where Ω is a bounded subdomain of ℝN, N ≥ 1 whose boundary is sufficiently regular (e.g. of class 𝒞1) and the heterogeneous terms satisfy $\begin{array}{}
\displaystyle
{a_{ij}},{b_i} \in {\mathscr C}\left( {\overline \Omega } \right)
\end{array}$ for all 1 ≤ i, j ≤ n,
As far as concerns the nonlinear terms, it is assumed that fi ∈ 𝒞1[0, ∞) is a nondecreasing function such that fi(0) = 0, for all 1 ≤ i ≤ n. A function u := (u1,...,un) ∈ [𝒞2+v(Ω)]n, v ∈ (0,1), is a solution of (1) if it satisfies the system of (1) and
These functions will be referred as large solutions. Its study goes back to the pioneering works of J. B. Keller [8] and R. Osserman [20], who considered the case of the single equation Δu = f(u) when f isa monotone function. Since them, many works have dealt with large solutions of elliptic equations (see, e.g. the lists of references of [11]), but almost all of them are focused in the study of the single equation. Some of the few existing references for systems are [6,7,12-14], although, except [12], they only study the special case in which n = 2. Moreover, the majority of the works are restricted to the case in which the nonlinearities are of power-type.
In order to ensure the existence of large solutions of (1) one should ask for the following Keller-Osserman type condition:
(KO) There exists f ∈ 𝒞1 [0, +∞) such that f(u) < fi(u) for every 1 ≤ i ≤ n and, for every a, b > 0,
This condition is a generalization of the one made by P. Álvarez-Caudevilla and J. López-Gómez in [1]. According to S. Dumont et al. [4], by the monotonicity of f we have that
so the second part of the Keller-Osserman condition described in [1] is satisfied. The reader is sent to [11, Chapter 3] and [4] for a detailed discussion on Keller-Ossermann conditions. Essentially, (KO) is a condition on the growth of f at infinity. It is not hard to check that the existence of p > 1 and C > 0 such that
$$\begin{array}{}
\displaystyle
{f_i}(u) \ge C{u^p},\quad 1 \le i \le n,
\end{array}$$
for every u > 0 sufficiently large, entails (KO).
The assumption made on the first inequalities of (2) guarantees that the system of (1) is cooperative, so the maximum principle for cooperative systems of J. López-Gómez and M. Molina-Meyer, [16], which is a fundamental tool for the analysis carried out in this paper, is available. Thus, the usual comparison principle works in our context (see Theorem 2 of Section 2). In particular, we can adapt the construction of a maximal and a minimal large solution given in [1, Sections 3 and 4]. Then, under the general hypotheses of this paper and (KO), there exists a minimal and a maximal solution to (1) for every subdomain Ω ⊂ ℝN with ∂Ω sufficiently regular.
The uniqueness of solutions of (1) is still a widely open question, even when (1) reduces to a single equation, and the usual uniqueness argument for the equation via the blow-up rates does not have a trivial extension to cover (1) (see [14]). Our main result establishes the uniqueness of large solution of (1) when Ω is a star-shaped domain, i.e. when there exists a point x0 ∈ Ω, called the center of Ω, such that the line segment between x0 and x belongs to Ω for every x ∈ Ω. It can be stated as follows.
Theorem 1
Suppose that Ω is star-shaped. Without loss of generality, we can suppose that the center of Ω is the origin; otherwise, the change of variables y = x − x0transforms x0into 0. Let D0be an open neighborhood of ∂Ω with the next property:
There exists ρ0 > 1 such that for every 1 ≤ ρ ≤ ρ0and x ∈ Ω ∩ D0withρx ∈ Ω ∩ D0,
Theorem 1 is a substancial extension of [12, Theorem 1.1] and [15, Theorem 1.1]. Indeed, the main result of [12] establishes the uniqueness of solution for the radially symmetric counterpart of (1) with constant coupling coefficients, aij ∈ ℝ+, 1 ≤ i ≠ j ≤ n, while [15, Theorem 1.1] only deals with (1) in the restricted case n = 1 and a11 ∈ ℝ. On the other hand, the case where Ω is an annular region is covered in [12] and [15].
