(x))in Lp′(Ω).\widehat{f}_I(\alpha_n,x,v_n(x))\longrightarrow\widehat{f}_I(\alpha,x,v(x))\quad\text{in }L^{p'}(\Omega).

Passing to the limit in the weak formulation shows that z=TI(α,v)z=\mathcal{T}_I(\alpha,v) by uniqueness for the prescribed right-hand side. Every convergent subsequence has this same limit, so the full sequence converges in C01(Ω‾)C_0^1(\overline{\Omega}).

Uniqueness in the above lemma concerns the problem with the given input vv. It does not assert uniqueness of fixed points of TI(α,⋅)\mathcal{T}_I(\alpha,\cdot).

Lemma 3.5. Fix k∈{1,…,K}k \in\{1,\ldots,K\}, let I∈{Ik−,Ik+}I \in\{I_k^-,I_k^+\} and let (α,u)∈I×C01(Ω‾)(\alpha,u) \in I \times C_0^1(\overline{\Omega}) be such that

u=TI(α,u).u=\mathcal{T}_I(\alpha,u).

Then

ℓI(α)≤u≤t∗a.e. in Ω.(28)\ell_I(\alpha) \le u \le t_* \quad\text{a.e. in } \Omega. \tag*{(28)}

Consequently, uu solves the untruncated frozen problem

−m(α)Δpu=f(x,u),u∈W01,p(Ω).(29)-m(\alpha)\Delta_p u=f(x,u), \qquad u\in W_0^{1,p}(\Omega). \tag*{(29)}

Proof. Write ℓ=ℓI(α)\ell=\ell_I(\alpha). Since u=TI(α,u)u=\mathcal{T}_I(\alpha,u),

m(α)∫Ω∣∇u∣p−2∇u⋅∇φ dx=∫Ωf^I(α,x,u)φ dxfor every φ∈W01,p(Ω).(30)m(\alpha)\int_\Omega|\nabla u|^{p-2}\nabla u\cdot\nabla\varphi\,dx =\int_\Omega\widehat{f}_I(\alpha,x,u)\varphi\,dx \quad\text{for every }\varphi\in W_0^{1,p}(\Omega). \tag*{(30)}

Take φ=(ℓ−u)+\varphi=(\ell-u)^+. On the set {u<ℓ}\{u<\ell\}, where φ\varphi may be nonzero, the truncated reaction equals f(x,ℓ)f(x,\ell). Subtracting (3.14) from the weak form of the subsolution inequality (3.3) for ℓ\ell, both tested with φ\varphi, yields

m(α)∫Ω(∣∇ℓ∣p−2∇ℓ−∣∇u∣p−2∇u)⋅∇(ℓ−u)+ dx≤0.m(\alpha)\int_\Omega \left(|\nabla\ell|^{p-2}\nabla\ell-|\nabla u|^{p-2}\nabla u\right) \cdot\nabla(\ell-u)^+\,dx\le0.

The integrand is nonnegative by the monotonicity of ξ↦∣ξ∣p−2ξ\xi\mapsto|\xi|^{p-2}\xi, and since m(α)>0m(\alpha)>0 it vanishes a.e. in Ω\Omega. By strict monotonicity, ∇ℓ=∇u\nabla\ell=\nabla u a.e. on {u<ℓ}\{u<\ell\}, that is, ∇(ℓ−u)+=0\nabla(\ell-u)^+=0 a.e. in Ω\Omega. Poincaré’s inequality gives (ℓ−u)+=0(\ell-u)^+=0, and hence u≥ℓu\ge\ell.

Next take φ=(u−t∗)+\varphi=(u-t_*)^+, which belongs to W01,p(Ω)W_0^{1,p}(\Omega). The truncated reaction vanishes on {u>t∗}\{u>t_*\}, so

m(α)∫Ω∣∇(u−t∗)+∣p dx=0.m(\alpha)\int_\Omega|\nabla(u-t_*)^+|^p\,dx=0.

Poincaré’s inequality now gives u≤t∗u\le t_*. The truncation therefore agrees with f(x,u)f(x,u) a.e., proving (3.13).

Lemma 3.6. Fix k∈{1,…,K}k\in\{1,\ldots,K\}. Every weak solution u∈W01,p(Ω)u\in W_0^{1,p}(\Omega) of

−m(τk)Δpu=f(x,u)in Ω,u=0on ∂Ω,(31)-m(\tau_k)\Delta_pu=f(x,u)\quad\text{in }\Omega,\qquad u=0\quad\text{on }\partial\Omega, \tag*{(31)}

with 0≤u≤t∗0\le u\le t_* a.e. satisfies

H(u)<τk.(32)\mathcal{H}(u)<\tau_k. \tag*{(32)}

Proof. Testing with uu and using (10) gives

m(τk)∥u∥p=∫Ωf(x,u)u dx≤∣Ω∣E.m(\tau_k)\lVert u\rVert^p = \int_{\Omega} f(x,u)u\,dx \leq|\Omega|E.

