General relative entropy inequality: an illustration on growth models (Q2571748): Difference between revisions

A general method to construct Lyapunov functionals for several linear evolution equations is presented, extending a principle known for evolution equations stemming from a Markov process. Roughly speaking, if \(A\) is a linear operator acting on real-valued functions of \({\mathbb R}^d\) with adjoint \(A^*\), \(n\), \(p>0\) and \(\psi\geq 0\) are solutions to \(\partial_t n=A n\), \(\partial_t p=A p\) and \(-\partial_t \psi=A^* \psi\), and \(H\) is a convex function, the \textit{general relative entropy inequality} asserts that \[ t\longmapsto \int_{{\mathbb R}^d} p(t,x)\;H\left( \frac{n(t,x)}{p(t,x)} \right)\;\psi(t,x)\;dx \] is a nonincreasing function of time. Such a property is in particular valid when \(A\) is a general second-order elliptic differential operator, but also for (i) the fragmentation equation with drift, (ii) the cell division equation and (iii) the renewal equation with periodic coefficients, which are the three equations studied in the paper. Suitable choices of \(H\), \(\psi\) and \(p\) yield contraction and positivity properties, and a priori estimates as well. Of particular interest is the case where it is possible to take \(p(t,x)=e^{\lambda t} N(t,x)\) and \(\psi(t,x)=e^{\lambda t} \phi(t,x)\), \((N,\phi)\) being either independent of time or periodic in time. In the above three cases, the existence of a unique pair \((N,\phi)\) (independent of time in cases (i,ii) and time-periodic in case (iii)) is established and the general relative entropy inequality is used to show the convergence of any solution towards \(p\).