Random gluing of metric spaces

Abstract: We construct random metric spaces by gluing together an infinite sequence of pointed metric spaces that we call blocks. At each step, we glue the next block to the structure constructed so far by randomly choosing a point on the structure and then identifying it with the distinguished point of the block. The random object that we study is the completion of the structure we obtain after an infinite number of steps. We introduce a sequence

(w_{n})_{n g e q 1}

that we call the weights of the blocks. The probability at each step that the next block is glued onto any of the preceding blocks is proportional to its weight. We suppose that the blocks are i.i.d. copies of the same random metric space, scaled by deterministic factors that we call

(l a m b d a_{n})_{n g e q 1}

. We work under some conditions on the distribution of the blocks ensuring that they a.s. have dimension

d

, for some

d > 0

. The main contribution of this paper is the computation of the Hausdorff dimension of the set

m a t h c a l L

of points which appear during the completion procedure, which we call the leaves, when

(l a m b d a_{n})_{n g e q 1}

and

(w_{n})_{n g e q 1}

typically behave like a power of

n

, say

n^{- a l p h a}

for the scaling factors and

for the weights. For a large domain of

a l p h a

and

, we have

m a t h r m d i m_{H} (m a t h c a l L) = a l p h a^{- 1}

. However for

and

a l p h a < 1 / d

, our results reveal an interesting phenomenon: the dimension has a non-trivial dependence in

a l p h a

,

and

d

, namely

.}

Review text

Denote by \((\lambda_n)_{n\ge 1}\) and \((\omega_n)_{n\ge 1}\) two sequences of nonnegative real numbers that will be the scaling factors and weights of the metric spaces, which are gluing. All the scaling factors \((\lambda_n)_{n\ge 1}\) are considered strictly positive, but the weights, except for \(\omega_1\), can possibly be zero. Note that if \((\lambda_n)_{n\ge 1} = (\omega_n)_{n\ge 1}\), the model coincides with the one studied in [\textit{N. Curien} and \textit{B. Haas}, Ann. Inst. Fourier 67, No. 5, 1963--2001 (2017; Zbl 1404.60020)]. Denote by \((\mathbf b_n)_{n\ge 1}\) segments of length \((\lambda_n)_{n\ge 1}\), rooted at one end and endowed with the Lebesque measure normalized so that their respective total measure is \((\omega_n)_{n\ge 1}\) (or endowed with the null measure for branches with vanishing weight). Let \((\mathcal T_n)_{n\ge 1}\) be a sequence of increasing trees by gluing those branches as follows. First, \(\mathcal T_1 = \mathbf b_1\). Then, if \(\mathcal T_n\) is constructed, \(\mathcal T_{n+1}\) is built by first sampling a point \(X_n\) chosen proportionally to the measure \(\mu_n\) obtained by aggregating the measures concentrated on the branches \(\mathbf b_1,\dots,\mathbf b_n\) and then gluing \(\mathbf b_{n+1}\) onto \(\mathcal T_n\) by identifying its root with \(X_n\). Let \(\mathcal T^*\) be the increasing union of the trees \(\mathcal T_n\) for \(n\ge 1\) and \(\mathcal T\) be the completion of \(\mathcal T^*\). The author computes the Hausdorff dimension of the resulting tree in the case where \((\lambda_n)_{n\ge 1}\) and \((\omega_n)_{n\ge 1}\) behave like powers of \(n\), say \(\lambda_n = n^{-\alpha}\) and \(\omega_n = n^{-\beta}\). In the particular case \(\mathcal L=\mathcal T\setminus\mathcal T^*\), \(\dim_H(\mathcal L)= \frac{1}{\alpha} (\dim_H(\cdot)\) stands for the Hausdorff dimension) as in [loc. cit.] for most values of \(\beta\), however, a new phenomenon, absent from [loc. cit.], happens in the case \(\beta>1\) (the sum of the weights is finite) and \(\alpha< 1\) (the total length is infinite). In this case, the Hausdorff dimension of \(\mathcal T\) depends a nontrivial manner on \(\alpha\) and \(\beta\). The author generalizes the model where the sequence \((\mathbf b_n)\) is a more general metric space (see Theorem 1). The assumptions of Theorem 1 are rather general and various known models fall into this setting, see [loc. cit.] and [\textit{N. Ross} and \textit{Y. Wen}, Electron. J. Probab. 23, Paper No. 5, 35 p. (2018; Zbl 1390.60046)]. The paper is organized as follows. Section 1 reviews the definition of the model, set up some notation, and discuss some general properties. In Section 2, the author studies the normalized \(\overline{\mu}_n\) on \(\mathcal T_n\) and prove that it converges to a measure \(\overline{\mu}\) on \(\mathcal T\) under suitable assumptions. In Section 3.1, the author proves the almost sure compactness of \(\mathcal T\) and some upper bounds on its Hausdorff dimension under some relatively weak hypotheses. Section 3.2 develops a new approach that allows to obtain a better upper bound for some parameters for which the former fails to be optimal. In Section 4, the lower bounds are constructed, that match the upper bounds obtained in Section 3. It is again divided in two subsections, each providing a proof that is only valid for some choices of parameters \(\alpha\) and \(\beta\). Some of the technical facts needed are collected in the appendices.

Recommendations

Cites work

Cited in

(10)

This page was built for publication: Random gluing of metric spaces

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2189459)