Two abstract methods of lower and upper solutions with applications

Abstract

In this paper, we present two abstract methods for constructing a lower and an upper solution for a fixed point equation. The first method applies when the nonlinear operator is a composition of a linear and a nonlinear mapping, while the second method applies when the nonlinear operator satisfies an inequality of Harnack type. An application is provided for each method.

Authors

Radu Precup
Faculty of Mathematics and Computer Science and Institute of Advanced Studies in Science and Technology, Babes-Bolyai, University, Cluj-Napoca, Romania
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, Romania

Andrei Stan
Tiberiu Popoviciu Institute of Numerical Analysis, Romanian Academy, Romania
Department of Mathematics, Babes-Bolyai University, Cluj-Napoca, Romania

Keywords

Critical point, Nehari manifold, Birkhoff-Kellogg invariant-direction, cone, p-Laplace operator, positive solution, multiple solutions

Paper coordinates

Precup R, Stan A, Two abstract methods of lower and upper solutions with applications, 2025.Β 

About this paper

Journal
Preprint
Publisher Name
Print ISSN
Online ISSN

google scholar link

Paper (preprint) in HTML form

Two abstract methods of lower and upper solutions with applications

Abstract

In this paper, we present two abstract methods for constructing a lower and an upper solution for a fixed point equation. The first method applies when the nonlinear operator is a composition of a linear and a nonlinear mapping, while the second method applies when the nonlinear operator satisfies an inequality of Harnack type. An application is provided for each method.

keywords:
Upper and lower solution, Harnack inequality, fixed point

1 Introduction and preliminaries

The method of upper and lower solutions proves to be extremely useful for solving nonlinear equations, as it not only guarantees the existence of a solution but also provides a localization of the solution within an interval. The literature on this subject is extensive; we mention a few reference works on the topic [2, 4, 14, 9, 8].

The structure of the paper is as follows. In Section 2, we present two abstract methods for determining lower and upper solutions for an abstract equation, which are then used to guarantee the existence of a fixed point. The first method concerns an abstract Hammerstein equation (see [14] for another approach for the same equation), while the second method applies to a fixed point equation in the case where the nonlinear operator satisfies an abstract Harnack-type inequality. We note that the conditions are inspired by [13], although in that work the method is entirely different, being based on the fixed point index approach. Section 3 is devoted to illustrative applications, each demonstrating one of the two abstract methods.

We conclude this section with two auxiliary results. The first result is a fundamental theorem from the theory of linear operators, which generalizes the classical Perron–Frobenius theorem for matrices ([15], see also [16, p. 266] or [5, Theorem 19.2]).

Theorem 1 (Krein-Rutman).

Let XX be a Banach space, KβŠ‚XK\subset X a total cone, and FF a linear compact operator with F(K)βŠ‚KF(K)\subset K and the spectral radius r(T)r(T) strictly positive. Then, r(T)r(T) is an eigenvalue of FF and the corresponding eigenvector lies in the cone KK.

We continue with a variant of Harnack’s inequality (see, [17, 13]).

Theorem 2.

Let Ξ©βŠ‚β„n\Omega\subset\mathbb{R}^{n} be a domain and DβŠ‚Ξ©D\subset\Omega a compact set. Then, for each p>1p>1 and q∈[1,n(pβˆ’1)nβˆ’p]q\in\left[1,\frac{n(p-1)}{n-p}\right] there exists a constant π’ž>0\mathcal{C}>0 such that

infDuβ‰₯π’ž(∫Duq)1q,\inf_{D}u\geq\mathcal{C}\left(\int_{D}u^{q}\right)^{\frac{1}{q}},

for all uu such that uβ‰₯0u\geq 0 and βˆ’Ξ”puβ‰₯0.-\Delta_{p}u\geq 0.

2 Main abstract results

Let XX be a Banach space with norm |β‹…||\cdot|, ordered by a cone KK. The induced order relation given by the cone KK is denote by ≀\leq, that is, for u,v∈Xu,v\in X we write u≀vu\leq v if and only if vβˆ’u∈Kv-u\in K. Throughout this section, we always assume that the norm |β‹…||\cdot| is semi-monotone, i.e., there exists Ξ³>0\gamma>0 such that

0≀u≀v implies |u|≀γ|v|.0\leq u\leq v\quad\text{ implies }\quad|u|\leq\gamma|v|. (1)

Each of the following two subsections presents a different method for constructing lower and upper solutions for a fixed point equation. The first approach concerns an abstract Hammerstein equation.

2.1 Abstract Hammerstein equations

We consider the equation

u=LF(u),u=LF(u), (2)

where L:X→XL\colon X\to X is a linear operator and F:K→KF\colon K\to K is a (nonlinear) continuous mapping. In the subsequent we denote

N:=LF.N:=LF.

Our first condition is related to the linear operator LL.

