Special fast diffusion with slow asymptotics. Entropy method and flow on a Riemannian manifold

TL;DR

This paper investigates the asymptotic behavior of solutions to a specific fast diffusion equation at a critical exponent, using geometric and entropy methods to reveal polynomial decay rates in contrast to exponential decay seen in other cases.

Contribution

The study introduces a novel geometric interpretation of the linearized flow as a heat flow on a Riemannian manifold, enabling new inequalities and asymptotic analysis for the critical exponent case.

Findings

Asymptotic behavior is polynomial decay for the critical exponent case.

The linearized flow corresponds to heat flow on a conformally related Riemannian manifold.

New Gagliardo-Nirenberg inequalities are established for the generator.

Abstract

We consider the asymptotic behaviour of positive solutions $u(t,x)$ of the fast diffusion equation $u_t=\Delta (u^{m}/m)={\rm div} (u^{m-1}\nabla u)$ posed for $x\in\RR^d$, $t>0$, with a precise value for the exponent $m=(d-4)/(d-2)$. The space dimension is $d\ge 3$ so that $m<1$, and even $m=-1$ for $d=3$. This case had been left open in the general study \cite{BBDGV} since it requires quite different functional analytic methods, due in particular to the absence of a spectral gap for the operator generating the linearized evolution. The linearization of this flow is interpreted here as the heat flow of the Laplace-Beltrami operator of a suitable Riemannian Manifold $(\RR^d,{\bf g})$, with a metric ${\bf g}$ which is conformal to the standard $\RR^d$ metric. Studying the pointwise heat kernel behaviour allows to prove {suitable Gagliardo-Nirenberg} inequalities associated to the generator. Such inequalities in turn allow to study the nonlinear evolution as well, and to determine its asymptotics, which is identical to the one satisfied by the linearization. In terms of the rescaled representation, which is a nonlinear Fokker--Planck equation, the convergence rate turns out to be polynomial in time. This result is in contrast with the known exponential decay of such representation for all other values of $m$.