Sub-exponential tail bounds for conditioned stable Bienaymé-Galton-Watson trees (Q2359736)

The author derives uniform sub-exponential bounds for the width, height, and maximal outdegree of critical Bienaymé-Galton-Watson (BGW) trees conditioned on having a fixed size. The offspring distribution is assumed to belong to the domain of attraction of a stable law and may have infinite variance. Essentially, the approach is similar to that of \textit{L. Addario-Berry} et al. [Ann. Probab. 41, No. 2, 1072--1087 (2013; Zbl 1278.60128)], who obtained sub-Gaussian tail bounds for the width and height in the finite-variance case. The conditioned BGW trees are coded by their Łukasiewicz paths, i.e., nonnegative spectrally positive random walks \(W_n\) conditioned on a late return to \(0\). Bounds on width and outdegree are obtained via bounds on the supremum of \(W_n\), using the fact that due to the spectral positivity, reaching high values and then returning to \(0\) has subexponential cost. The idea remains intact in the infinite variance case, but utilizing it requires techniques differing from those of [loc. cit.], where a subexponential bound on \(P(W_n=-m)\) is used which does not hold in the infinite variance case. Proving the bounds for the height also starts with an observation already found in [loc. cit.]: If a conditioned BGW tree \(t_n\) has a large height, the vertices at late generations will have many children branching off their ancestral line to th e root. This results in a large value for the supremum of the Łukasiewicz path, already known to have subexponential cost. The proof for the bound in [loc. cit.] relies on the fact that in the finite variance case width and height of \(t_n\) both are of the order \(n^{1/2}\). That, however, does not hold when the offspring law has infinite variance. So again, different techniques of proof come into play.