(h1)

The cone KK is invariant under the linear operator LL, i.e., L(K)βŠ‚KL(K)\subset K. Moreover, LL admits a positive eigenvalue Ξ»1>0\lambda_{1}>0 with the corresponding eigenfunction Ο†1\varphi_{1} from the cone KK (Ο†1∈K\varphi_{1}\in K).

From (h1) it follows that LL is increasing (order-preserving), that is, for all u,v∈Ku,v\in K, one has

0≀u≀vimpliesLu≀Lv.0\leq u\leq v\quad\text{implies}\quad Lu\leq Lv.

Indeed, if 0≀u≀v0\leq u\leq v, then vβˆ’u∈Kv-u\in K. Since L(vβˆ’u)∈KL(v-u)\in K, i.e., L(vβˆ’u)β‰₯0L(v-u)\geq 0, it follows that Lvβˆ’Luβ‰₯0Lv-Lu\geq 0, which proves the claim.

Under certain conditions on LL and FF, we show that there exists a lower solution uΒ―\underline{u} , i.e.,

u¯≀NuΒ―.\underline{u}\leq N\underline{u}.

and an upper solution

uΒ―β‰₯NuΒ―,\overline{u}\geq N\>\overline{u},

for the equation (2).

Before stating our assumptions, let Ξ¦:K→ℝ+\Phi\colon K\to\mathbb{R}_{+} be a positively homogeneous mapping, that is,

Ξ¦(Ξ±u)=Ξ±Ξ¦(u)for all Ξ±>0,u∈K,\Phi(\alpha u)=\alpha\Phi(u)\quad\text{for all }\alpha>0,\;u\in K,

and with the additional property that Ξ¦(u)=0\Phi(u)=0 if and only if u=0u=0.

Remark 1.

The mapping Ξ¦\Phi can, for instance, to be a seminorm.

The following conditions are assumed to be satisfied:

(h2)

The operator FF is increasing (order-preserving), that is, for u,v∈Ku,v\in K one has

0≀u≀vimpliesF(u)≀F(v).0\leq u\leq v\quad\text{implies}\quad F(u)\leq F(v).
(h3)

There exists a constant r>0r>0 such that

F(Ξ»1u)β‰₯ufor all u∈K with Ξ¦(u)=r.F(\lambda_{1}u)\geq u\quad\text{for all }u\in K\text{ with }\Phi(u)=r.
(h4)

There exists Ξ±>0\alpha>0 and μ∈Kβˆ–{0}\mu\in K\setminus\{0\} such that

Lμ≀αμL\mu\leq\alpha\mu

and

F(Ξ±ΞΌ)≀μ.F(\alpha\mu)\leq\mu.

Denote

uΒ―=rΞ¦(Ο†1)L(Ο†1) and uΒ―=LΞΌ.\underline{u}=\frac{r}{\Phi(\varphi_{1})}L(\varphi_{1})\quad\text{ and }\quad\overline{u}=L\mu. (3)

We now show that uΒ―\underline{u} and uΒ―\overline{u} are a lower and an upper solution, respectively, for the equation (2).

Theorem 3.

Assume that (h1)–(h4) hold true. Then, uΒ―\underline{u} is a lower solution and uΒ―\overline{u} is an upper solution for the equation (2).

Proof.

To show that uΒ―\underline{u} is a lower solution, first observe that since

Φ(rΦ(φ1)φ1)=rΦ(φ1)Φ(φ1)=r,\Phi\left(\frac{r}{\Phi(\varphi_{1})}\varphi_{1}\right)=\frac{r}{\Phi(\varphi_{1})}\Phi\left(\varphi_{1}\right)=r,

and Ο†1∈Kβˆ–{0}\varphi_{1}\in K\setminus\{0\}, condition (h3) implies

rΞ¦(Ο†1)Ο†1≀T(Ξ»1rΞ¦(Ο†1)Ο†1)=T(rΞ¦(Ο†1)L(Ο†1))=F(uΒ―).\frac{r}{\Phi(\varphi_{1})}\varphi_{1}\leq T\!\left(\frac{\lambda_{1}r}{\Phi(\varphi_{1})}\varphi_{1}\right)=T\!\left(\frac{r}{\Phi(\varphi_{1})}L\!\left(\varphi_{1}\right)\right)=F(\underline{u}).

By the order-preserving property of LL, it follows that

uΒ―=L(rΞ¦(Ο†1)Ο†1)≀LF(uΒ―)=N(uΒ―),\underline{u}=L\!\left(\frac{r}{\Phi(\varphi_{1})}\varphi_{1}\right)\leq LF(\underline{u})=N(\underline{u}),

as desired.

Concerning the function uΒ―,\overline{u}, using the monotonicity property of FF from (h2), together with (h4), one obtains

ΞΌβ‰₯F(Ξ±ΞΌ)β‰₯F(LΞΌ)=F(uΒ―).\mu\geq F(\alpha\mu)\geq F(L\mu)=F(\overline{u}).

