Layered solutions with unbounded mass for the Keller-Segel equation (Q523349)

The authors build families of radial solutions to the (steady) Keller-Segel equation \(-\Delta u+u=e^{\lambda u}\) posed in a bounded and smooth domain \( \Omega \) of \(\mathbb R^{n}\) with the constraint \(u>0\) in \(\Omega \) and the Neumann boundary condition \(\frac{\partial u}{\partial \nu }=0\) on \(\partial \Omega \) . The first main result proves that for given \(0<R_{0}<R_{1}\) there exists \( \overline{\lambda }>0\) such that for every \(\lambda \in (0,\overline{\lambda })\) there exists a radial solution \(u_{\lambda }\) to the Keller-Segel equation in the annulus \(B_{R_{1}}(0)\setminus B_{R_{0}}(0)\) with \(\frac{ \partial u_{\lambda }}{\partial \nu }=0\) on \(\partial (B_{R_{1}}(0)\setminus B_{R_{0}}(0))\) and which satisfies \[ \lim_{\lambda \rightarrow 0}\int_{B_{R_{1}}(0)\setminus B_{R_{0}}(0)}e^{u_{\lambda }(r)}dr=\infty , \quad \lim_{\lambda \rightarrow 0}u_{\lambda }=\sqrt{2}U \] uniformly on the compact subsets of \(B_{R_{1}}(0)\setminus B_{R_{0}}(0)\), where \(U\) is the solution to \(-U^{\prime \prime }-\frac{n-1}{r}U^{\prime }+U=0\) in \((R_{0},R_{1})\) with the boundary conditions \(U(R_{0})=U(R_{1})=1\), and \[ \lim_{\lambda \rightarrow 0}\varepsilon _{\lambda }\lambda e^{u_{\lambda }}=-\left| \partial _{\nu }U(R_{1})\right| ^{-1}\delta _{R_{1}}-\left| \partial _{\nu }U(R_{2})\right| ^{-1}\delta _{R_{2}}. \] Here the positive number \( \varepsilon _{\lambda }\), \(\lambda >0\), is the solution to the equation \(\ln \frac{4}{\varepsilon _{\lambda }^{2}}-\ln \lambda =\frac{\sqrt{2}}{ \varepsilon _{\lambda }}\). The authors then quote from their paper [Calc. Var. Partial Differ. Equ. 56, No. 3, Paper No. 74, 35 p. (2017; Zbl 1375.35097)] the construction of multilayered solutions, that is solutions in \( \cup _{i=0}^{k-1}[\alpha _{i},\alpha _{i+1}]\) for some configuration \( 0=\alpha _{0}<\alpha _{1}<\ldots <\alpha _{k}=1\), of \(-U_{k}^{\prime \prime }-\frac{n-1}{r}U_{k}^{\prime }+U_{k}=0\) with the boundary conditions \( U_{k}^{\prime }(0)=0\), \(U_{k}(\alpha _{i})=1\) and satisfying the reflection law \[ \lim_{\varepsilon \rightarrow 0^{-}}\frac{U_{k}(\alpha _{i}+\varepsilon )-U_{k}(\alpha _{i})}{\varepsilon }=-\lim_{\varepsilon \rightarrow 0^{+}} \frac{U_{k}(\alpha _{i}+\varepsilon )-U_{k}(\alpha _{i})}{\varepsilon }. \] They also recall that for other configurations \(0=\widetilde{\alpha }_{0}< \widetilde{\alpha }_{1}<\ldots <\widetilde{\alpha }_{k}=1\), multilayered solutions may be obtained with the same problem but with the boundary conditions \(U_{k}^{\prime }(0)=0=U_{k}^{\prime }(1)\), \(U_{k}(\alpha _{i})=1\) and satisfying the same reflection law. They then prove the existence of radial solutions \(\widetilde{u}_{\lambda }\) for \(\lambda <\lambda (1)\) to the Keller-Segel equation posed in the unit ball \(B_{1}(0)\) such that \( \lim_{\lambda \rightarrow 0}\widetilde{u}_{\lambda }=\sqrt{2}\widetilde{U} _{1}\) uniformly on the compact subsets of \(B_{\widetilde{\alpha } _{1}}(0)\cup B_{1}(0)\setminus B_{\widetilde{\alpha }_{1}}(0)\) and \( \lim_{\lambda \rightarrow 0}\varepsilon _{\lambda }\lambda e^{\widetilde{u} _{\lambda }}=-\left| \partial _{\nu }\widetilde{U}_{1}(\widetilde{\alpha }_{1})\right| ^{-1}\delta _{\widetilde{\alpha }_{1}}\). The main idea of the proof of these results is to look for a solution in successive annuli: \(u_{\mathrm{peak}}^{0}\) in \((R_{0},R_{0}+\delta )\), \(u_{\mathrm{trans}}^{0}\) in \( (R_{0}+\delta ,R_{0}+2\delta )\), \(u_{\mathrm{int}}\) in \((R_{0}+2\delta ,R_{1}-2\delta )\), \(u_{\mathrm{trans}}^{1}\) in \((R_{1}-2\delta ,R_{1}-\delta )\) and \(u_{\mathrm{peak}}^{1}\) in \((R_{1}-\delta ,R_{1})\). Then the authors adapt ideas developed by \textit{A. Pistoia} and \textit{G. Vaira} in [Proc. R. Soc. Edinb., Sect. A, Math. 145, No. 1, 203--222 (2015; Zbl 1321.35027)] and by \textit{M. del Pino} et al. in [J. Differ. Equations 261, No. 6, 3414--3462 (2016; Zbl 1345.35039)]. They start building \(u_{\mathrm{peak}}^{0}\) and \(u_{\mathrm{peak}}^{1}\) through direct computations and estimates on approximate solutions. They then build \(u_{\mathrm{int}}\) using the solution \(U\) and a fixed point argument. They end proving properties of multilayered solutions.