Since uqi≤t∗qi−1uu^{q_i} \leq t_*^{q_i-1}u for 0≤u≤t∗0 \leq u \leq t_*, Hölder’s inequality and (8) imply

H(u)≤∑i=1Mt∗qi−1∥μi∥p′∥u∥p≤λ1−1/p∑i=1Mt∗qi−1∥μi∥p′∥u∥≤(∣Ω∣Eλ1m(τk))1/p∑i=1Mt∗qi−1∥μi∥p′<τk.(33)\begin{aligned} \mathcal{H}(u) \leq\sum_{i=1}^{M}t_*^{q_i-1}\lVert\mu_i\rVert_{p'}\lVert u\rVert_p \leq\lambda_1^{-1/p}\sum_{i=1}^{M}t_*^{q_i-1}\lVert\mu_i\rVert_{p'}\lVert u\rVert\\ &\leq\left(\frac{|\Omega|E}{\lambda_1m(\tau_k)}\right)^{1/p}\sum_{i=1}^{M}t_*^{q_i-1}\lVert\mu_i\rVert_{p'}<\tau_k. \tag*{(33)} \end{aligned}

The last inequality is (Hτ)(H_{\tau}) in equivalent form. □\square

Proofs of the main results

Proof of Theorem 1.2. Set X=C01(Ω‾)X=C^1_0(\overline{\Omega}). The uniform bounds in (27) and the finiteness of the family of intervals Ik±I_k^{\pm} give R>0R>0 such that each TIk±T_{I_k^{\pm}} maps Ik±×BRI_k^{\pm}\times B_R into BR={v∈X:∥v∥C1(Ω‾)≤R}B_R=\{v\in X:\lVert v\rVert_{C^1(\overline{\Omega})}\leq R\}. By Lemmas 3.1 and 3.4, Φ=H\Phi=\mathcal{H} is continuous and these solution operators are continuous and compact. Thus the functional and operator assumptions of Lemma 2.2 hold.

Fix kk. If u=TIk−(βk,u)u=T_{I_k^-}(\beta_k,u) or u=TIk+(γk,u)u=T_{I_k^+}(\gamma_k,u), then Lemma 3.5 and (20) give u≥wu\geq w. Since H\mathcal{H} preserves order on nonnegative functions,

H(u)≥H(w)>α2K≥α2k>γk>βk.(34)\mathcal{H}(u)\geq\mathcal{H}(w)>\alpha_2K\geq\alpha_{2k}>\gamma_k>\beta_k. \tag*{(34)}

At the common endpoint τk\tau_k, every fixed point of either operator solves (31) and lies between 00 and t∗t_* by Lemma 3.5. Lemma 3.6 therefore gives

H(u)<τk.\mathcal{H}(u)<\tau_k.

These estimates hold for every fixed point at the respective endpoints, as required in Lemma 2.2.

Apply Lemma 2.2(a) on [βk,τk][\beta_k,\tau_k] and part (b) on [τk,γk][\tau_k,\gamma_k]. We obtain pairs (ak,1,uk,1)(a_{k,1},u_{k,1}) and (ak,2,uk,2)(a_{k,2},u_{k,2}) with

uk,1=TIk−(ak,1,uk,1),uk,2=TIk+(ak,2,uk,2),ak,j=H(uk,j),u_{k,1}=T_{I_k^-}(a_{k,1},u_{k,1}),\qquad u_{k,2}=T_{I_k^+}(a_{k,2},u_{k,2}),\qquad a_{k,j}=\mathcal{H}(u_{k,j}),

and

βk<ak,1<τk<ak,2<γk.\beta_k<a_{k,1}<\tau_k<a_{k,2}<\gamma_k.

By Lemma 3.5, each uk,ju_{k,j} solves

−m(ak,j)Δpuk,j=f(x,uk,j),0<uk,j≤t∗in Ω.-m(a_{k,j})\Delta_pu_{k,j}=f(x,u_{k,j}),\qquad0<u_{k,j}\leq t_* \quad\text{in }\Omega.

Substituting ak,j=H(uk,j)a_{k,j}=\mathcal{H}(u_{k,j}) proves that these are positive solutions of (P). They belong to C01(Ω‾)C^1_0(\overline{\Omega}) by construction. Since α2k−1<βk\alpha_{2k-1}<\beta_k and γk<α2k\gamma_k<\alpha_{2k} by (3.2), this gives (1.4). Distinct nonlocal values separate the two solutions in each pair, and the disjoint positivity intervals of mm separate the pairs. Hence all 2K2K solutions are distinct.