Applying again the order-preserving property of LL to this relation yields

uΒ―=LΞΌβ‰₯N(uΒ―),\overline{u}=L\mu\geq N(\overline{u}),

hence u¯\overline{u} is an upper solution. ∎

Further, assume that

(h5)

The lower and upper solutions uΒ―\underline{u} and uΒ―\overline{u} satisfy (are comparable)

u¯≀uΒ―.\underline{u}\leq\overline{u}.

Then the following invariance result holds.

Theorem 4.

Under conditions (h1)-(h5), the interval [uΒ―,uΒ―][\underline{u},\overline{u}] is invariant under the operator NN, that is,

N([uΒ―,uΒ―])βŠ‚[uΒ―,uΒ―].N\big([\underline{u},\overline{u}]\big)\;\subset\;[\underline{u},\overline{u}].
Proof.

Let u∈[u¯,u¯]u\in[\underline{u},\overline{u}]. By the monotonicity of LL and FF, we have

u¯≀u≀uΒ― implies N(uΒ―)=LF(uΒ―)≀LF(u)≀LF(uΒ―)=N(uΒ―).\underline{u}\leq u\leq\overline{u}\quad\text{ implies }\quad N(\underline{u})=LF(\underline{u})\leq LF(u)\leq LF(\overline{u})=N(\overline{u}).

Since u¯≀N(uΒ―)\underline{u}\leq N(\underline{u}) and uΒ―β‰₯N(uΒ―)\overline{u}\geq N(\overline{u}), it follows that

u¯≀Nu≀uΒ―,\underline{u}\leq Nu\leq\overline{u},

which completes our proof. ∎

Remark 2.

One easily sees that condition (h5) holds if

rΞ¦(Ο†1)Ο†1≀μ,\frac{r}{\Phi(\varphi_{1})}\varphi_{1}\leq\mu,

which proves to be useful in applications, as we are about to see in the next section.

Under an additional compactness condition on LL, we obtain the following fixed-point result.

Theorem 5.

Assume that conditions (h1)-(h5) are satisfied. If, in addition, the operator LL is completely continuous, then there exists uβˆ—βˆˆ[uΒ―,uΒ―]u^{\ast}\in[\underline{u},\overline{u}] such that

uβˆ—=N(uβˆ—).u^{\ast}=N(u^{\ast}).
Proof.

Clearly, the set [uΒ―,uΒ―][\underline{u},\overline{u}] is convex. Since the cone KK is closed, it follows that [uΒ―,uΒ―][\underline{u},\overline{u}] is also closed. Moreover, as the norm |β‹…||\cdot| is semi-monotone, for all u∈[uΒ―,uΒ―]u\in[\underline{u},\overline{u}] one has

|u|≀γ|uΒ―|,|u|\leq\gamma|\overline{u}|,

and hence the set [uΒ―,uΒ―][\underline{u},\overline{u}] is bounded.

The complete continuity of LL, together with the continuity and monotonicity of FF, implies that the operator NN is continuous and maps the set [uΒ―,uΒ―][\underline{u},\overline{u}] into a relatively compact set. Indeed, since FF is increasing and the norm is semi-monotone, one obtains that F([uΒ―,uΒ―])F([\underline{u},\overline{u}]) is bounded, more exactly, |F(u)|≀γ|F(uΒ―)||F(u)|\leq\gamma\,|F(\overline{u})| for all u∈[uΒ―,uΒ―]u\in[\underline{u},\overline{u}]. Thus, by the compactness of LL, it follows that N([uΒ―,uΒ―])N([\underline{u},\overline{u}]) is relatively compact, which establishes the desired property.

Finally, by Theorem 4, the interval [uΒ―,uΒ―][\underline{u},\overline{u}] is invariant under the operator NN. Hence, Schauder’s fixed point theorem applies and ensures the existence of a fixed point uβˆ—βˆˆ[uΒ―,uΒ―]u^{\ast}\in[\underline{u},\overline{u}] for NN, which finishes our proof.

∎

Remark 3.

From the proof of Theorem 3, we observe that instead of Ο†1\varphi_{1} and Ξ»1\lambda_{1}, one may take any nonzero element Ο†βˆˆKβˆ–{0}\varphi\in K\setminus\{0\} and any Ξ»>0\lambda>0 such that

L(Ο†)β‰₯λφ.L(\varphi)\geq\lambda\varphi.

In this case, Ξ»1\lambda_{1} in (h3) should be replaced by Ξ»\lambda, and uΒ―\underline{u} to be defined by

u¯=rΦ(φ)L(φ).\underline{u}=\frac{r}{\Phi(\varphi)}\,L(\varphi). (4)

Indeed, since

rΞ¦(Ο†)φ≀T(Ξ»rΞ¦(Ο†)Ο†)≀T(rΞ¦(Ο†)L(Ο†))=F(uΒ―),\frac{r}{\Phi(\varphi)}\varphi\leq T\!\left(\frac{\lambda r}{\Phi(\varphi)}\varphi\right)\leq T\!\left(\frac{r}{\Phi(\varphi)}L(\varphi)\right)=F(\underline{u}),

the order-preserving property of LL implies that uΒ―\underline{u} given in (4) is a lower solution for the equation (2).

