Asymptotic behavior of the growth-fragmentation equation with bounded fragmentation rate (Q517107)

The paper under review deals with the large time behaviour of solutions to the growth-fragmentation equation \[ \partial_t f(t,x) + \partial_x\left( \tau(x) f(t,x) \right) = (\mathcal{F}f)(t,x)\;, \qquad (t,x)\in (0,\infty)^2\;, \] supplemented with no-flux boundary condition \(\tau(0) f(t,0)=0\) for \(t>0\) and initial condition \(f(0,x)=f^{in}(x)\), \(x\in (0,\infty)\). The fragmentation operator \(\mathcal{F}\) only acts on the size variable \(x\) and reads \[ (\mathcal{F}f)(x) = - B(x) f(x) + \int_0^1 B\left( \frac{x}{z} \right) f\left( \frac{x}{z} \right) \frac{\mathfrak{p}(dz)}{z}\;, \qquad x>0\;, \] the fragmentation rate \(\mathfrak{p}\) being a positive measure on \((0,1)\) which satisfies the mass conservation \[ \int_0^1 z \mathfrak{p}(dz) = 1\;. \] Under suitable conditions on \(\tau\), \(B\), and \(\mathfrak{p}\), it is known that there are \(\lambda\geq 0\) and two non-negative functions \(G\in L^1(0,\infty)\) and \(\phi\in W_{\mathrm{loc}}^{1,\infty}(0,\infty)\) such that \[ \partial_x (\tau G) + \lambda G = \mathcal{F}G \;\text{ in }\; \mathcal{D}'(0,\infty)\;, \qquad \|G\|_1=1\;, \] and \(\phi\) is a strong solution to the dual problem satisfying \(\langle G , \phi \rangle = 1\). It is also known that \(f(t)\) behaves as \(e^{\lambda t} \langle f^{in} , \phi \rangle G\) as \(t\to\infty\) in \(L^1_\phi = L^1(0,\infty;\phi(x)dx)\). It is then tempting to figure out whether this convergence takes place at an exponential rate and whether an inequality of the form \[ \| e^{-\lambda t} f(t) - \langle f^{in} , \phi \rangle G \|_{L_\phi^1} \leq C e^{-\mu t} \| f^{in} - \langle f^{in} , \phi \rangle G \|_{L_\phi^1} \] holds true for some \(C>0\) and \(\mu>0\). While weaker versions of the previous estimate are available but for a norm stronger than \(\|\cdot\|_{L_\phi^1}\) on the right-hand side, the purpose is here to show that this is not true if \(B\) is constant for sufficiently large sizes, that is, \(B(x)=B_0>0\) for \(x\geq A_0\). More precisely, assume also that the growth rate \(\tau\in C^1((0,\infty))\) is a positive function satisfying \(\tau_1 x^{\alpha_1} \leq \tau(x) \leq \tau_2 x^{\alpha_2}\) for \(x\) large enough with two positive constants \(\tau_1\) and \(\tau_2\) and exponents \(\alpha_1\leq \alpha_2 < 1\). Assume finally that the function \[ p(r) = \int_0^1 z^r \mathfrak{p}(dz) \] is well defined on \((\underline{r},\infty)\) for some \(\underline{r}\in [-\infty,0]\) and that \(p(r)\to \infty\) as \(r\searrow \underline{r}\). Under these assumptions, it is shown that the operator \(\mathcal{A}\) defined in \(L_\phi^1\) by \(\mathcal{A}g = - \partial_x(\tau g) - \lambda g + \mathcal{F}g\) with domain \(D(\mathcal{A}) = \{ g \in L_\phi^1\;: \;\partial_x(\tau g)\in L_\phi^1\;, \;(\tau g)(0) = 0 \}\) generates a continuous semigroup \((T_t)_{t\geq 0}\) on \(L_\phi^1\) which satisfies \[ \| T_t - P \|_{\mathcal{L}(L_\phi^1)} \geq 1\;, \qquad t\geq 0\;, \] where \(Pg = \langle g , \phi \rangle G\). Several steps are needed to prove this result. First, the precise algebraic growth of \(\phi\) as \(x\to\infty\) is identified. Next, assuming further that \(\mathrm{supp}\, \mathfrak{p} \subset (z_0,1)\) for some \(z_0>0\), the Dyson-Phillips representation of \((T_t)_{t\geq 0}\) is used to show that \(\| T_t - P \|_{\mathcal{L}(L_\phi^1)}=2\) for all \(t\geq 0\). The final step is to remove the previous assumption on the support of \(\mathfrak{p}\) by an approximation argument and relies on functional analytic tools such as the characterization of quasi-compactness for semigroups. Some care is needed here to guarantee that the function \(\phi\) associated to the approximation has the same behaviour for large \(x\) as the original one.