betaFeedback

WorksMathematics

On the dimension of VV-variable fractals

Abstract

Random recursive sets have been used to model a wide range of fractal phenomena, but they can be slow and difficult to generate. VV-variability address this problem by allowing these sets at most VV distinct shapes at each level of magnification. The formulae for the almost sure Hausdorff dimension of both classes is known, and it has been conjectured that (under mild assumptions) the VV-variable dimension converges to the random recursive dimension as V→∞V\to\infty. We prove this conjecture.

Introduction

Iterated function systems (IFSs) are a popular method for constructing fractals exhibiting self-similarity, and there is a rich and growing literature studying their attractor’s dynamics and dimension. Random iterated function systems generalise classical IFSs by introducing randomness into the construction of the attractor over a family of IFSs. Well studied cases include the random homogeneous model, whereby a single IFS is randomly chosen at each level and applied uniformly across that level, and the random recursive model, where an IFS is chosen independently for each piece at every level of the construction (see [5, 8]). VV-variable sets were introduced by Barnsley, Hutchinson, and Stenflo to interpolate between these two models by requiring there to be at most VV distinct local patterns at each level of magnification [2, 3], with V=1V = 1 the random homogeneous case and V=∞V = \infty the random recursive case.

For random systems made up of similarities and satisfying an open set condition, the Hausdorff dimension of almost every random homogeneous and recursive set can be readily calculated as a probabilistic version of the classical similarity dimension. The case for VV-variable sets however relies on calculating the Lyapunov exponent of the decision trees [4]. Methods for such dimension calculations have since been researched and extended to more general settings, see for example [7, 10]. Nevertheless, it has remained a folklore conjecture since 2012, supported by the computational results in [4], that the dimension of a typical VV-variable set approaches the dimension of a typical random recursive set as V→∞V \to\infty. This is explicitly formulated in [Tro21], which further conjectures the almost sure dimension as a function of VV is strictly increasing (provided its endpoint values are distinct). We prove the first of these conjectures by piecing together techniques in [3, 4] and [GMW88].

Date: November 20, 2026

The author would like to thank Jonathan Fraser for conversations regarding the problem, Antti Käenmäki for an inspiring minicourse on fractal geometric applications of subadditivity, and Johannes Livengood van Vliet for revisions. The author was financially supported by a PhD studentship provided by the University of St Andrews in associated with the EPSRC Open Fellowship (EP/Z533440/1) held by Jonathan Fraser.

V-variable fractals

We will first precisely define the construction of VV-variable iterated function systems.

Random Iterated Function Systems

Definition 2.1. Let F={Fλ={fi,λ}}λ∈ΛF=\{F_\lambda=\{f_{i,\lambda}\}\}_{\lambda\in\Lambda} be a family of IFSs consisting of contracting similitudes on Rd\mathbb{R}^d, each with NλN_\lambda maps and associated contraction factors ri,λr_{i,\lambda}. Let μ\mu be a probability measure on the (possibly infinite) indexing set Λ\Lambda. We say the pair (F,μ)(F,\mu) is a random iterated function system (RIFS). We make the following assumptions:

inf⁡Nλ≥2,N:=sup⁡Nλ<∞,\inf N_\lambda\ge2,\qquad N:=\sup N_\lambda<\infty,
0<rmin⁡:=inf⁡ri,λ≤sup⁡ri,λ=:rmax⁡<1,sup⁡∣fi,λ(0)−0∣<∞.0<r_{\min}:=\inf r_{i,\lambda}\le\sup r_{i,\lambda}=:r_{\max}<1,\qquad\sup\lvert f_{i,\lambda}(0)-0\rvert<\infty.

As mentioned in the introduction, the goal of an RIFS is to apply different function sets at each iteration depth. The choices of such IFSs can be written as a tree. Let [N]∗[N]^* be the set of finite words (including the empty word ∅\varnothing) where [n]:={1,2,…,n}[n]:=\{1,2,\ldots,n\}, let [N]n[N]^n denote the length nn words, and let ∣v∣\lvert v\rvert denote the length of v∈[N]∗v\in[N]^*. For 0≤m<n0\le m<n and v∈[N]nv\in[N]^n, we write v∣m=v1v2⋯vmv|m=v_1v_2\cdots v_m for truncation, and for v,w∈[N]∗v,w\in[N]^* we will write vwvw for their concatenation. We will also view [N]∗[N]^* as the rooted NN-ary tree, where ∅\varnothing is the root, the children of any v∈[N]∗v\in[N]^* are v1,v2,…,vNv1,v2,\ldots,vN, and [N]n[N]^n is the nn-th level of the tree. A subtree of [N]∗[N]^* is a nonempty subset τ⊂[N]∗\tau\subset[N]^* such that v∈τv\in\tau implies v∣k∈τv|k\in\tau for all k<∣v∣k<\lvert v\rvert. We will write τn\tau_n for the nn-th level of τ\tau and τ∣v\tau|_v for the subtree rooted at vv. We will consider coloured subtrees of [N]∗[N]^* defined inductively by associating Λ\Lambda-colourings to nodes at each level.