If, instead of conditions (h3) and (h4), we consider

(h3)’

There exists R>0R>0 such that

F(Ξ»1u)≀ufor all u∈K with Ξ¦(u)=R,F(\lambda_{1}u)\leq u\quad\text{for all }u\in K\text{ with }\Phi(u)=R,
(h4)’

There exist Ξ±>0\alpha>0 and μ∈Kβˆ–{0}\mu\in K\setminus\{0\} such that

LΞΌβ‰₯Ξ±ΞΌL\mu\geq\alpha\mu

and

F(Ξ±ΞΌ)β‰₯ΞΌ,F(\alpha\mu)\geq\mu,

an analogue of Theorem 3 can be established.

Theorem 6.

Assume that conditions (h1), (h2), (h3)’ and (h4)’ are satisfied. Then

u¯=Lμandu¯=RΦ(φ1)L(φ1),\underline{u}=L\mu\quad\text{and}\quad\overline{u}=\frac{R}{\Phi(\varphi_{1})}\,L(\varphi_{1}),

are a lower and an upper solution, respectively, for the equation (2).

Proof.

Under similar reasoning as in the proof of Theorem 3, we obtain

uΒ―=Lμ≀LF(Ξ±ΞΌ)≀LF(LΞΌ)=N(uΒ―),\underline{u}=L\mu\leq LF(\alpha\mu)\leq LF\left(L\mu\right)=N(\underline{u}),

and

uΒ―=L(R|Ο†1|Ο†1)β‰₯LF(Ξ»1RΞ¦(Ο†1)Ο†1)=LF(RΞ¦(Ο†1)L(Ο†1))=N(uΒ―).\overline{u}=L\left(\frac{R}{|\varphi_{1}|}\varphi_{1}\right)\geq LF\!\left(\frac{\lambda_{1}R}{\Phi(\varphi_{1})}\varphi_{1}\right)=LF\left(\frac{R}{\Phi(\varphi_{1})}L(\varphi_{1})\right)=N(\overline{u}).

∎

Remark 4.

We emphasize that, under the same assumptions as in Theorem 5, the method of monotone iterations can be applied to obtain two extremal (not necessarily distinct) fixed points for the operator NN. That is, there exist uΒ―βˆ—,uΒ―βˆ—βˆˆ[uΒ―,uΒ―]\underline{u}^{\ast},\overline{u}^{\ast}\in[\underline{u},\overline{u}] such that N(uΒ―βˆ—)=uΒ―βˆ—N(\underline{u}^{\ast})=\underline{u}^{\ast} and N(uΒ―βˆ—)=uΒ―βˆ—N(\overline{u}^{\ast})=\overline{u}^{\ast}, and every other fixed point of NN from [uΒ―,uΒ―][\underline{u},\overline{u}] lies in the interval [uΒ―βˆ—,uΒ―βˆ—][\underline{u}^{\ast},\overline{u}^{\ast}]. We refer the reader to [4] or [12] for further details on the monotone iterative method.

In the next section, we present another method for constructing a lower and an upper solution for a fixed point equation, applicable when the nonlinear operator satisfies an abstract Harnack inequality.

2.2 Fixed point equations via abstract Harnack inequality

We consider the fixed point problem

u=N(u),u=N(u), (5)

where N:Xβ†’XN\colon X\to X is a (nonlinear) increasing operator, i.e., for u,v∈Ku,v\in K one has

0≀u≀vimpliesN(u)≀N(v).0\leq u\leq v\quad\text{implies}\quad N(u)\leq N(v).

On XX, we consider a seminorm βˆ₯β‹…βˆ₯\|\cdot\|, which is assumed to be increasing, i.e., for u,v∈Ku,v\in K one has

0≀u≀vimpliesβ€–u‖≀‖vβ€–.0\leq u\leq v\quad\text{implies}\quad\|u\|\leq\|v\|.

Moreover, we assume that there exists ψ∈Kβˆ–{0}\psi\in K\setminus\{0\} such that

u≀|u|ψfor all u∈K.u\leq|u|\,\psi\quad\text{for all }\,\,u\in K. (6)

The following condition plays a key role in the subsequent analysis and can be regarded as a weak type Harnack inequality [13].

(a1)

There exists Ο‡βˆˆKβˆ–{0}\chi\in K\setminus\{0\} such that

N(u)β‰₯β€–N(u)β€–Ο‡for all u∈K.N(u)\geq\|N(u)\|\chi\quad\text{for all }u\in K. (7)

The next two additional conditions are required.

(a2)

There exists r>0r>0 such that

β€–N(rΟ‡)β€–β‰₯r.\|N(r\chi)\|\geq r.
(a3)

There exists R>0R>0 such that

|N(Rψ)|≀R.|N(R\psi)|\leq R.

Now, we are ready to present the main result of this subsection.

Theorem 7.