The hypothesis (5) goes back to [10, Eq. (11)] and [3, Eq. (6)]. It is easily seen that (5) implies (KO), which ensures the existence of positive solutions of (1): Assuming (5), we have that (3) is satisfied for the choice p := min{pi : 1 ≤ i ≤ n} and C := min{fi(1) : 1 ≤ i < n}. Moreover, (5) implies that
$$\begin{array}{}
\displaystyle
{f_i}(u)/u\;\;{\rm{is increasing,}}\quad 1 \le i \le n,
\end{array}$$
because
$$\begin{array}{}
\displaystyle
\frac{{{f_i}(\theta u) - {f_i}(u)}}{{\theta u - u}} \ge \frac{{{\theta ^{{p_i}}}{f_i}(u) - {f_i}(u)}}{{\theta u - u}} = \frac{{{\theta ^{{p_i}}} - 1}}{{\theta - 1}}\frac{{{f_i}(u)}}{u} > \frac{{{f_i}(u)}}{u},\quad 1 \le i \le n,
\end{array}$$
for every θ > 1 and u > 0. The last inequalities entail that $\begin{array}{}
\displaystyle
f_i^\prime (u) \ge {f_i}(u)/u
\end{array}$, and hence, (fi(u)/u)′ ≥ 0 for all 1 ≤ i ≤ n. In the proof of Theorem 1, we will not use (5) directly, but the following equivalent condition.
The assumption (4) is a condition on the growth of the heterogeneous terms, aij, bi, along the rays of Ω as x approximates ∂Ω. More precisely, if for every z ∈ ∂Ω we define the functions
$$\begin{array}{}
\displaystyle
\begin{array}{*{20}{l}}
{a_{ij}^z(t): = {a_{ij}}(tz),}\\
{b_i^z(t): = {b_i}(tz),}
\end{array}\quad t \in [0,1],\;\;1 \le i,j \le n,
\end{array}$$
then, the second line of (4) means that $\begin{array}{}
\displaystyle
b_i^z(t)
\end{array}$ is non-increasing when t ~ 1, for every 1 ≤ i ≤ n. Note that, if the following conditions are satisfied,
for some t0 ∈ (0,1), then the inequalities of the first line of (4) hold. It is remarkable that (4) is weaker than hypothesis (ii) established in [12, Remark 4.2] if we assume Ω is a ball. Indeed, suppose Ω = BR(0) := {x ∈ ℝN : ||x|| < R} and
$$\begin{array}{}
\displaystyle
{a_{ij}}(x): = a_{ij}^*({\rm{dist}}(x,\partial {\rm{\Omega }})),\quad 1 \le i \ne j \le n,
\end{array}$$
for some positive continuous non-increasing functions $\begin{array}{}
\displaystyle
a_{ij}^*
\end{array}$. Then,
for every 1 ≤ i ≠ j ≤ n, ρ > 1 and x ∈ Ω with ρx ∈ Ω.
The distribution of this paper is the following. Section 2 sketches the existence of a minimal and a maximal solution to (1) and provides us with some necessary results for proving the main result. Finally, in Section 3 we show the proof Theorem 1.
Existence and previous results
The existence of a minimal and a maximal solution to (1) can be obtained simply by adapting the abstract results of J. López-Gómez and P. Álvarez-Caudevilla [1, Section 3] to the case of n equations. For our purpose, it is enough if we show a scheme of this construction, with special attention in the construction of a supersolution for the non singular counterpart of (1).