Definition 2.2. We say τ\tau is a Λ\Lambda-coloured tree for an RIFS (F,μ)(F,\mu) if it is a subtree of [N]∗[N]^* constructed in the following way: for every n∈N0n\in\mathbb{N}_0 and node vv at level nn of the partially constructed tree,

  1. Colour vv with some λ∈Λ\lambda\in\Lambda,

  2. Build level n+1n+1 nodes v1,v2,…,vNλv1,v2,\ldots,vN_\lambda.

Abusing notation, let τ\tau denote both the constructed coloured tree as well as the colouring map which determined it (i.e. the partial function τ:[N]∗⇀Λ\tau:[N]^*\rightharpoonup\Lambda). Let Ω\Omega denote the collection of such trees. For k∈Nk\in\mathbb{N} and v,w∈τkv,w\in\tau_k, we say the two coloured trees τ∣v\tau|_v and τ∣w\tau|_w are isomorphic if vu∈τ⟺wu∈τvu\in\tau\Longleftrightarrow wu\in\tau and τ(vu)=τ(wu)\tau(vu)=\tau(wu) for every u∈[N]∗u\in[N]^*, and say τ\tau is VV-variable if there are at most VV isomorphism classes of coloured subtrees rooted at level kk, for every k∈N0k\in\mathbb{N}_0. Let ΩV\Omega_V denote the set of such trees, so Ω∞=Ω\Omega_\infty=\Omega.

Definition 2.3. The attractor for the RIFS (F,μ)(F,\mu) with respect to τ\tau is

Fτ:=lim⁡n→∞⋃v∈τnfv1,τ(∅)∘fv2,τ(v1)∘⋯∘fvn,τ(v∣n−1)(K)(1)F_\tau:=\lim_{n\to\infty}\bigcup_{v\in\tau_n}f_{v_1,\tau(\varnothing)}\circ f_{v_2,\tau(v_1)}\circ\cdots\circ f_{v_n,\tau(v|_{n-1})}(K) \tag*{(1)}

with respect to the Hausdorff metric, independent of the choice of non-empty compact set K⊆RdK\subseteq\mathbb{R}^d. We assume FF satisfies the Uniform Open Set Condition (UOSC), that is there exists an open set U⊆RdU \subseteq\mathbb{R}^{d} such that fi,λ(U)⊆Uf_{i,\lambda}(U) \subseteq U and fi,λ(U)∩fj,λ(U)=∅f_{i,\lambda}(U) \cap f_{j,\lambda}(U) = \varnothing for every λ∈Λ\lambda\in\Lambda and 1≤i≠j≤Nλ1 \leq i \ne j \leq N_{\lambda}. Further, set

Sλs=∑i=1Nλri,λs,Ss:=Eλ∼μ(Sλs),S‾s:=inf⁡λ{Sλs},S‾s:=sup⁡λ{Sλs}\mathcal{S}_{\lambda}^{s} = \sum_{i=1}^{N_{\lambda}} r_{i,\lambda}^{s}, \qquad\mathcal{S}^{s} := \mathbb{E}_{\lambda\sim\mu}(\mathcal{S}_{\lambda}^{s}), \qquad\underline{\mathcal{S}}^{s} := \inf_{\lambda}\{\mathcal{S}_{\lambda}^{s}\}, \qquad\overline{\mathcal{S}}^{s} := \sup_{\lambda}\{\mathcal{S}_{\lambda}^{s}\}

Note our assumption on Nλ≥2N_{\lambda} \geq2 implies Eλ∼μ(log⁡Sλ0)>0\mathbb{E}_{\lambda\sim\mu}(\log\mathcal{S}_{\lambda}^{0}) > 0 and S0>1\mathcal{S}^{0} > 1.

Theorem 2.4. [5 GMW88] The Hausdorff dimension of a typical random homogenous attractor is the unique ss such that Eλ∼μ(log⁡Sλs)=0\mathbb{E}_{\lambda\sim\mu}(\log\mathcal{S}_{\lambda}^{s}) = 0, and the Hausdorff dimension of a typical random recursive attractor is the unique ss such that Ss=1\mathcal{S}^{s} = 1.

Another perspective of trees

Definition 2.2 constructs arbitrarily coloured trees, but does not impose VV-variability. Our aim in this section is to introduce an alternative tree construction which guarantees the property by tracking the isomorphism class information of each subtree at each level and encoding it in their root vertices. To motivate a formal definition, we will demonstrate a conversion procedure using a tree constructed by our current definition.

Choose some τ∈ΩV\tau\in\Omega_{V}. Fix some λ0∈Λ\lambda_{0} \in\Lambda and i0∈[V]i_{0} \in[V], these will act as ‘default’ choices for colour and isomorphism class. For every n∈N0n \in\mathbb{N}_{0}, let VnV_{n} denote the number of isomorphism classes of subtrees at level nn, index them 1,2,…,Vn1, 2, \ldots, V_{n}, and let vi(n)v_{i}^{(n)} denote the root word of the ii-th isomorphism class representative. Now build the VV-tuple