Assume that conditions (a1)-(a3) are satisfied. Then

uΒ―=rΟ‡anduΒ―=Rψ,\underline{u}=r\chi\quad\text{and}\quad\overline{u}=R\psi,

represent a lower and an upper solution, respectively, for the equation (5). Moreover, if

u¯≀uΒ―,\underline{u}\leq\overline{u}, (8)

and the operator NN maps the interval [uΒ―,uΒ―][\underline{u},\overline{u}] into a relatively compact set, then NN admits a fixed point in the interval [uΒ―,uΒ―][\underline{u},\overline{u}], i.e., there exists uβˆ—βˆˆKu^{\ast}\in K such that

u¯≀uβˆ—β‰€uΒ―andN(uβˆ—)=uβˆ—.\underline{u}\leq u^{\ast}\leq\overline{u}\quad\text{and}\quad N(u^{\ast})=u^{\ast}. (9)
Proof.

Note that, since the seminorm βˆ₯β‹…βˆ₯\|\cdot\| is increasing, and using the Harnack inequality (7) together with condition (a2), we obtain

N(uΒ―)=N(rΟ‡)β‰₯β€–N(rΟ‡)β€–Ο‡β‰₯rΟ‡=uΒ―,N(\underline{u})=N(r\chi)\geq\|N(r\chi)\|\chi\geq r\chi=\underline{u}, (10)

so uΒ―\underline{u} is a lower solution of problem (5). Moreover, for any uβ‰₯uΒ―u\geq\underline{u}, by the monotonicity of NN and relation (10), we have

N(u)β‰₯N(uΒ―)β‰₯uΒ―.N(u)\geq N(\underline{u})\geq\underline{u}. (11)

Taking u=N(Rψ)u=N(R\psi) in (6), we have

N(uΒ―)=N(Rψ)≀|N(Rψ)|ψ=|N(uΒ―)|ψ.N(\overline{u})=N(R\psi)\leq|N(R\psi)|\psi=|N(\overline{u})|\psi.

Thus, by (a3), one obtains

N(uΒ―)≀|N(uΒ―)|ψ=|N(Rψ)|Οˆβ‰€Rψ=uΒ―,N(\overline{u})\leq|N(\overline{u})|\psi=|N(R\psi)|\psi\leq R\psi=\overline{u},

whence uΒ―\overline{u} is an upper solution for the problem (5). For any u≀uΒ―u\leq\overline{u}, by the monotonicity of NN, we further obtain that

N(u)≀N(uΒ―)≀uΒ―.N(u)\leq N(\overline{u})\leq\overline{u}. (12)

Assume now that relation (8) holds. Then, from (11) and (12) it follows immediately that

N([uΒ―,uΒ―])βŠ‚[uΒ―,uΒ―].N\big([\underline{u},\overline{u}]\big)\subset[\underline{u},\overline{u}].

Moreover, the set [uΒ―,uΒ―][\underline{u},\overline{u}] is convex, closed, and bounded (the boundedness follows from the semi-monotonicity of the norm). Now, if NN maps [uΒ―,uΒ―][\underline{u},\overline{u}] into a relatively compact set, then the Schauder’s fixed point theorem applies and guarantees that there exists a fixed point for NN in the interval [uΒ―,uΒ―][\underline{u},\overline{u}], that is, relation (9) holds. ∎

Remark 5.

By the monotonicity of the seminorm βˆ₯β‹…βˆ₯\|\cdot\| and the Harnack inequality (7), we have ‖χ‖≀1\|\chi\|\leq 1. Consequently, compared to condition (2.18) in [13], assumption (a2) represents a weaker requirement. Indeed, if condition (2.18) in [13] holds, that is,

β€–N(rΟ‡)β€–β‰₯rβ€–Ο‡β€–,\|N(r\chi)\|\geq\frac{r}{\|\chi\|},

then

β€–N(rΟ‡)β€–β‰₯rβ€–Ο‡β€–β‰₯r,\|N(r\chi)\|\geq\frac{r}{\|\chi\|}\geq r,

and hence assumption (a2) is also satisfied.

3 Applications

In this section, we present one application for each of the two abstract results from Section 2.

3.1 Positive solution of the classical Hammerstein equation

In this subsection, we consider the fixed point problem

u(t)=∫01k(t,s)f(u(s))𝑑s,u(t)=\int_{0}^{1}k(t,s)f(u(s))\,ds, (13)

for which we show how Theorem 5 can be applied. Here, kk is a symmetric (k(t,s)=k(s,t)k(t,s)=k(s,t) for all t,s∈[0,1]t,s\in[0,1]), nonnegative continuous function on [0,1]2[0,1]^{2}, and f∈C(ℝ,ℝ+)f\in C(\mathbb{R},\mathbb{R}_{+}) is nondecreasing on ℝ+\mathbb{R}_{+}.

