Grad-Caflisch Pointwise Decay Estimates Revisited

Abstract

In the influential paper (Caflisch in Commun Pure Appl Math 33:651–666, 1980) which was the starting point of the employment of Hilbert expansion method to the rigorous justifications of the fluid limits of the Boltzmann equation, Caflisch discovered an elegant and crucial estimate on each expansion term (Proposition 3.1 in Caflisch (Commun Pure Appl Math 33:651–666, 198)). The proof essentially relied on an estimate of Grad as reported by Grad (in: Proceedings of the 3rd international symposium, held at the Palais de l’UNESCO, Paris, 1962), which was on the pointwise decay properties of \(\mathcal {L}^{-1}\), the pseudo-inverse operator of the linearized Boltzmann collision operator \(\mathcal {L}\), for the hard potential collision kernel, i.e. the power \(0\le \gamma \le 1\). Caflisch’s arguments need the exponential version of Grad’s estimate. However, Grad’s original paper was only on the polynomial decay. In this paper, we revisit and provide a full proof of the Caflisch-Grad type decay estimates and the corresponding applications in the compressible Euler limit of the Boltzmann equaiton. The main novelty is that for the case collision kernel power \(-\frac{3}{2}<\gamma \le 1\), the proof of the pointwise estimate does not use any derivatives. So the potential applications of this estimate could be wider than in the Hilbert expansion. For the completeness of the result, we also prove the almost everywhere pointwise estimate using derivatives for the case \(-3<\gamma \le -\frac{3}{2}\). Furthermore, in the application to fluid limits, \(\mathcal {L}^{-1}\) and the derivatives with respect to the parameters (for example, (t, x), this must happen when \(\mathcal {L}\) is linearized around local Maxwellian which depends on (t, x)) are not commutative. We detailed analyze the estimate of commutators, which was missing in previous literatures of fluid limits of the Boltzmann equation. This estimate is needed in all compressible fluid limits from Boltzmann equation.