λ(n)=(τ(v1(n)),τ(v2(n)),…,τ(vVn(n)),λ0,…,λ0⏟V−Vn)\lambda^{(n)} = \left(\tau(v_{1}^{(n)}), \tau(v_{2}^{(n)}), \ldots, \tau(v_{V_{n}}^{(n)}), \underbrace{\lambda_{0}, \ldots, \lambda_{0}}_{V-V_{n}}\right)

and a V×NV \times N matrix J(n)J^{(n)}, setting its ijij-th element to be the unique kk such that τ∣vi(n)=τ∣vk(n+1)\tau|_{v_{i}^{(n)}} = \tau|_{v_{k}^{(n+1)}} for j≤Nλi(n)j \leq N_{\lambda_{i}^{(n)}} and i0i_{0} otherwise, i.e. the index of the isomorphism class which the jj-th child of vi(n)v_{i}^{(n)} belongs to. So the tuple an=(λ(n),J(n))a_{n} = (\lambda^{(n)}, J^{(n)}) records the root node colour of each subtree at the nn-th level, and determines the isomorphism class index of each subtree at the (n+1)(n+1)-th level in τ\tau’s construction. Moreover, the sequence (an)n∈N(a_{n})_{n \in\mathbb{N}} determines the tree. One should note we assigned constants in redundant situations, and chose a specific ordering of the isomorphism classes at each level, so the assignment of the sequence (an)n∈N0(a_{n})_{n \in\mathbb{N}_{0}} to τ\tau is many-to-one. We remark this information can be turned into a VV-tuple of attractors (see [1]), but that isn’t relevant to the work in this paper.

Definition 2.5. Let AV=ΛV×[V]V×N\mathcal{A}_{V} = \Lambda^{V} \times[V]^{V \times N}, and fix some i∈[V]i \in[V] and

a=(λ(0),J(0))(λ(1),J(1))⋯∈AVN0,a = (\lambda^{(0)}, J^{(0)})(\lambda^{(1)}, J^{(1)}) \cdots\in\mathcal{A}_{V}^{\mathbb{N}_{0}},

where λ(n)=(λ1(n),λ2(n),…,λV(n))\lambda^{(n)} = (\lambda_{1}^{(n)}, \lambda_{2}^{(n)}, \ldots, \lambda_{V}^{(n)}), J(n)=(Jij(n))J^{(n)} = (J_{ij}^{(n)}).

Define the Λ\Lambda-coloured tree τa,i\tau^{a,i} and isomorphism class indexing function ι:τa,i→[V]\iota: \tau^{a,i} \to[V] the following way: for each n∈N0n \in\mathbb{N}_{0}, each v∈[N]nv \in[N]^{n} will already be assigned some index ι(v)\iota(v) (with initialisation ι(∅)=i\iota(\varnothing) = i), so

  1. For each vv at level nn of the partially constructed tree,

(i) Set the colour τa,i(v)\tau^{a,i}(v) of vv to λι(v)(n)\lambda^{(n)}_{\iota(v)} and

(ii) Build nodes v1,v2,…,vNτa,i(v)v1,v2,\ldots,vN_{\tau^{a,i}(v)}

(2) For each v∈[N]nv \in[N]^n and j∈[N]j \in[N], let ι(vj)=(J(n))ι(v)j\iota(vj)=(J^{(n)})_{\iota(v)j}.

We will write τa:=τa,1\tau^a := \tau^{a,1} (i.e. giving the root node the sensible isomorphism class indexing). With UVU_V the uniform probability measure on [V][V] for V∈NV \in\mathbb{N}, let

PVμ:=(μ⊗V⊗UV⊗VN)⊗N0\mathbb{P}^{\mu}_V := \left(\mu^{\otimes V} \otimes U_V^{\otimes V N}\right)^{\otimes\mathbb{N}_0}

be the canonical measure (with respect to μ\mu and UVU_V) on the product space AVN0\mathcal{A}_V^{\mathbb{N}_0}. Note PVμ\mathbb{P}^{\mu}_V is invariant under σ\sigma and permutation of [V][V]. Finally, let BV=B(AVN0)\mathcal{B}_V=\mathcal{B}(\mathcal{A}_V^{\mathbb{N}_0}) be the Borel σ\sigma-algebra generated by cylinder sets, and Bn,V⊂BV\mathcal{B}_{n,V}\subset\mathcal{B}_V be the σ\sigma-algebra generated by cylinder sets of words up to length nn (i.e. the smallest σ\sigma-algebra which makes projections onto the kk-th coordinate of the string a0a1…a_0a_1\ldots, for k<nk<n, measurable).

Remark 1. We will similarly let the probability measure on Ω\Omega, denoted P∞μ\mathbb{P}_{\infty}^{\mu}, be the push-forward of μ[N]∗\mu^{[N]^*} under the mapping Λ[N]∗→Ω\Lambda^{[N]^*}\to\Omega. Let B∞\mathcal{B}_{\infty} be the σ\sigma-algebra generated by sets {τ∈Ω∣v∈τ,τ(v)∈Λ′}\{\tau\in\Omega\mid v\in\tau,\tau(v)\in\Lambda'\} for v∈[N]∗v\in[N]^* and μ\mu-measurable Λ′⊂Λ\Lambda'\subset\Lambda, and Bn,∞⊂B∞\mathcal{B}_{n,\infty}\subset\mathcal{B}_{\infty} be the σ\sigma-algebra generated by such sets with ∣v∣<n|v|<n. Additionally one can note that, denoting σ\sigma the left shift on AVN0\mathcal{A}_V^{\mathbb{N}_0}, for v∈τna,iv\in\tau_n^{a,i}

τa,i∣v=τσn(a),ι(v).(2)\tau^{a,i}|_v=\tau^{\sigma^n(a),\iota(v)}. \tag*{(2)}

Flows on trees

We can now define the flow of IFS ss-mass between two subtrees rooted at different levels belonging to different isomorphism classes and use it to calculate the dimension of VV-variable sets.

