We revisit the discrete additive and multiplicative coalescents, startingwith $n$ particles with unit mass. These cases are known to be related to some"combinatorial coalescent processes": a time reversal of a fragmentation ofCayley trees or a parking scheme in the additive case, and the random graphprocess $(G(n,p))_p$ in the multiplicative case. Time being fixed, encodingthese combinatorial objects in real-valued processes indexed by the line is thekey to describing the asymptotic behaviour of the masses as $n\to +\infty$. We propose to use the Prim order on the vertices instead of the classicalbreadth-first (or depth-first) traversal to encode the combinatorial coalescentprocesses. In the additive case, this yields interesting connections betweenthe different representations of the process. In the multiplicative case, itallows one to answer to a stronger version of an open question of Aldous [Ann.Probab., vol. 25, pp. 812--854, 1997]: we prove that not only the sequence of(rescaled) masses, seen as a process indexed by the time $\lambda$, convergesin distribution to the reordered sequence of lengths of the excursions abovethe current minimum of a Brownian motion with parabolic drift $(B_t+\lambda t -t^2/2, t\geq 0)$, but we also construct a version of the standard augmentedmultiplicative coalescent of Bhamidi, Budhiraja and Wang [Probab. Theory Rel.,to appear] using an additional Poisson point process.
展开▼
机译:我们从具有单位质量的$ n $粒子开始,重新讨论离散的累加和乘法合并。已知这些情况与某些“组合合并过程”有关:在加性情况下,Cayley树碎片的时间反转或停车方案,以及在乘法中的随机图过程$(G(n,p))_ p $案件。固定时间后,将这些组合对象编码为按行索引的实值过程是将质量的渐近行为描述为$ n \至+ \ infty $的关键。我们建议在顶点上使用基本顺序,而不是经典的广度优先(或深度优先)遍历来对组合的合并过程进行编码。在加性情况下,这会在过程的不同表示之间产生有趣的联系。在乘法的情况下,它允许人们回答Aldous [Ann.Probab。,vol。 25,第812--854页,1997年]:我们证明,不仅(重定标的)质量序列(被视为以时间\\ lambda $索引的过程)还收敛到分布在上方的偏移长度的重排序序列抛物线漂移$(B_t + \ lambda t -t ^ 2/2,t \ geq 0)$的布朗运动的当前最小值,但是我们还构造了Bhamidi,Budhiraja和Wang [Probab。使用相对的Poisson点过程。
展开▼