Proof of Theorem 1.4. Fix the function ww given by Lemma 2.5 and the intervals Ik−=[βk,τk]I_k^-=[\beta_k,\tau_k] and Ik+=[τk,γk]I_k^+=[\tau_k,\gamma_k], where βk\beta_k and γk\gamma_k are chosen as in Lemma 3.2. Define

Rk=(∣Ω∣Eλ1m(τk))1/p∑i=1Mq∗i−1∥μi∥p′,(35)R_k=\left(\frac{|\Omega|E}{\lambda_1m(\tau_k)}\right)^{1/p}\sum_{i=1}^{M}q_*^{i-1}\lVert\mu_i\rVert_{p'}, \tag*{(35)}
d=min⁡{H(w)−α2K,min⁡1≤k≤K(τk−Rk)}>0,0<ε0<d.(36)d=\min\left\{\mathcal{H}(w)-\alpha_{2K},\min_{1\leq k\leq K}(\tau_k-R_k)\right\}>0,\qquad0<\varepsilon_0<d. \tag*{(36)}

Positivity follows from (3.1) and (Hτ)(H_\tau). Any choice of ε0∈(0,d)\varepsilon_0\in(0,d) gives a sufficient perturbation tolerance.

Fix a continuous functional H~:X→[0,∞)\widetilde{\mathcal{H}}:X\to[0,\infty) satisfying (1.5). Use the barriers ℓk±\ell_k^\pm and operators TIk±T_{I_k^\pm} constructed in Section 3. Their construction uses ff, mm and the fixed barriers, and does not require the functional appearing in the scalar constraint to preserve order. Choose the same ball BR⊂XB_R\subset X as in the proof of Theorem 1.2. Lemma 3.4 gives continuous compact maps Ik±×BR→BRI_k^\pm\times B_R\to B_R. We apply Lemma 2.2 with Φ=H~\Phi=\widetilde{\mathcal{H}}.

Every fixed point lies in DD by Lemma 3.5. At either outer endpoint, (3.4) gives u≥wu\geq w. The reference functional H\mathcal{H} preserves order, so

H~(u)≥H(u)−ε0≥H(w)−ε0≥α2K+d−ε0>α2K>γk>βk.\begin{aligned} \widetilde{\mathcal{H}}(u)\geq\mathcal{H}(u)-\varepsilon_0\geq\mathcal{H}(w)-\varepsilon_0 \\ &\geq\alpha_{2K}+d-\varepsilon_0>\alpha_{2K}>\gamma_k>\beta_k. \end{aligned}

At the common endpoint τk\tau_k, the fixed point solves the untruncated frozen equation. The estimate (3.17) gives H(u)≤Rk\mathcal{H}(u)\leq R_k, and hence

H~(u)≤Rk+ε0≤τk−d+ε0<τk.\widetilde{\mathcal{H}}(u)\leq R_k+\varepsilon_0\leq\tau_k-d+\varepsilon_0<\tau_k.

Both estimates hold for all fixed points at the corresponding endpoints. Lemma 2.2(a) on Ik−I_k^- and part (b) on Ik+I_k^+ therefore provide points (ak,j,uk,j)(a_{k,j},u_{k,j}) satisfying

uk,j=TIk±(ak,j,uk,j),ak,j=H~(uk,j),u_{k,j}=T_{I_k^\pm}(a_{k,j},u_{k,j}),\qquad a_{k,j}=\widetilde{\mathcal{H}}(u_{k,j}),
βk<ak,1<τk<ak,2<γk.\beta_k<a_{k,1}<\tau_k<a_{k,2}<\gamma_k.

By Lemma 3.5, substitution of this scalar identity into the frozen equation gives (1.6). Positivity follows from the lower barrier, and regularity follows from the definition of TIT_I. The disjoint intervals containing the values of H~\widetilde{\mathcal{H}} give (1.7) and separate all 2K2K solutions.

Examples

Example 5.1. Let N=2N = 2, p=3p = 3, t∗>1t_\ast> 1 and Ω=B(x0,3)\Omega= B(x_0,3). Choose x2∈B(x0,2)x_2 \in B(x_0,2), set Ω1=B(x0,1)\Omega_1 = B(x_0,1) and Ω2=B(x2,1)\Omega_2 = B(x_2,1) (so that Ω2⊂Ω\Omega_2 \subset\Omega), and define

μ1=χΩ1,μ2=2χΩ2.\mu_1 = \chi_{\Omega_1}, \qquad\mu_2 = 2\chi_{\Omega_2}.

These weights satisfy (Hμ)(H_\mu). The two regions may overlap, and their union does not cover Ω\Omega.