Let X=C[0,1]X=C[0,1] be endowed with the supremum norm |β‹…|∞|\cdot|_{\infty}, and let KK denote the cone of continuous nonnegative functions. Also, let L:Xβ†’XL\colon X\to X be the linear Hammerstein operator

(Lu)(t)=∫01k(t,s)u(s)𝑑s,u∈C[0,1],(Lu)(t)=\int_{0}^{1}k(t,s)u(s)\,ds,\quad u\in C[0,1],

and F:Kβ†’KF\colon K\to K the Nemytskii operator, which assigns to each u∈Ku\in K the function Fu:[0,1]→ℝ+Fu\colon[0,1]\to\mathbb{R}_{+} given by

F(u)(s)=f(u(s)),s∈[0,1].F(u)(s)=f(u(s)),\quad s\in[0,1].

Note that FF is well defined since ff is continuous and nonnegative on ℝ\mathbb{R}.

The complete continuity of LL follows from standard arguments based on the Arzelà–Ascoli theorem (see, e.g., [7, 10, 3, 11]). To guarantee that LL has an eigenvalue, we need the following additional condition on the kernel kk (see [3]).

(H1)

The function kk satisfies a Green like inequality, that is, there exists a continuous function ΞΈ:[0,1]→ℝ+\theta\colon[0,1]\to\mathbb{R}_{+} such that

k(t,s)β‰₯ΞΈ(t)k(q,s)for allt,s,q∈[0,1],k(t,s)\geq\theta(t)k(q,s)\quad\text{for all}\quad t,s,q\in[0,1],

and

|Lθ|∞>0.|L\theta|_{\infty}>0.

The following result is of great importance for the subsequent analysis.

Lemma 1.

If condition (H1) is satisfied, then the linear Hammerstein operator LL has a positive eigenvalue Ξ»1>0\lambda_{1}>0, and moreover, the corresponding eigenfunction Ο†1\varphi_{1} is positive, i.e., Ο†1∈K\varphi_{1}\in K.

Proof.

The proof relies on the Krein–Rutman theorem (Theorem 1). If one can show that the spectral radius r(L)r(L) of LL is strictly positive, then, since LL is completely continuous and the cone KK is reproducing (see [5, Chapter 19]), and hence total, Theorem 1 applies, and the conclusion follows.

Now we show that r(L)>0r(L)>0. To this end, we use Gelfand’s formula (see, e.g., [5, p. 79])

r(L)=limkβ†’βˆž|Lk|op1/k,r(L)=\lim_{k\to\infty}\left|L^{k}\right|_{\text{op}}^{1/k},

where |β‹…|op|\cdot|_{\text{op}} denotes the operator norm, i.e.,

|L|op=supuβ‰ 0|Lu|∞|u|∞.|L|_{\text{op}}=\sup_{u\neq 0}\frac{|Lu|_{\infty}}{|u|_{\infty}}.

Simple computations yields that

(LΞΈ)(t)=∫01k(t,s)ΞΈ(s)𝑑sβ‰₯ΞΈ(t)∫01k(q,s)ΞΈ(s)𝑑s=ΞΈ(t)(LΞΈ)(q),(L\theta)(t)=\int_{0}^{1}k(t,s)\theta(s)\,ds\geq\theta(t)\int_{0}^{1}k(q,s)\theta(s)\,ds=\theta(t)\,(L\theta)(q),

for all q∈[0,1]q\in[0,1]. Hence,

(LΞΈ)(t)β‰₯ΞΈ(t)|LΞΈ|∞for all t∈[0,1].(L\theta)(t)\geq\theta(t)\,|L\theta|_{\infty}\quad\text{for all }t\in[0,1].

Thus, for any kβˆˆβ„•k\in\mathbb{N}, one has

(LkΞΈ)(t)β‰₯ΞΈ(t)|LΞΈ|∞k.(L^{k}\theta)(t)\geq\theta(t)\,|L\theta|_{\infty}^{k}.

Taking the supremum norm, we obtain

|LkΞΈ|∞β‰₯|ΞΈ|∞|LΞΈ|∞k.|L^{k}\theta|_{\infty}\geq|\theta|_{\infty}\,|L\theta|_{\infty}^{k}.

Consequently, since

|Lk|opβ‰₯|LkΞΈ|∞|ΞΈ|∞β‰₯|LΞΈ|∞k,|L^{k}|_{\text{op}}\geq\frac{|L^{k}\theta|_{\infty}}{|\theta|_{\infty}}\geq|L\theta|_{\infty}^{k},

it follows that

r(L)β‰₯|LΞΈ|∞>0,r(L)\geq|L\theta|_{\infty}>0,

where the latter inequality follows by condition (H1). ∎

Remark 6.

Other conditions than (H1) can ensure that the conclusion of Lemma 1 remains valid, for instance, mint∈[0,1]⁑k(t,t)>0\min_{t\in[0,1]}k(t,t)>0 (see [11, Lemma 1]).