Definition 2.6. The V×VV\times V flow matrix MaM_a for some a=(λ,J)∈AVa=(\lambda,J)\in\mathcal{A}_V is given by

(Ma(s))ij=∑k=1Nλi{rk,λis∣Jik=j},(3)(M_a(s))_{ij}=\sum_{k=1}^{N_{\lambda_i}}\{r^s_{k,\lambda_i}\mid J_{ik}=j\}, \tag*{(3)}

and for a∈AVN0a\in\mathcal{A}_V^{\mathbb{N}_0} write Ma∣n(s)=Ma0(s)Ma1(s)⋯Man−1(s)M_a|_n(s)=M_{a_0}(s)M_{a_1}(s)\cdots M_{a_{n-1}}(s). Let ∥M∥=∑i,jMi,j\lVert M\rVert=\sum_{i,j}M_{i,j}.

Theorem 2.7 ([4 Main Theorem]). Let (F,μ)(F,\mu) be a RIFS. The Lyapunov exponent of a VV-variable system is equal to

γV(s)=lim⁡n→∞Ea∼PVμ(1nlog⁡∥Ma∣n(s)∥),(4)\gamma_V(s)=\lim_{n\to\infty}\mathbb{E}_{a\sim\mathbb{P}^{\mu}_V}\left(\frac{1}{n}\log\lVert M_a|_n(s)\rVert\right), \tag*{(4)}

and is monotonically decreasing in ss. Additionally, for almost every a∈AVN0a\in\mathcal{A}_V^{\mathbb{N}_0}, the Hausdorff dimension of FτaF_{\tau_a} is the unique zero of γV(s)\gamma_V(s), so define the function D:N∪{∞}→R+D:\mathbb{N}\cup\{\infty\}\to\mathbb{R}^{+} by

D(V)=a.s.-dim⁡H(Fτa),D(∞)=the unique s such that Ss=1.D(V)=\mathop{\mathrm{a.s.}\text{-}\dim}_{H}(F_{\tau_a}),\qquad D(\infty)=\text{the unique }s\text{ such that }\mathcal{S}^s=1.

Proof of the dimension limit

Properties of flows

In [4 §5], they look at the sum over the ii-th row of the finite product of matrices in the definition of γV(s)\gamma_V(s), however this direct definition can only be discussed in the VV-variable setting. We will look at an object studied in [GMW88] for the V=∞V=\infty case which extends the aforementioned notion and contains results similar to Lemma 3.1 and Proposition 3.6. For K∈N0K\in\mathbb{N}_0, s≥0s\geq0 and τ∈Ω\tau\in\Omega, let

F(K,s,τ):=∑v∈τK∏k<Krvk+1,τ(v∣k)s(5)\mathcal{F}(K,s,\tau):=\sum_{v\in\tau_K}\prod_{k<K}r_{v_{k+1},\tau(v|k)}^s \tag*{(5)}

be the total flow of IFS ss-mass over the first KK levels of τ\tau. We will first note the recursive nature of this definition, then relate this flow to the product of matrices in Equation (4).

Lemma 3.1. For all K∈N0K\in\mathbb{N}_0, s≥0s\geq0 and τ∈Ω\tau\in\Omega,

F(K+1,s,τ)=∑i=1Nτ(∅)ri,τ(∅)sF(K,s,τ∣i)(6)\mathcal{F}(K+1,s,\tau)=\sum_{i=1}^{N_{\tau(\varnothing)}}r_{i,\tau(\varnothing)}^s\mathcal{F}(K,s,\tau|i) \tag*{(6)}

and

(S‾s)K≤F(K,s,τ)≤(S‾s)K.(7)\left(\underline{\mathcal{S}}^s\right)^K\leq\mathcal{F}(K,s,\tau)\leq\left(\overline{\mathcal{S}}^s\right)^K. \tag*{(7)}

Proof. We can isolate the first or last term in the product in Eq. (5). The former gives

F(K+1,s,τ)=∑v∈τK+1rv1,τ(∅)s∏1≤k<K+1rvk+1,τ(v∣k)s(8)\mathcal{F}(K+1,s,\tau)=\sum_{v\in\tau_{K+1}}r_{v_1,\tau(\varnothing)}^s\prod_{1\leq k<K+1}r_{v_{k+1},\tau(v|k)}^s \tag*{(8)}

which is equivalent to Eq. (6). The latter gives

F(K,s,τ)=∑v∈τKrvK,τ(v∣K−1)s∏k<K−1rvk+1,τ(v∣k)s,\mathcal{F}(K,s,\tau)=\sum_{v\in\tau_K}r_{v_K,\tau(v|K-1)}^s\prod_{k<K-1}r_{v_{k+1},\tau(v|k)}^s,

thus

S‾sF(K−1,s,τ)≤F(K,s,τ)≤S‾sF(K−1,s,τ).\underline{\mathcal{S}}^s\mathcal{F}(K-1,s,\tau)\leq\mathcal{F}(K,s,\tau)\leq\overline{\mathcal{S}}^s\mathcal{F}(K-1,s,\tau).

Iterating the above gives Eq. (7).

Proposition 3.2. For any a∈AN0a\in\mathcal{A}^{\mathbb{N}_0}, K∈N0K\in\mathbb{N}_0, s≥0s\geq0 and i∈[V]i\in[V],

(Ma∣K(s)1)i=F(K,s,τa,i).(M_{a|K}(s)\mathbf{1})_i=\mathcal{F}(K,s,\tau^{a,i}).