Fix any q1,q2≥1q_1,q_2 \ge1. Since ∣Ω1∣=∣Ω2∣=π|\Omega_1| = |\Omega_2| = \pi and t∗>1t_\ast> 1,

M∗=πt∗q1+2πt∗q2>3π>32.M_\ast= \pi t_\ast^{q_1} + 2\pi t_\ast^{q_2} > 3\pi> \frac{3}{2}.

Thus (HM)(H_M) holds for the coefficient mm defined in (5.1) below, for which K=1K = 1 and α2K=α2=3/2\alpha_{2K} = \alpha_2 = 3/2.

We consider the autonomous Carathéodory reaction f(x,t)=f(t)f(x,t) = f(t), where

f(t)=λ1(t∗−t)2(1+34sin⁡4πtt∗),t∈R.f(t) = \lambda_1(t_\ast-t)^2\left(1+\frac{3}{4}\sin\frac{4\pi t}{t_\ast}\right), \qquad t \in\mathbb{R}.

Since

1+34sin⁡4πtt∗≥14for all t∈R,1+\frac{3}{4}\sin\frac{4\pi t}{t_\ast} \ge\frac{1}{4} \qquad\text{for all } t \in\mathbb{R},

we have f(t)≥0f(t) \ge0 for all t∈Rt \in\mathbb{R}. Moreover, f(t∗)=0f(t_\ast) = 0. If J⋐(0,t∗)J \Subset(0,t_\ast), then inf⁡t∈Jf(t)>0\inf_{t\in J} f(t) > 0, and

sup⁡t∈[0,t∗]f(t)≤74λ1t∗2=:T.\sup_{t\in[0,t_\ast]} f(t) \le\frac{7}{4}\lambda_1t_\ast^2 =: T.

Thus (Hf)(H_f) holds. Furthermore, since p=3p = 3,

lim inf⁡t→0+f(t)tp−1=lim inf⁡t→0+λ1(t∗−t)2t2(1+34sin⁡4πtt∗)=+∞.\liminf_{t\to0^+}\frac{f(t)}{t^{p-1}} = \liminf_{t\to0^+}\lambda_1\frac{(t_\ast-t)^2}{t^2}\left(1+\frac{3}{4}\sin\frac{4\pi t}{t_\ast}\right) = +\infty.

Therefore (Hλ)(H_\lambda) holds for every continuous coefficient mm satisfying (Hm)(H_m), in particular for the one defined in (5.1) below.

Let us also observe that the quotient t↦f(t)/tp−1t \mapsto f(t)/t^{p-1} is not monotone on (0,t∗)(0,t_\ast). Indeed, for t∈(0,t∗)t \in(0,t_\ast) set

Q(t):=f(t)t2=λ1(t∗−t)2t2(1+34sin⁡4πtt∗).Q(t) := \frac{f(t)}{t^2} = \lambda_1\frac{(t_\ast-t)^2}{t^2}\left(1+\frac{3}{4}\sin\frac{4\pi t}{t_\ast}\right).

Since Q(t)→+∞Q(t) \to+\infty as t→0+t \to0^+, QQ is not nondecreasing. On the other hand, writing Q=λ1hgQ = \lambda_1hg with h(t)=(t∗−t)2t2h(t) = \frac{(t_\ast-t)^2}{t^2} and g(t)=1+34sin⁡4πtt∗g(t) = 1+\frac{3}{4}\sin\frac{4\pi t}{t_\ast}, we have h(t∗/2)=g(t∗/2)=1h(t_\ast/2) = g(t_\ast/2) = 1, h′(t∗/2)=−8/t∗h'(t_\ast/2) = -8/t_\ast and g′(t∗/2)=3π/t∗g'(t_\ast/2) = 3\pi/t_\ast, whence

Q′(t∗2)=λ1(h′g+hg′)(t∗2)=λ1t∗(3π−8)>0.Q'\left(\frac{t_\ast}{2}\right) = \lambda_1\left(h'g+hg'\right)\left(\frac{t_\ast}{2}\right) = \frac{\lambda_1}{t_\ast}(3\pi-8) > 0.

Thus QQ is not nonincreasing either. This reaction violates the decrease condition used for the multiplicity results in [4, 15, 16, 17].

Let α1=1<α2=32\alpha_1=1<\alpha_2=\frac{3}{2} and define

m(s)=As(s−1)(32−s),s≥0,(37)m(s)=As(s-1)\left(\frac{3}{2}-s\right),\qquad s\geq0, \tag*{(37)}

where

A>286721875π3t∗3(t∗q1−1+2t∗q2−1)3.(38)A>\frac{28672}{1875}\pi^3t_*^3\left(t_*^{q_1-1}+2t_*^{q_2-1}\right)^3. \tag*{(38)}

Then m>0m>0 in (1,3/2)(1,3/2) and m(1)=m(3/2)=0m(1)=m(3/2)=0, hence (Hm)(H_m) holds with K=1K=1.