By Lemma 1, the operator LL has a positive eigenvalue Ξ»1=r(L)>0\lambda_{1}=r(L)>0 with the corresponding eigenfunction Ο†1∈K\varphi_{1}\in K, hence condition (h1) is fulfilled. Moreover, since the function ff is continuous and nondecreasing, the Nemytskii operator FF is continuous and increasing, therefore, condition (h2) is also satisfied.

To ensure that (h3) and (h4) are valid as well, the following asymptotic conditions on ff are required:

(H2)

One has,

limtβ†˜0f(t)t>1Ξ»1 and limtβ†’βˆžf(t)t<1|L1|∞,\lim_{t\searrow 0}\frac{f(t)}{t}>\frac{1}{\lambda_{1}}\quad\text{ and }\quad\lim_{t\to\infty}\frac{f(t)}{t}<\frac{1}{|L1|_{\infty}},

where

(L1)(t)=∫01k(t,s)𝑑s.(L1)(t)=\int_{0}^{1}k(t,s)ds.

Note that, by (H2), there exists 0<r<R0<r<R such that

f(t)β‰₯tΞ»1 for allt∈[0,Ξ»1r],f(t)\geq\frac{t}{\lambda_{1}}\quad\text{ for all}\,\,t\in[0,\lambda_{1}r], (14)

and

f(t)≀t|L1|∞ for alltβ‰₯|L1|∞R.f(t)\leq\frac{t}{|L1|_{\infty}}\quad\text{ for all}\,\,t\geq|L1|_{\infty}R. (15)

Now, the following result holds.

Theorem 8.

Under conditions (H1) and (H2), there exists a positive solution uβˆ—βˆˆKu^{\ast}\in K for the problem (13), and moreover,

r|Ο†1|∞(LΟ†1)(t)≀uβˆ—(t)≀(LR)(t) for allt∈[0,1].\frac{r}{|\varphi_{1}|_{\infty}}(L\varphi_{1})(t)\leq u^{\ast}(t)\leq(LR)(t)\quad\text{ for all}\,\,t\in[0,1].
Proof.

We verify that all the assumptions of Theorem 5 are satisfied. Conditions (h1) and (h2) are valid, as explained above.

Check of condition (h3). Letting Ξ¦(u)=|u|∞\Phi(u)=|u|_{\infty}, we see that for any u∈Ku\in K with Ξ¦(u)=r\Phi(u)=r, one has Ξ»1u(t)≀λ1r\lambda_{1}u(t)\leq\lambda_{1}r (t∈[0,1]t\in[0,1]). Thus, using (14), we deduce that

f(Ξ»1u(t))≀λ1u(t)Ξ»1=u(t) for all t∈[0,1],f(\lambda_{1}u(t))\leq\frac{\lambda_{1}u(t)}{\lambda_{1}}=u(t)\quad\text{ for all }\,\,t\in[0,1],

so condition (h2) is verified.

Check of condition (h4). Let μ≑R\mu\equiv R. One has,

(LΞΌ)(t)=(LR)(t)=R(L1)(t)≀αR(t∈[0,1]),(L\mu)(t)=(LR)(t)=R(L_{1})(t)\leq\alpha R\quad(t\in[0,1]),

where α=|L1|∞\alpha=|L1|_{\infty}, and

F(Ξ±ΞΌ)=f(R|L1|∞)≀R|L1|∞|L1|∞=ΞΌ,F(\alpha\mu)=f(R|L1|_{\infty})\leq\frac{R|L1|_{\infty}}{|L1|_{\infty}}=\mu,

whence condition (h4) holds.

Check of condition (h5). Based on Remark 2, condition (h5) is satisfied if

r|Ο†|βˆžΟ†1(t)≀μ=R,\frac{r}{|\varphi|_{\infty}}\varphi_{1}(t)\leq\mu=R,

which is clearly true since r≀Rr\leq R.

Therefore, Theorem 5 applies and gives the conclusion. ∎

3.2 Positive solutions of pp-Laplace equations

In this subsection, inspired by [13], we apply Theorem 7 for the pp-Laplace problem

{βˆ’Ξ”pu=f(u)in Ξ©u=0on βˆ‚Ξ©u>0,\begin{cases}-\Delta_{p}u=f(u)\quad\text{in }\,\,\Omega\\ u=0\quad\text{on }\,\,\partial\Omega\\ u>0,\end{cases} (16)

where p>1p>1, Ξ©\Omega is a smooth domain in ℝn\mathbb{R}^{n} and f:ℝ+→ℝ+f\colon\mathbb{R}_{+}\to\mathbb{R}_{+} is a nondecreasing continuous function. Following [1, Lemma 1.1] (see also [6]), for each h∈L∞(Ξ©)h\in L^{\infty}(\Omega), there exists a unique (weak) solution S(h)∈C01(Ω¯)S(h)\in C_{0}^{1}(\overline{\Omega}) to the problem

{βˆ’Ξ”pu=h in Ξ©,u∈W01,p(Ξ©).\begin{cases}-\Delta_{p}u=h\quad\text{ in }\Omega,\\[4.0pt] u\in W^{1,p}_{0}(\Omega).\end{cases}