Given a regular subdomain D ⊂ ℝN, we define the operator $\begin{array}{}
\displaystyle
\mathfrak{L}:[\mathscr{C}^{2+\nu}(D)]^n\rightarrow [\mathscr{C}^{\nu}(D)]^n
\end{array}$ by
Thanks to the cooperative structure of $\begin{array}{}
\displaystyle
\mathfrak{L}
\end{array}$, given by (2), it is well known that there exists a unique σ ∈ ℝ such that the linear eigenvalue problem
admits apositive eigenfunction φ ∈ [𝒞2+v(D)]n, φi(x) > 0 for all x ∈ D, 1 ≤ i ≤ n. This value σ ∈ ℝ is called the principal eigenvalue of $\begin{array}{}
\displaystyle
\mathfrak{L}
\end{array}$ under Dirichlet homogeneous boundary conditions. From here on, it will be denoted by $\begin{array}{}
\displaystyle
\sigma [\mathfrak{L},D]
\end{array}$.
The following theorem goes back to M. Molina-Meyer, [17-19]. It can be obtained by adapting the classical theory of sub and supersolution provided in H. Amann [2] and the maximum principle for cooperative systems of J. López-Gómez and M. Molina-Meyer, [16].
Theorem 2
Suppose that (2) and (5) are satisfied. Then, for every w ∈ [𝒞(∂Ω)]n, w ≥ 0, the boundary value problem
Sketch of the proof. By the general assumptions concerning to the nonlinear terms, $\begin{array}{}
\displaystyle
\underline{u}: = \vec 0
\end{array}$ is a (strict) subsolution of (1), for every w > 0. Then, for the existence of a positive solution, it only remains to construct a supersolution of (1).
In the special case when
$$\begin{array}{}
\displaystyle
\mathop {\min }\limits_{z \in \partial {\rm{\Omega }}} {b_i}(z) > 0,\quad 1 \le i \le n,
\end{array}$$
the function ū := (M,M) provides us with a supersolution of (1) for M sufficiently large. Indeed, by (5),
where λ1[−Δ, D] stands for the classical first eigenvalue of −Δ in D under Dirichlet homogeneous boundary conditions. On the other hand, thanks to the Faber-Krahn inequality, [5,9],
for every regular subdomain D ⊂ ℝN such that |D| < δ0.
Set D′ a neighborhood of ∂Ω satisfying (4) and φ = (φ1,...,φn) an eigenfunction associated to $\begin{array}{}
\displaystyle
\sigma \left[ {\mathfrak{L},D'} \right]
\end{array}$, i.e.,
$$\begin{array}{}
\displaystyle
- {\rm{\Delta }}{\varphi _i} - \mathop \sum\limits_{j = 1}^n {a_{ij}}(x){\varphi _j} = \sigma [ - \mathfrak{L} ,D']{\varphi _i} \gt 0,\quad x \in D',\,\,1 \le i \le n,
\end{array}$$
and φi(x) > 0 for every x ∈ D′. Clearly, we can consider another open neighborhood of ∂Ω, namely D*, such that $\begin{array}{}
\displaystyle
{\bar D^*} \subset D'
\end{array}$ and
This finishes the construction of a supersolution. The last assertion of the theorem is due to the uniqueness of the solution of (1), which is a consequence of the maximum principle and (6). □
where $\begin{array}{}
\displaystyle
\vec m: = (m, \ldots ,m)
\end{array}$, is increasing. Moreover, by adapting the construction provided in [11, Chapter 3] for the single equation, we obtain the following result, which in case n = 2 is given by [1, Theorem 4.10].
is well defined, and it provides us with the minimal solution of (1), throughout denoted by$\begin{array}{}
\displaystyle
L_{\rm{\Omega }}^{{\rm{min}}}
\end{array}$. Furthermore, the maximal solution of (1) is given by
Hence, applying Theorem 2 to the solution $\begin{array}{}
\displaystyle
L_\Omega ^{{\rm{max}}}
\end{array}$ and the supersolution ūρ, we obtain that, for every 1 ≤ i ≤ n,
To conclude the proof, note that $\begin{array}{}
\displaystyle
(1 + \varepsilon )L_{\Omega ,i}^{{\rm{min}}}
\end{array}$ is a supersolution of the problem