It remains to verify (Hτ)(H_\tau). Since K=1K=1, we choose τ1=5/4\tau_1=5/4 and write τ:=τ1\tau:=\tau_1 for brevity. By (5.1) and (5.2),

τ3m(τ)=Aτ4(τ−1)(32−τ)=A6254096>73π3t∗3(t∗q1−1+2t∗q2−1)3.(39)\tau^3m(\tau)=A\tau^4(\tau-1)\left(\frac{3}{2}-\tau\right)=A\frac{625}{4096}>\frac{7}{3}\pi^3t_*^3\left(t_*^{q_1-1}+2t_*^{q_2-1}\right)^3. \tag*{(39)}

Moreover, for t∈[0,t∗]t\in[0,t_*],

tf(t)≤74λ1t(t∗−t)2.tf(t)\leq\frac{7}{4}\lambda_1t(t_*-t)^2.

Since

max⁡t∈[0,t∗]t(t∗−t)2=4t∗327,\max_{t\in[0,t_*]}t(t_*-t)^2=\frac{4t_*^3}{27},

we get

max⁡t∈[0,t∗][tf(t)]≤λ17t∗327.\max_{t\in[0,t_*]}[tf(t)]\leq\lambda_1\frac{7t_*^3}{27}.

Since ∣Ω∣=9π|\Omega|=9\pi, p′=3/2p'=3/2, and

∥μ1∥3/2=∥χΩ1∥3/2=π2/3,∥μ2∥3/2=∥2χΩ2∥3/2=2π2/3,\|\mu_1\|_{3/2}=\|\chi_{\Omega_1}\|_{3/2}=\pi^{2/3},\qquad \|\mu_2\|_{3/2}=\|2\chi_{\Omega_2}\|_{3/2}=2\pi^{2/3},

we obtain

∣Ω∣λ1(t∗q1−1∥μ1∥3/2+t∗q2−1∥μ2∥3/2)3max⁡t∈[0,t∗][tf(t)]≤9πλ1(t∗q1−1π2/3+2t∗q2−1π2/3)3λ17t∗327=73π3t∗3(t∗q1−1+2t∗q2−1)3.\begin{aligned} \frac{|\Omega|}{\lambda_1}\left(t_*^{q_1-1}\|\mu_1\|_{3/2}+t_*^{q_2-1}\|\mu_2\|_{3/2}\right)^3\max_{t\in[0,t_*]}[tf(t)] \\ &\leq\frac{9\pi}{\lambda_1}\left(t_*^{q_1-1}\pi^{2/3}+2t_*^{q_2-1}\pi^{2/3}\right)^3\lambda_1\frac{7t_*^3}{27} =\frac{7}{3}\pi^3t_*^3\left(t_*^{q_1-1}+2t_*^{q_2-1}\right)^3. \end{aligned}

Combining this estimate with (5.3), we obtain (Hτ)(H_\tau).

All assumptions of Theorem 1.2 are now verified. Problem (P), with the weights, reaction and coefficient specified above, therefore has two positive solutions u1,u2∈C01(Ω‾)u_1,u_2\in C^1_0(\overline{\Omega}) satisfying

1<H(u1)<54<H(u2)<32,H(u)=∥u∥q1,Ω1q1+2∥u∥q2,Ω2q2.1<\mathcal{H}(u_1)<\frac{5}{4}<\mathcal{H}(u_2)<\frac{3}{2},\qquad \mathcal{H}(u)=\|u\|_{q_1,\Omega_1}^{q_1}+2\|u\|_{q_2,\Omega_2}^{q_2}.

The normalized reaction is nonmonotone, so the cited multiplicity theorems requiring a decreasing quotient do not apply to these data.

Example 5.2 (Explicit data for a quadratic coefficient). Let p=3p=3, M=1M=1, q1=1q_1=1, μ1=1\mu_1=1, and t∗>0t_* > 0, on any bounded smooth domain as above. Then H(u)=∫Ω∣u∣ dx\mathcal{H}(u)=\int_{\Omega}|u|\,dx, M∗=t∗∣Ω∣M_* = t_*|\Omega|. Set b=M∗/2b=M_*/2, a=b/2a=b/2, τ=3b/4\tau=3b/4, and define

f(x,t)=(t∗−t)2(t∈R),m(s)=A0(s−a)(b−s)(s≥0).f(x,t)=(t_*-t)^2 \qquad(t\in\mathbb{R}), \qquad m(s)=A_0(s-a)(b-s) \qquad(s\ge0).