Moreover, the operator S:L∞(Ξ©)β†’C01(Ω¯)S\colon L^{\infty}(\Omega)\to C_{0}^{1}(\overline{\Omega}) is completely continuous, order preserving (increasing), and invariant with respect to the cone of positive functions, i.e., S(h)β‰₯0S(h)\geq 0 whenever hβ‰₯0h\geq 0. Thus, letting X=L∞(Ξ©)X=L^{\infty}(\Omega) be endowed with the supremum norm |β‹…|∞|\cdot|_{\infty} (hence increasing), and denoting by KK the cone of positive functions, the problem (16) allows for the fixed point formulation

u=N(u),u∈K,u=N(u),\quad u\in K, (17)

where N=SFN=SF, and FF is the Nemytskii operator that associates to each function u∈L∞(Ω)u\in L^{\infty}(\Omega) the function

F(u)(x)=f(u(x)),x∈Ω.F(u)(x)=f(u(x)),\quad x\in\Omega.

Note that, since ff is continuous and nonnegative, the operator FF is continuous and F(K)βŠ‚KF(K)\subset K.

We immediately observe that relation (6) holds with Οˆβ‰‘1\psi\equiv 1. Moreover, given that the function ff is nondecreasing, the operator TT is increasing, and because SS is order-preserving, it follows that NN is also increasing.

By Theorem 2, for some fixed compact set DβŠ‚Ξ©D\subset\Omega, there exists M>0M>0 such that for every pp-superharmonic function u∈Ku\in K, one has

infDuβ‰₯M∫Du(x)𝑑x.\inf_{D}u\geq M\int_{D}u(x)dx.

Let us consider the seminorm βˆ₯β‹…βˆ₯\|\cdot\| be given by

β€–uβ€–=M∫Du(x)𝑑x,u∈L∞(Ξ©).\|u\|=M\int_{D}u(x)\,dx,\quad u\in L^{\infty}(\Omega).

Then, since S(h)S(h) is nonnegative and pp-superharmonic for each h∈Kh\in K, condition (a1) (inequality  (7)) holds with

Ο‡(x)={1,x∈D,0,otherwise.\chi(x)=\begin{cases}1,&x\in D,\\[4.0pt] 0,&\text{otherwise}.\end{cases}

Note for any q>0q>0, one has (see [13])

Sq=q1pβˆ’1S1.Sq=q^{\frac{1}{p-1}}S1.

Thus, it follows that

β€–N(qΟ‡)β€–=β€–SF(qΟ‡)β€–=f(q)1pβˆ’1β€–S1β€–\|N(q\chi)\|=\|SF(q\chi)\|=f(q)^{\frac{1}{p-1}}\|S1\|

and

|N(qψ)|∞=|N(q)|βˆžβ‰€f(q)1pβˆ’1|S1|∞.|N(q\psi)|_{\infty}=|N(q)|_{\infty}\leq f(q)^{\frac{1}{p-1}}|S1|_{\infty}.

If

(A1)

There exists 0<r≀R0<r\leq R such that

f(r)β‰₯rpβˆ’1β€–S1β€–pβˆ’1,f(r)\geq\frac{r^{p-1}}{\|S1\|^{p-1}},

and

f(R)≀rpβˆ’1|S1|∞pβˆ’1,f(R)\leq\frac{r^{p-1}}{|S1|_{\infty}^{p-1}},

then conditions (a2) and (a3) are verified. In addition, since r≀Rr\leq R then relation (8) is also valid, where

uΒ―:=rΟ‡ and uΒ―=R.\underline{u}:=r\chi\quad\text{ and }\quad\overline{u}=R. (18)

We easily see that, since FF is continuous, the operator NN is also continuous. Moreover, as FF is increasing, the set F([uΒ―,uΒ―])F([\underline{u},\overline{u}]) is bounded. Hence, the complete continuity of SS implies that the set N([uΒ―,uΒ―])N([\underline{u},\overline{u}]) is relatively compact.

Therefore, all the requirements of Theorem 7 are satisfied, so the following result holds.

Theorem 9.

Assume that condition (A1) is satisfied. Then, uΒ―\underline{u} and uΒ―\overline{u} given in (18) represent a lower and upper solution, respectively, for the equation (17). Moreover, there exists uβˆ—βˆˆKu^{\ast}\in K such that it is a solution for the problem (16) and

rχ≀uβˆ—β‰€R.r\chi\leq u^{\ast}\leq R.
\bmhead

Acknowledgements The author would like to thank Prof. Radu Precup for insightful discussions on the subject and verification of the paper.

Declarations

Funding

The authors declare that no funds, grants, or other support were received during the preparation of this manuscript.

Conflict of interest

The author have no relevant financial or non-financial interests to disclose.

Data availability

Not applicable

Ethics approval and consent to participate

Not applicable

Materials availability

Not applicable

Code availability

Not applicable

Related Posts

No results found.