Proof. We will prove this inductively over KK. With the base case trivial, assume true for a fixed KK, and denote a0=(λ,J)a_0=(\lambda,J). Then

(Ma0(s)Mσ(a)∣K(s)1)i=∑j(Ma0(s))ijF(K,s,τσ(a),j)=∑jF(K,s,τσ(a),j)∑k=1Nλi{rk,λis∣Jik=j}=∑k=1Nλirk,λisF(K,s,τσ(a),Jik).\begin{aligned} (M_{a_0}(s)M_{\sigma(a)|K}(s)\mathbf{1})_i=\sum_j(M_{a_0}(s))_{ij}\mathcal{F}(K,s,\tau^{\sigma(a),j}) \\ &= \sum_{j} \mathfrak{F}(K,s,\tau^{\sigma(a),j}) \sum_{k=1}^{N_{\lambda_i}} \{r_{k,\lambda_i}^{s}\mid J_{ik}=j\} \\ &= \sum_{k=1}^{N_{\lambda_i}} r_{k,\lambda_i}^{s}\mathfrak{F}(K,s,\tau^{\sigma(a),J_{ik}}). \end{aligned}

Since τa,i(∅)=λi\tau^{a,i}(\varnothing)=\lambda_i by the first step of tree construction in Definition 2.5, and τa,i∣k=τσ(a),Jik\tau^{a,i}|_k=\tau^{\sigma(a),J_{ik}} by (2), then (6) gives us the desired equality.

This has two immediate consequences.

Corollary 3.3. For any a∈AVN0a\in\mathcal{A}_{V}^{\mathbb{N}_0}, s≥0s\geq0 and K≥1K\geq1,

∥Ma∣K(s)∥=∑i=1VF(K,s,τa,i).\lVert M_{a|K}(s)\rVert=\sum_{i=1}^{V}\mathfrak{F}(K,s,\tau^{a,i}).

Corollary 3.4. For every s≥0s\geq0, K≥1K\geq1 and 1≤i≤V1\leq i\leq V,

Ea∼PVμ(F(K,s,τa,i))=(Ss)K=Eτ∼P∞μ(F(K,s,τ)).\mathbb{E}_{a\sim\mathbb{P}_{V}^{\mu}}\left(\mathfrak{F}(K,s,\tau^{a,i})\right)=(\mathcal{S}^{s})^{K}=\mathbb{E}_{\tau\sim\mathbb{P}_{\infty}^{\mu}}\left(\mathfrak{F}(K,s,\tau)\right).

Proof. Since each ana_n are independent, then

\mathbb{E}_{a\sim\mathbb{P}_{V}^{\mu}}\left(M_{a|K}(s)\right)_{ij}=\left(\mathbb{E}_{a\sim\mu^{\otimes V}\otimes\mathop{\mu}\limits^{\otimes V\mathbb{N}}\left(M_a(s)\right)^{K}\right)_{ij}.}

Additionally, (Ma(s)1)i=Sλis(M_a(s)1)_i=\mathcal{S}^{s}_{\lambda_i} for any a∈AVa\in\mathcal{A}_{V}, so Proposition 3.2 gives the first equality. Additionally, since the colouring of the root is independent to the colouring of the rest of the tree with respect to P∞μ\mathbb{P}_{\infty}^{\mu}, we get the second equality by taking the expectation of both sides in (6) and iterating.

Expectation results

By Corollary 3.3, it suffices to work with these flows when studying γV(s)\gamma_V(s). We will first adapt [3 Theorem 49] to our setting of flows. For a fixed ss we will write

ΦK(τ):=1Klog⁡F(K,s,τ),\Phi_K(\tau):=\frac{1}{K}\log\mathfrak{F}(K,s,\tau),

and will write Φ(τ)\Phi(\tau) when KK is fixed.

Proposition 3.5. For all K,V∈NK,V\in\mathbb{N} and $s\geq 0

∣Ea∼PVμ(ΦK(τa))−Eτ∼P∞μ(ΦK(τ))∣≤N2KVlog⁡SsS‾s(9)\left|\mathbb{E}_{a\sim\mathbb{P}_{V}^{\mu}}\left(\Phi_K(\tau^a)\right)-\mathbb{E}_{\tau\sim\mathbb{P}_{\infty}^{\mu}}\left(\Phi_K(\tau)\right)\right|\leq\frac{N^{2K}}{V}\log\frac{\mathcal{S}^{s}}{\underline{\mathcal{S}}^{s}} \tag*{(9)}

Proof. Fix KK, VV and ss. For any a∈AVN0a \in\mathcal{A}^{N_0}_{V} and τ∈Ω\tau\in\Omega, by Eq. (7)

(S~sSs)−K≤F~(K,s,τa)F~(K,s,τ)≤(S~sSs)K,\left(\frac{\widetilde{\mathcal{S}}^{s}}{\mathcal{S}^{s}}\right)^{-K} \le\frac{\widetilde{\mathcal{F}}(K,s,\tau^{a})}{\widetilde{\mathcal{F}}(K,s,\tau)} \le\left(\frac{\widetilde{\mathcal{S}}^{s}}{\mathcal{S}^{s}}\right)^{K},

so taking the logarithm of both sides and dividing by KK gives the general bound

∣Φ(τa)−Φ(τ)∣≤log⁡S~sSs.(10)\left|\Phi(\tau^{a})-\Phi(\tau)\right| \le\log\frac{\widetilde{\mathcal{S}}^{s}}{\mathcal{S}^{s}}. \tag*{(10)}