The reaction satisfies (Hf)(H_f), with T=t∗2T=t_*^2, and f(x,t)/t2→+∞f(x,t)/t^2\to+\infty as t→0+t\to0^+, so (Hλ)(H_\lambda) holds for every finite A0>0A_0>0. The coefficient satisfies (Hm)(H_m) with K=1K=1, α1=a\alpha_1=a and α2=b\alpha_2=b. Since b<M∗b<M_*, (HM)(H_M) also holds. Moreover,

E=max⁡0≤t≤t∗t(t∗−t)2=4t∗327,∥μ1∥3/2=∣Ω∣2/3.E=\max_{0\le t\le t_*}t(t_*-t)^2=\frac{4t_*^3}{27}, \qquad\|\mu_1\|_{3/2}=|\Omega|^{2/3}.

Consequently (Hτ)(H_\tau) holds whenever

A0>4∣Ω∣t∗327λ1τ3(τ−a)(b−τ).(40)A_0>\frac{4|\Omega|t_*^3}{27\lambda_1\tau^3(\tau-a)(b-\tau)}. \tag*{(40)}

For every such A0A_0, Theorem 1.2 gives two positive solutions u1,u2u_1,u_2 with

a<H(u1)<τ<H(u2)<b.a<\mathcal{H}(u_1)<\tau<\mathcal{H}(u_2)<b.

These data will also be used in Example 5.3.

Example 5.3 (A perturbation depending on an additional spatial integral). This example illustrates Theorem 1.4 through a perturbation involving an additional spatial integral. We verify that order preservation may fail and show that in general the effect of the perturbation cannot be reproduced by changing mm while keeping H\mathcal{H} unchanged.

Use the data from Example 5.2, with A0A_0 satisfying (40), and let d>0d>0 be the margin defined in (36) in the proof of Theorem 1.4. Let U⊆ΩU\subseteq\Omega be any measurable set with ∣U∣>0|U|>0, and put

J(u)=∫U∣u∣ dx,0<η<min⁡{1/5,d/M∗}.J(u)=\int_U |u|\,dx, \qquad0<\eta<\min\{1/5,d/M_*\}.

For ω>0\omega>0 define

H~η,ω(u)=H(u)[1+ηsin⁡(ωJ(u))],u∈X.(41)\widetilde{\mathcal{H}}_{\eta,\omega}(u)=\mathcal{H}(u)[1+\eta\sin(\omega J(u))], \qquad u\in X. \tag*{(41)}

This functional is continuous and nonnegative on XX. For u∈Du\in D,

∣H~η,ω(u)−H(u)∣≤ηH(u)≤ηM∗<d.|\widetilde{\mathcal{H}}_{\eta,\omega}(u)-\mathcal{H}(u)|\le\eta\mathcal{H}(u)\le\eta M_*<d.

The proof of Theorem 1.4 allows any ε0∈(0,d)\varepsilon_0\in(0,d). Choosing ε0∈(ηM∗,d)\varepsilon_0\in(\eta M_*,d) therefore gives two positive solutions for every ω>0\omega>0.

We now choose ω\omega so that the functional fails to preserve order. Set s0=5b/8=5M∗/16s_0=5b/8=5M_*/16 and choose v0∈Xv_0\in X such that

0<v0<t∗in Ω,H(v0)=s0.0<v_0<t_* \quad\text{in }\Omega,\qquad\mathcal{H}(v_0)=s_0.

Such a choice is possible since s0<M∗s_0 < M_*, by approximating the constant t∗t_* from below with positive functions in XX and rescaling. Since ∣U∣>0|U| > 0 and v0>0v_0 > 0 in Ω\Omega,

ρ=J(v0)s0∈(0,1].\rho= \frac{J(v_0)}{s_0} \in(0,1].

Choose an integer j≥1j \ge1 such that η(2j−1)π>1\eta(2j-1)\pi> 1, and set

ω=(4j+1)π2ρs0,s1=(2j−1)πωρ∈(0,s0).\omega= \frac{(4j+1)\pi}{2\rho s_0}, \qquad s_1 = \frac{(2j-1)\pi}{\omega\rho} \in(0,s_0).

Along the ordered family us=(s/s0)v0u_s = (s/s_0)v_0, 0<s<s00 < s < s_0, we have H(us)=s\mathcal{H}(u_s) = s, J(us)=ρsJ(u_s) = \rho s and us∈Du_s \in D. Consequently, for G(s)=s[1+ηsin⁡(ωρs)]G(s) = s[1+\eta\sin(\omega\rho s)],

H~η,ω(us)=G(s),G′(s1)=1−η(2j−1)π<0.\widetilde{\mathcal{H}}_{\eta,\omega}(u_s) = G(s), \qquad G'(s_1) = 1-\eta(2j-1)\pi< 0.

