On conditioning a self-similar growth-fragmentation by its intrinsic area (Q2041840)

A Markovian growth-fragmentation is a Crump-Mode-Jagers type branching process that can be considered as a model to describe the masses of individuals in a family of living cells. These evolve independently, and the dynamics of the mass of a typical cell is determined by a Markov process on the positive real half-line. The times of negative jumps are interpreted as times at which daughter cells are born with initial masses given by the absolute magnitudes of the corresponding jumps (preserving the total mass at birth events). Moreover, if these dynamics are self-similar, the process that records the masses of cells at birth is a branching random walk. This naturally leads to an additive (Biggins') martingale whose terminal value \(\mathtt{A}\) is called the intrinsic area of the growth fragmentation, see [\textit{J.~Bertoin} et al., Probab.~Theory Relat.~Fields 172, No. 3-4, 663--724, (2018, Zbl 1451.60047)]. The main results of the paper under review are that the density \(a\) of \(\mathtt{A}\) exists, is a smooth function and does not vanish at every positive point. Moreover, \[ a(x)\sim c \gamma x^{-1 -\gamma}\quad \text{as }x\to\infty. \] The proofs rely on the fact that the random variable \(\mathtt{A}\) can be viewed as the limit of an additive (Biggins) martingale in an associated branching random walk. Consequently, the law of \(\mathtt{A}\) is a fixed point of a smoothing transformation. Many of the properties of \(\mathtt{A}\) can be lifted from the existing literature on fixed points of smoothing transforms. In particular, conditions for the existence of the density are known as well as conditions implying that \(\mathtt{A}\) has a power tail: \(P(\mathtt{A} > r) \sim c r^{-\gamma}\) for some \(\gamma\). The tail asymptotics of \(\mathtt{A}\) are then lifted to the level of the density. The main results then enable the authors to study the limit self-similar growth-frag\-mentation process conditioned on \(\mathtt{A}=r\) as \(r\) goes to \(\infty\) (a natural question arising in random planar geometry). Proofs rely on the fact that the displacements of the underlying branching random walk are governed by the discontinuities of an exponent of some Lévy process with negative jumps. This enables the authors, e.g.,\ to derive for \(\mathtt{A}^-\), a biased version of \(\mathtt{A}\), a conditional perpetuity-type distributional equation. Moreover, it is a nice observation that for some fixed points of particular smoothing transformations, results on the tail probabilities can be lifted to corresponding statements for their densities. A possible route to such extensions is sketched in Section 3.4 of the paper, where an alternative approach to Theorem 1.2 is outlined.