On the stability of critical chemotactic aggregation (Q2249464)

It is by now well-known that non-negative solutions to the two-dimensional Smoluchowski-Poisson equation (also referred to as the parabolic-elliptic Keller-Segel equation) \[ \partial_t u = \mathrm{div}(\nabla u + u \nabla \phi_u) \;\;\text{ in }\;\; (0,\infty)\times \mathbb{R}^2\;, \quad \phi_u = \frac{1}{2\pi}\log{|x|} * u\;, \] have a strikingly different behavior according to the value of the mass \(M=\|u_0\|_1\) of the initial condition \(u_0\geq 0\). More precisely, solutions exist for all times provided \(M\in [0,8\pi]\) and blow up in finite time as soon as \(M>8\pi\). The issue investigated in the paper under review is the blowup rate and the behavior of the solution as time approaches the blowup time, and both are identified for a class of nonnegative radially symmetric initial data with a mass \(M\in (8\pi,8\pi+\alpha^*)\) close to the critical mass \(8\pi\). To this end, an important observation is that the Smoluchowski-Poisson equation in \(\mathbb{R}^2\) shares common features with some critical parabolic and dispersive equations (such as the mass-critical nonlinear Schrödinger equation) and it turns out that the approach developed for the latter can be extended to the former, but in a non-trivial way due in particular to the nonlocal structure of the equation. One of these features is the existence of a steady state \(Q(x)=8\pi (1+|x|^2)^{-2}\) with critical mass \(8\pi\), which is the unique minimizer (up to symmetry) of the logarithmic Hardy-Littlewood-Sobolev inequality. It is in fact shown that there is an open set of nonnegative and radially symmetric initial data \(u_0\) which are close to \(Q\) in a suitably weighted \(H^2\)-space \(H_Q^2\) and have mass \(\|u_0\|_1\in (8\pi,8\pi+\alpha^*)\) for some small \(\alpha^*>0\) with the following properties: the corresponding solution \(u\) to the Smoluchowski-Poisson equation blows up at some finite time \(T>0\) and \[ u(t,x) = \frac{1}{\lambda(t)^2} \left[ Q\left( \frac{x}{\lambda(t)} \right) + \varepsilon\left( t , \frac{x}{\lambda(t)} \right) \right] \] with \(\|\varepsilon(t)\|_{H_Q^2}\to 0\) as \(t\to T\) and \[ \lambda(t) = \sqrt{T-t}\;\exp{\left( -\sqrt{\frac{|\log{T-t}|}{2}} +O(1) \right)} \;\;\text{ as }\;\; t\to T\;. \] The proof involves in particular a detailed study of the linearised equation near \(Q\).