A priori estimates and multiplicity for systems of elliptic PDE with natural gradient growth (Q2309171)

The authors investigate fully nonlinear uniformly elliptic cooperative systems \[ \!\!\!\!\!\begin{cases} -F_i(x,Du_i,D^2u_i) - \langle M_i(x)Du_i,Du_i\rangle \!=\! \lambda\displaystyle\sum\limits^n_{j=1} c_{ij}(x)u_j + h_i(x) & \!\mbox{in}\;\Omega \\ u_1 = \cdots = u_n = 0 &\!\mbox{on}\; \partial\Omega\end{cases} \eqno{(P_\lambda)} \] where $\Omega$ is a bounded $C^{1,1}$ domain in $\mathbb{R}^N$, $\lambda \in \mathbb{R}\), \(n, N \ge 1$, $c_{ij}, h_i\in L^\infty(\Omega)$, $c_{ij}\ge 0$, $M_i$ is a bounded nondegenerate matrix and $F_i$ is an uniformly elliptic Isaacs operator. The authors assume that the matrices $M_i$ satisfy the nondegeneracy condition \[ \mu_1 I \le M_i(x) \le \mu_2 I \quad \mbox{a.e. in}\;\Omega \eqno{(M)} \] for some $\mu_1, \mu_2 > 0$, and that $F_i$ in $(P_\lambda)$ has the following structure \[ \!\begin{cases} F_i(x, 0,X) \; \; \mbox{is continuous in}\;\; x\in \overline{\Omega}, \\ \mathcal M^-(X\! -\! Y ) - b|\vec{p} - \vec{q}| \le F_i(x, \vec{p}, X)-F_i(x, \vec{q}, Y )\le \mathcal M^+(X\! - \!Y ) + b|\vec{p} - \vec{q}| \end{cases} \eqno{(SC)} \] for a.e. $x\in \Omega$, where $b \ge 0$ and $\mathcal M^-,\mathcal M^+$ are the Pucci extremal operators with constants $0 < \lambda_P \le \Lambda_P$. Solutions of the Dirichlet problem $(P_\lambda)$ are understood in the $L^p$-viscosity sense and the authors also use the notion of strong solutions, which are functions in $W^{2,p}_{\mathrm{loc}} (\Omega)$ satisfying the equation almost everywhere. The authors denote $F[u] := (F_1[u_1], \dots , F_n[u_n] )$, $u = (u_1, \dots , u_n)$, $f = (f_1, \dots , f_n)$, fix $p > N$, and consider the Dirichlet problem $$ -F[u] = f(x) \quad \mbox{in}\;\Omega,\quad u = 0 \quad\mbox{on} \;\partial\Omega \eqno{(2)} $$ and suppose $$ \!\!\mbox{For each }f\in L^p(\Omega)^n,\text{ there exists a unique }L^p\text{-viscosity solution of (2).}\!\!\! \eqno{(H_1)} $$ $$ \mbox{For each }f\in L^p(\Omega)^n,\text{ any solution }u\text{ of (2) belongs to }W^{2,p}(\Omega)^n. \eqno{(H_2)} $$ $$ \mbox{The problem }(P_0)\text{ has a strong solution }u_0 = (u^0_1, \dots, u^0_n). \eqno{(H_0)} $$ The authors also assume that $C$ in $(P_\lambda)$ is in the block triangular form $$ C(x) = (C_{kl}(x))^{n'}_{k,l=1}, \eqno{(3)} $$ where $1 \le n' \le n$, $C_{kl}$ are $t_k\times t_l$ matrices, $\sum^{n'}_{k=1} t_k = n$, $C_{kk}$ is irreducible for each $k = 1, \dots n'$, and $C_{kl} \equiv 0$ in $\Omega$, for all $k, l\in \{1, \dots n'\}$ with $k < l$ and $$ \begin{matrix} \mbox{in (3) there is no }1\times 1\text{ block with a zero coefficient,}\cr \qquad\qquad\qquad\qquad\qquad\qquad\mbox{ i.e., if }t_k = 1,\text{ then }C_{kk} \not\equiv 0. \end{matrix} \eqno{(H3)} $$ The main results of this article are the following theorems. Theorem 2.2. Suppose that (M), (SC), (H3) hold. Let $\Lambda_1, \Lambda_2$ with $0 < \Lambda_1 < \Lambda_2$. Then every $L^p$-viscosity solution $(u_1,\dots, u_n)$ of $(P_\lambda)$ satisfies $$ \|u_i\|_\infty \le C, \; \mbox{for all} \; \lambda\in [ \Lambda_1,\Lambda_2], \quad i = 1, \dots , n, $$ where $C$ depends on $n, N, p, \mu_1,\mu_2$, diam($\Omega$), $\Lambda_1, \Lambda_2,$ $\|b\|_\infty , \|c_{ij}\|_\infty,$ $\|h_i\|_\infty, \lambda_p,$ $ \Lambda_p$, and on a lower bound on the measure of the sets where the $c_{ij}$ are positive, for those $i, j$ which determine the irreducibility of the blocks in the form (3). The next theorems describe the solution set of $(P_\lambda)$. Theorem 2.3. Assume (M), (SC), (H0), (H1), and (H3). 1. Then, for $\lambda \le 0$, the problem $(P_\lambda)$ has an $L^p$-viscosity solution $u_\lambda$ that converges to $u_0$ in $E$ as $\lambda \to 0^-$. Moreover, the set $\Sigma = \{(\lambda, u)\in \mathbb{R}\times E, \text{ u solves } (P_\lambda)\}$ possesses an unbounded component $\mathcal{C}^+ \subset [0,+\infty]\times E$ such that $\mathcal{C}^+ \cap (\{0\}\times E) =\{u_0\}.$ 2. This component is such that: either it bifurcates from infinity to the right of the axis $\lambda = 0$ with the corresponding solutions having a positive part blowing up to infinity in $C(\overline{\Omega})$ as $\lambda \to 0^+$; or its projection on the $\lambda$ axis is $[0,+\infty)$. 3. There exists $\overline{\lambda}\in (0,+\infty]$ such that, for every $\lambda\in (0,\overline{\lambda})$, the problem $(P_\lambda)$ has at least two $L^p$-viscosity solutions, $u_{\lambda,1}$ and $u_{\lambda,2}$, satisfying $u_{\lambda,1} \to u_0$ in $E$; $\max_{\overline{\Omega}} u_{\lambda,2}\to +\infty$ as $\lambda \to 0^+$; and if $\overline{\lambda}< +\infty$, the problem $(P_{\overline{\lambda}})$ has at least one $L^p$-viscosity solution. The latter is unique if $F(x, \vec{p}, X)$ is convex in $(\vec{p},X).$ 4. If (H2) holds, the solutions $u_\lambda$ for $\lambda \le 0$ are unique among $L^p$-viscosity solutions; whereas the solutions from 3. for $\lambda > 0$ are ordered in some block. If in addition the system is fully coupled, $u_{\lambda,1}\ll u_{\lambda,2}$, for all $\lambda > 0.$ The authors also show that it is possible to obtain a more precise description of the set $\Sigma$, provided we know the sign of $u_0$.