There are therefore s−<s+s_- < s_+ near s1s_1 with us−≤us+u_{s_-} \le u_{s_+} but

H~η,ω(us−)>H~η,ω(us+).\widetilde{\mathcal{H}}_{\eta,\omega}(u_{s_-}) > \widetilde{\mathcal{H}}_{\eta,\omega}(u_{s_+}).

In particular, this functional cannot be a sum of integral powers with non-negative weights, since every functional of that form preserves order on non-negative functions.

To rule out absorption of the perturbation into mm, assume in addition that ∣U∣<∣Ω∣|U| < |\Omega|. Choose h∈C∞(Ω)h \in C^\infty(\Omega) such that

∫Ωh dx=0,∫Uh dx=1.\int_{\Omega} h\,dx = 0, \qquad\int_U h\,dx = 1.

Such an hh exists because the two integration functionals are linearly independent on C∞(Ω)C^\infty(\Omega). Otherwise the characteristic function of UU would be constant a.e. in Ω\Omega, contrary to 0<∣U∣<∣Ω∣0 < |U| < |\Omega|. For sufficiently small ε>0\varepsilon> 0, the function vε=v0+εhv_\varepsilon= v_0 + \varepsilon h belongs to XX and satisfies 0<vε<t∗0 < v_\varepsilon< t_* in Ω\Omega. Moreover,

H(vε)=H(v0)=s0,J(vε)=J(v0)+ε.\mathcal{H}(v_\varepsilon) = \mathcal{H}(v_0) = s_0, \qquad J(v_\varepsilon) = J(v_0) + \varepsilon.

Since ωJ(v0)=(4j+1)π/2\omega J(v_0) = (4j+1)\pi/2, choosing also 0<ωε<π/20 < \omega\varepsilon< \pi/2 gives

H~η,ω(vε)=s0[1+ηcos⁡(ωε)]<s0(1+η)=H~η,ω(v0).\widetilde{\mathcal{H}}_{\eta,\omega}(v_\varepsilon) = s_0[1+\eta\cos(\omega\varepsilon)] < s_0(1+\eta) = \widetilde{\mathcal{H}}_{\eta,\omega}(v_0).

As η<1/5\eta< 1/5, both values belong to (a,τ)(a,\tau), since

a<s0<H~η,ω(vε)<H~η,ω(v0)=s0(1+η)<τ=(a+b)/2.a < s_0 < \widetilde{\mathcal{H}}_{\eta,\omega}(v_\varepsilon) < \widetilde{\mathcal{H}}_{\eta,\omega}(v_0) = s_0(1+\eta) < \tau= (a+b)/2.

The coefficient m(s)=A0(s−a)(b−s)m(s) = A_0(s-a)(b-s) is strictly increasing on (a,τ)(a,\tau). Hence both coefficient values are positive and

m(H~η,ω(vε))<m(H~η,ω(v0)).m\bigl(\widetilde{\mathcal{H}}_{\eta,\omega}(v_\varepsilon)\bigr) < m\bigl(\widetilde{\mathcal{H}}_{\eta,\omega}(v_0)\bigr).

As the reference values are equal, no function m^:[0,∞)→R\widehat{m} : [0,\infty) \to\mathbb{R} can satisfy

m(H~η,ω(u))=m^(H(u))for every u∈D.m\left(\widetilde{\mathcal{H}}_{\eta,\omega}(u)\right) = \widehat{m}\left(\mathcal{H}(u)\right) \quad\text{for every } u \in\mathcal{D}.

The obstruction already occurs for strictly positive functions in D\mathcal{D}, since both v0v_0 and vεv_\varepsilon are positive in Ω\Omega.

If ∣U∣=∣Ω∣|U| = |\Omega|, then J=HJ = \mathcal{H}, and the perturbation can be absorbed into mm by setting

m^(s)=m(s[1+ηsin⁡(ωs)]),s≥0.\widehat{m}(s) = m\left(s[1+\eta\sin(\omega s)]\right), \qquad s \ge0.

Acknowledgements

The first author is a member of the Gruppo Nazionale per l’Analisi Matematica, la Probabilità e le loro Applicazioni (GNAMPA) of the Istituto Nazionale di Alta Matematica (INdAM). This work was partially supported by the INdAM GNAMPA Project with CUP E5324001950001.

Conflict of interest

The authors declare that they have no conflict of interest.