For NK≥VN^{K} \ge V this immediately implies Eq. (9), so assume NK<VN^{K}<V. Then there exist a∈AVN0a \in\mathcal{A}^{N_0}_{V} such that ι\iota is injective on [N]k[N]^{k} for k<Kk<K; let E∈BK,VE \in\mathcal{B}_{K,V} denote the collection of such aa. For all k≤Kk \le K, note: (a) ι∣[N]k\iota|_{[N]^{k}} is only determined by J(l)J^{(l)} for l<kl<k and so is independent of λ\lambda, and (b) the injectivity of ι∣[N]k\iota|_{[N]^{k}} implies the colouring λι(v)(k)\lambda^{(k)}_{\iota(v)} of each v∈[N]kv \in[N]^{k} is chosen i.i.d. Since F~(K,s,τ)\widetilde{\mathcal{F}}(K,s,\tau) only sees the first KK levels of τ\tau, then

Ea∼PVμ(Φ(τa)∣E)=Eτ∼P∞μ(Φ(τ)),(11)\mathbb{E}_{a\sim\mathbb{P}^{\mu}_{V}}\left(\Phi(\tau^{a})\mid E\right)=\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\Phi(\tau)\right), \tag*{(11)}

So expanding the left hand side of Eq. (9) and applying Eq. (11), we get

∣Ea∼PVμ(Φ(τa)1Ec)+[Ea∼PVμ(Φ(τa)1E)−PVμ(E)⋅Eτ∼P∞μ(Φ(τ))]−PVμ(Ec)⋅Eτ∼P∞μ(Φ(τ))∣=∣Ea∼PVμ([Φ(τa)−Eτ∼P∞μ(Φ(τ))]1Ec)∣≤PVμ(Ec)sup⁡a∈AVN0∣Φ(τa)−Eτ∼P∞μ(Φ(τ))∣≤PVμ(Ec)log⁡S~sSs,\begin{aligned} \left|\mathbb{E}_{a\sim\mathbb{P}^{\mu}_{V}}\left(\Phi(\tau^{a})\mathbb{1}_{E^{c}}\right)+\left[\mathbb{E}_{a\sim\mathbb{P}^{\mu}_{V}}\left(\Phi(\tau^{a})\mathbb{1}_{E}\right)-\mathbb{P}^{\mu}_{V}(E)\cdot\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\Phi(\tau)\right)\right]-\mathbb{P}^{\mu}_{V}(E^{c})\cdot\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\Phi(\tau)\right)\right| \\ = \left|\mathbb{E}_{a\sim\mathbb{P}^{\mu}_{V}}\left(\left[\Phi(\tau^{a})-\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\Phi(\tau)\right)\right]\mathbb{1}_{E^{c}}\right)\right| \\ \le\mathbb{P}^{\mu}_{V}(E^{c})\sup_{a\in\mathcal{A}^{N_0}_{V}}\left|\Phi(\tau^{a})-\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\Phi(\tau)\right)\right| \\ \le\mathbb{P}^{\mu}_{V}(E^{c})\log\frac{\widetilde{\mathcal{S}}^{s}}{\mathcal{S}^{s}}, \end{aligned}

where the second inequality is a result of applying the general bound Eq. (10). Finally, for an a∈Eca \in E^{c} there is a least k<Kk<K such that two vertices at level kk are labelled the same isomorphism class index. By minimality of kk, this is equivalent to there being two distinct entries of J(k−1)J^{(k-1)} which have equal value. Since entries are chosen i.i.d., the probability a fixed pair agrees is 1/V1/V, and since there are fewer than (Nk2)\binom{N^{k}}{2} choices of pairs at each level, then

PVμ(Ec)≤1V∑k=0K−1(Nk2)≤N2KV.\mathbb{P}^{\mu}_{V}(E^{c})\le\frac{1}{V}\sum_{k=0}^{K-1}\binom{N^{k}}{2}\le\frac{N^{2K}}{V}.

Thus the difference between the expected log-normed flow to a fixed level of a VV-variable and random tree shrinks to zero as V→∞V\to\infty. We now want to find a Lyapunov exponent representation for RIFS systems equivalent to Eq. (4).

Proposition 3.6. For any ss such that S~2s<(Ss)2\widetilde{\mathcal{S}}^{2s}<(\mathcal{S}^{s})^{2},

Eτ∼P∞μ(ΦK(τ))→K→∞log⁡(Ss).\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\Phi_{K}(\tau)\right)\xrightarrow[K\to\infty]{}\log(\mathcal{S}^{s}).

Proof. Let F~(K,s,τ):=F~(K,s,τ)/(Ss)K\widetilde{\mathcal{F}}(K,s,\tau) := \widetilde{\mathcal{F}}(K,s,\tau)/(\mathcal{S}^{s})^{K} denote the normalised flow. We claim its second moment is uniformly bounded in KK. By (6) and the independence of the colouring of the root to the rest of the tree with respect to P∞μ\mathbb{P}^{\mu}_{\infty},

