Quantitative fluid approximation in transport theory: a unified approach (Q2693721)

The authors present a unified method for the large space-time scaling limit of the linear transport equation written as \(\partial _{t}f+v\cdot \nabla _{x}f=Lf\), where \(f=f(t,x,v)\geq 0\) is the time-dependent density of particles, \((x,v)\in \mathbb{R}^{d}\times \mathbb{R}^{d}\), and \(t\geq 0\). Considering a solution \(f\in L_{t}^{\infty }([0,+\infty );L_{x,v}^{2}( \mathcal{M}^{-1}))\) to the above linear transport equation with initial data \(f_{in}^{(\varepsilon )}\), the authors rescale the solution through \[ f_{\varepsilon }(t,x,v)=\frac{1}{\varepsilon ^{d}}f(\frac{t}{\theta (\varepsilon )},\frac{x }{\varepsilon },v)\in L_{t}^{\infty }([0,+\infty );L_{x,v}^{2}(\mathcal{M} ^{-1})), \] with \[ \theta (\varepsilon )= \begin{cases} \varepsilon ^{\zeta }, &\text{if }\alpha \in (0,+\infty ]\setminus \{2+\beta \},\\ \varepsilon ^{2}\left\vert \ln \varepsilon \right\vert ,&\text{if }\alpha =2+\beta ,\\ \varepsilon ^{\beta /(1+\beta )}/\left\vert \ln \varepsilon \right\vert , &\text{if }\alpha =0, \end{cases} \] where \[ \zeta =\begin{cases}\zeta (\alpha ,\beta )=2&\text{when }\alpha \geq 2+\beta , \\ \frac{\alpha +\beta }{1+\beta }&\text{when }\alpha <2+\beta, \end{cases} \] and they define the weighted rescaled spatial density \( r_{\varepsilon }(t,x)=\int_{\mathbb{R}^{d}}f_{\varepsilon }(t,x,v)\left\lfloor v\right\rceil ^{-\beta }dv\). The rescaled solution \( f_{\varepsilon }\) satisfies the equation \(\theta (\varepsilon )\partial _{t}f_{\varepsilon }+\varepsilon v\cdot \nabla _{x}f_{\varepsilon }=\mathcal{ L}f_{\varepsilon }\). Assuming appropriate hypotheses on the operators \( \mathcal{M}\) and \(L\) and on the initial data \(f_{\varepsilon }(0,.,.)\), the authors prove that \[ \left\Vert f_{\varepsilon }/\mathcal{M}-r\right\Vert _{L_{t}^{2}([0,T];H_{x}^{-\zeta }L_{v}^{2}(\mathcal{M}_{\beta }))}\rightarrow _{\varepsilon \rightarrow 0}0, \quad\text{if }\alpha >\beta , \] \[ \left\Vert \left\vert \ln\frac{2\left\vert \nabla x\right\vert }{1+\left\vert \nabla x\right\vert }\right\vert (\frac{f_{\varepsilon }}{\mathcal{M}} -r)\right\Vert _{L_{t}^{2}([0,T];H_{x}^{-\zeta }L_{v}^{2}(\mathcal{M}_{\beta }))}\rightarrow _{\varepsilon \rightarrow 0}0, \quad\text{if }\alpha =\beta \] and \[ \left\Vert \left\vert \nabla x\right\vert ^{\frac{\beta -\left\vert \alpha \right\vert }{2(1+\beta )}}\left\lfloor \nabla x\right\rceil ^{\frac{-\beta -\left\vert \alpha \right\vert }{2(1+\beta )}}(\frac{f_{\varepsilon }}{ \mathcal{M}}-r)\right\Vert _{L_{t}^{2}([0,T];H_{x}^{-\zeta }L_{v}^{2}( \mathcal{M}_{\beta }))}\rightarrow _{\varepsilon \rightarrow 0}0, \quad\text{if } \alpha \in \lbrack 0,\beta ), \] where \(r=r(t,x)\) solves the equation \(\partial _{t}r=\kappa \Delta _{x}^{\frac{\zeta }{2}}r\), \(t>0\), the initial data \( r(0,\cdot )\) being the limit in \(H^{-\zeta }(\mathbb{R}^{d})\) of \(r_{\varepsilon }(0,\cdot )\). The authors then specialize this result in the cases of scattering or linear Boltzmann operators defined as: \[ \mathcal{L}f(v)=(\int_{ \mathbb{R}^{d}}b(v,v^{\prime })f(v^{\prime })dv^{\prime })\mathcal{M}(v)-\nu (v)f(v), \] where \(\nu (v)=\int_{\mathbb{R}^{d}}b(v,v^{\prime })\mathcal{M} (v^{\prime })dv^{\prime }\) is the collision frequency, \(b=b(v,v^{\prime })\) the collisional kernel and \(\mathcal{M}(v)\) an equilibrium distribution, or in that of the Lévy-Fokker-Planck operators: \[ \mathcal{L}(f)=\Delta _{v}^{s}f+\nabla _{v}\cdot (Uf), \] \(\Delta _{v}^{s}\) being the fractional Laplacian defined for \(s\in (0,1)\) as \(\Delta _{v}^{s}f(v)=-\mathcal{F} ^{-1}[\left\vert \cdot \right\vert ^{2s}\mathcal{F}f(\cdot )](v)\), where \( \mathcal{F}\) is Fourier's transform, \(U(v)=U(\left\vert v\right\vert )\) radially symmetric so that \(\Delta _{v}^{s}\mathcal{M}+\nabla _{v}\cdot (U \mathcal{M})=0\). They verify that the hypotheses they impose are satisfied in these cases.