References

  1. [1]C. O. Alves, D.-P. Covei, Existence of solution for a class of nonlocal elliptic problem via sub–supersolution method, Nonlinear Anal. Real World Appl. 23 (2015), 1–8. 10.1016/j.nonrwa.2014.11.003arxiv.org/abs/1306.1211
  2. [2]H. Brézis, L. Oswald, Remarks on sublinear elliptic equations, Nonlinear Anal. 10 (1986), no. 1, 55–64. 10.1016/0362-546X(86)90011-8DOI
  3. [3]P. Candito, G. Failla, R. Livrea, Pairs of positive solutions for a Carrier p(x)-Laplacian type equation, Mathematics 12 (2024), no. 16, Paper No. 2441. 10.3390/math12162441DOI
  4. [4]P. Candito, L. Gasiński, R. Livrea, J. R. Santos Júnior, Multiplicity of positive solutions for a degenerate nonlocal problem with p-Laplacian, Adv. Nonlinear Anal. 11 (2022), no. 1, 357–368. 10.1515/anona-2021-0200DOI
  5. [5]S. Carillo, M. Chipot, On some nonlocal in time and space parabolic problem, Appl. Numer. Math. 208 (2025), 314–322. 10.1016/j.apnum.2024.11.001DOI
  6. [6]G. F. Carrier, A note on the vibrating string, Quart. Appl. Math. 7 (1949), 97–101. 10.1090/qam/28511DOI
  7. [7]G. F. Carrier, On the non-linear vibration problem of the elastic string, Quart. Appl. Math. 3 (1945), 157–165. 10.1090/qam/12351DOI
  8. [8]M. Chipot, B. Lovat, Some remarks on nonlocal elliptic and parabolic problems, Nonlinear Anal. 30 (1997), no. 7, 4619–4627. 10.1016/S0362-546X(97)00169-7DOI
  9. [9]M. Chipot, J. F. Rodrigues, On a class of nonlocal nonlinear elliptic problems, RAIRO Modél. Math. Anal. Numér. 26 (1992), no. 3, 447–467. 10.1051/m2an/1992260304471DOI
  10. [10]Á. Crespo-Blanco, G. Failla, B. Vassallo, Multiple solutions for a nonlocal Neumann p-Laplacian problem, Nonlinear Anal. Real World Appl. 87 (2026), Paper No. 104449. 10.1016/j.nonrwa.2025.104449
  11. [11]M. Delgado, C. Morales-Rodrigo, J. R. Santos Júnior, A. Suárez, Non-local degenerate diffusion coefficients break down the components of positive solutions, Adv. Nonlinear Stud. 20 (2020), no. 1, 19–30. 10.1515/ans-2019-2046DOI
  12. [12]J. I. Díaz, J. E. Saa, Existence et unicité de solutions positives pour certaines équations elliptiques quasilinéaires, C. R. Acad. Sci. Paris Sér. I Math. 305 (1987), no. 12, 521–524.
  13. [13]G. Failla, L. Gasiński, Nonlocal Carrier’s double phase problem, Proc. Roy. Soc. Edinburgh Sect. A, Published online, 1–24 (2025). 10.1017/prm.2025.10099DOI
  14. [14]G. M. Figueiredo, T. S. Figueiredo-Sousa, C. Morales-Rodrigo, A. Suárez, Existence of positive solutions of an elliptic equation with local and nonlocal variable diffusion coefficient, Discrete Contin. Dyn. Syst. Ser. B 24 (2019), no. 8, 3689–3711. 10.3934/dcdsb.2018311DOI
  15. [15]L. Gasiński, J. R. Santos Júnior, Multiplicity of positive solutions for an equation with degenerate nonlocal diffusion, Comput. Math. Appl. 78 (2019), no. 1, 136–143. 10.1016/j.camwa.2019.02.029arxiv.org/abs/1807.01900
  16. [16]L. Gasiński, J. R. Santos Júnior, Nonexistence and multiplicity of positive solutions for an equation with degenerate nonlocal diffusion, Bull. Lond. Math. Soc. 52 (2020), no. 3, 489–497. 10.1112/blms.12342DOI
  17. [17]L. Gasiński, J. R. Santos Júnior, G. Siciliano, Positive solutions for a class of nonlocal problems with possibly singular nonlinearity, J. Fixed Point Theory Appl. 24 (2022), no. 3, Paper No. 65, 15 pp. 10.1007/s11784-022-00982-5arxiv.org/abs/2006.13674
  18. [18]G. M. Lieberman, Boundary regularity for solutions of degenerate elliptic equations, Nonlinear Anal. 12 (1988), no. 11, 1203–1219. 10.1016/0362-546X(88)90053-3DOI
  19. [19]D. Motreanu, V. V. Motreanu, N. S. Papageorgiou, Topological and variational methods with applications to nonlinear boundary value problems. Springer, New York, 2014. 10.1007/978-1-4614-9323-5DOI

Paper details

vlnc:W_nqbcuyby
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

Not recorded for this paper

View statement
Looking for related papers…
Contents