Eτ∼P∞μ(F~(K+1,s,τ)2)=1(Ss)2K+2Eτ∼P∞μ((∑i=1Nτ(∅)ri,τ(∅)sF~(K,s,τ∣i))2)=1(Ss)2K+2∑i=1NEλ∼μ(ri,λ2s)Eτ∼P∞μ(F~(K,s,τ)2)+1(Ss)2K+2∑i≠jEλ∼μ(ri,λsrj,λs)Eτ∼P∞μ(F~(K,s,τ)2)=1(Ss)2S2sEτ∼P∞μ(F~(K,s,τ)2)+1(Ss)2(∑i,jEλ∼μ(ri,λsrj,λs)−∑i=1NEλ∼μ(ri,λ2s))=S2s(Ss)2Eτ∼P∞μ(F~(K,s,τ)2)+Eλ∼μ((Ss)2)−S2s(Ss)2,\begin{aligned} \mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\widetilde{\mathcal{F}}(K+1,s,\tau)^{2}\right) &=\frac{1}{(\mathcal{S}^{s})^{2K+2}}\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\left(\sum_{i=1}^{N_{\tau(\varnothing)}}r_{i,\tau(\varnothing)}^{s}\widetilde{\mathcal{F}}(K,s,\tau\mid i)\right)^{2}\right)\\ &=\frac{1}{(\mathcal{S}^{s})^{2K+2}}\sum_{i=1}^{N}\mathbb{E}_{\lambda\sim\mu}\left(r_{i,\lambda}^{2s}\right)\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\widetilde{\mathcal{F}}(K,s,\tau)^{2}\right)\\ &\quad+\frac{1}{(\mathcal{S}^{s})^{2K+2}}\sum_{i\ne j}\mathbb{E}_{\lambda\sim\mu}\left(r_{i,\lambda}^{s}r_{j,\lambda}^{s}\right)\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\widetilde{\mathcal{F}}(K,s,\tau)^{2}\right)\\ &=\frac{1}{(\mathcal{S}^{s})^{2}}\mathcal{S}^{2s}\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\widetilde{\mathcal{F}}(K,s,\tau)^{2}\right)\\ &\quad+\frac{1}{(\mathcal{S}^{s})^{2}}\left(\sum_{i,j}\mathbb{E}_{\lambda\sim\mu}\left(r_{i,\lambda}^{s}r_{j,\lambda}^{s}\right)-\sum_{i=1}^{N}\mathbb{E}_{\lambda\sim\mu}\left(r_{i,\lambda}^{2s}\right)\right)\\ &=\frac{\mathcal{S}^{2s}}{(\mathcal{S}^{s})^{2}}\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\widetilde{\mathcal{F}}(K,s,\tau)^{2}\right)+\frac{\mathbb{E}_{\lambda\sim\mu}\left((\mathcal{S}^{s})^{2}\right)-\mathcal{S}^{2s}}{(\mathcal{S}^{s})^{2}}, \end{aligned}

where we let ri,λ=0r_{i,\lambda}=0 for Nλ<i≤NN_{\lambda}<i\le N, and use Corollary 3.4 for the third equality. Also note F~(1,s,τ)<∞\widetilde{\mathcal{F}}(1,s,\tau)<\infty for every τ∈Ω\tau\in\Omega, so by our assumption on ss this completes the claim, and moreover shows sup⁡K∥F~(K,s,τ)∥2<∞\sup_{K}\|\widetilde{\mathcal{F}}(K,s,\tau)\|_{2}<\infty. Now notice that Eτ∼P∞μ(F~(K,s,τ))=1\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}(\widetilde{\mathcal{F}}(K,s,\tau))=1 for every KK, and further

Eτ∼P∞μ(F~(K+1,s,τ)∣BK,∞)=F~(K,s,τ),\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\widetilde{\mathcal{F}}(K+1,s,\tau)\mid\mathcal{B}_{K,\infty}\right)=\widetilde{\mathcal{F}}(K,s,\tau),

so the sequence (τ↦F~(K,s,τ))K∈N(\tau\mapsto\widetilde{\mathcal{F}}(K,s,\tau))_{K\in\mathbb{N}} is a martingale adapted to (BK,∞)K∈N(\mathcal{B}_{K,\infty})_{K\in\mathbb{N}}, and converges P∞μ\mathbb{P}^{\mu}_{\infty}-a.e. and in L2L^{2}, to some F~(s,τ)\widetilde{\mathcal{F}}(s,\tau) by the Martingale Convergence Theorem [9].

This has two consequences. First, the convergence paired with uniform boundedness implies that Eτ∼P∞μ(F~(s,τ))=1\mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}(\widetilde{\mathcal{F}}(s,\tau))=1. Second, dividing both sides of (6) by (Ss)K+1(\mathcal{S}^{s})^{K+1} and taking the limit as K→∞K\to\infty gives

F~(s,τ)=1Ss∑i=1Nτ(∅)ri,τ(∅)sF~(s,τ∣i)P∞μ-a.e.,\widetilde{\mathcal{F}}(s,\tau)=\frac{1}{\mathcal{S}^{s}}\sum_{i=1}^{N_{\tau(\varnothing)}}r_{i,\tau(\varnothing)}^{s}\widetilde{\mathcal{F}}(s,\tau\mid i)\quad\mathbb{P}^{\mu}_{\infty}\text{-a.e.},

so F~(s,τ)=0\widetilde{\mathcal{F}}(s,\tau)=0 if and only if F~(s,τ∣i)=0\widetilde{\mathcal{F}}(s,\tau\mid i)=0 for all 1≤i≤Nτ(∅)1\le i\le N_{\tau(\varnothing)}. Moreover,

P∞μ(F~(s,τ)=0)=Eλ∼μ(P∞μ(F~(s,τ)=0)Nλ).\mathbb{P}^{\mu}_{\infty}(\widetilde{\mathcal{F}}(s,\tau)=0)=\mathbb{E}_{\lambda\sim\mu}\left(\mathbb{P}^{\mu}_{\infty}(\widetilde{\mathcal{F}}(s,\tau)=0)^{N_{\lambda}}\right).

By our assumption that Nλ≥2N_{\lambda}\ge2, then P∞μ(F~(s,τ)=0)∈{0,1}\mathbb{P}^{\mu}_{\infty}(\widetilde{\mathcal{F}}(s,\tau)=0)\in\{0,1\}. The first consequence rules out the latter option, so F~(s,τ)>0\widetilde{\mathcal{F}}(s,\tau)>0 almost surely, and log⁡F~(s,τ)∈(−∞,∞)\log\widetilde{\mathcal{F}}(s,\tau)\in(-\infty,\infty). Thus

ΦK(τ)=log⁡Ss+1Klog⁡F~(K,s,τ)→K→∞log⁡SsP∞μ-a.e.\Phi_{K}(\tau)=\log\mathcal{S}^{s}+\frac{1}{K}\log\widetilde{\mathcal{F}}(K,s,\tau)\xrightarrow[K\to\infty]{}\log\mathcal{S}^{s}\quad\mathbb{P}^{\mu}_{\infty}\text{-a.e.}

Finally by Eq. (7), ∣ΦK(τ)∣|\Phi_K(\tau)| is bounded by a constant, so by the Dominated Convergence Theorem

lim⁡K→∞Eτ∼P∞μ(ΦK(τ))=Eτ∼P∞μ(lim⁡K→∞ΦK(τ))=log⁡Ss.\lim_{K \to\infty} \mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\Phi_K(\tau)\right) = \mathbb{E}_{\tau\sim\mathbb{P}^{\mu}_{\infty}}\left(\lim_{K \to\infty}\Phi_K(\tau)\right) = \log\mathcal{S}^{s}.

Corollary 3.7. For any ss such that S2s<(Ss)2\mathcal{S}^{2s} < (\mathcal{S}^{s})^{2}, setting K(V):=⌊14log⁡NV⌋K(V) := \left\lfloor\frac{1}{4}\log_N V\right\rfloor,

Ea∼PVμ(ΦK(V)(τa))→V→∞log⁡(Ss).\mathbb{E}_{a \sim\mathbb{P}^{\mu}_{V}}\left(\Phi_{K(V)}(\tau^{a})\right) \xrightarrow[V \to\infty]{} \log(\mathcal{S}^{s}).

Proof. Combine Propositions 3.5 and 3.6, noting that N2K(V)/V≤V−1/2N^{2K(V)}/V \le V^{-1/2}.

Proof of D(V)→D(∞)D(V) \to D(\infty)

We will now bound γV(s)\gamma_V(s) by sequence and limit in Corollary 3.7, allowing us to use a Squeeze theorem argument to prove the dimension limit.

Proposition 3.8. For any K,V∈NK,V \in\mathbb{N} and s≥0s \ge0,

Ea∼PVμ(ΦK(τa))≤γV(s)≤log⁡Ss.\mathbb{E}_{a \sim\mathbb{P}^{\mu}_{V}}\left(\Phi_K(\tau^{a})\right) \le\gamma_V(s) \le\log\mathcal{S}^{s}.

Proof. By Jensen’s inequality,

Ea∼PVμ(1Klog⁡∑i=1VS(K,s,τa,i))≤1Klog⁡Ea∼PVμ(∑i=1VS(K,s,τa,i))=1Klog⁡V+log⁡Ss,\mathbb{E}_{a \sim\mathbb{P}^{\mu}_{V}}\left(\frac{1}{K}\log\sum_{i=1}^{V}\mathcal{S}(K,s,\tau^{a,i})\right) \le\frac{1}{K}\log\mathbb{E}_{a \sim\mathbb{P}^{\mu}_{V}}\left(\sum_{i=1}^{V}\mathcal{S}(K,s,\tau^{a,i})\right) = \frac{1}{K}\log V + \log\mathcal{S}^{s},

where the equality comes from Corollary 3.4. Thus as K→∞K \to\infty the first term vanishes and Corollary 3.3 gives us the upper bound.

To show the lower bound, letting xK:=Ea∼PVμlog⁡S(K,s,τa)x_K := \mathbb{E}_{a \sim\mathbb{P}^{\mu}_{V}}\log\mathcal{S}(K,s,\tau^{a}) we claim (xK)K∈N(x_K)_{K \in\mathbb{N}} is superadditive. By isolating the KK terms of the product in Eq. (5), for any τ∈Ω\tau\in\Omega we get

S(L+K,s,τ)=∑v∈τK(∏k<Krvk+1,τ(v∣k)s)S(L,s,τ∣v).\mathcal{S}(L+K,s,\tau) = \sum_{v \in\tau_K}\left(\prod_{k<K}r_{v_{k+1},\tau(v|k)}^{s}\right)\mathcal{S}(L,s,\tau|v).

Now let wv:=∏k<Krvk+1,τa(v∣k)s/S(K,s,τa)w_v := \prod_{k<K}r_{v_{k+1},\tau^{a}(v|k)}^{s}/\mathcal{S}(K,s,\tau^{a}), so ∑v∈τKwv=1\sum_{v \in\tau_K}w_v=1. Then by Jensen’s inequality and Eq. (2),

log⁡S(L+K,s,τa)≥log⁡S(K,s,τa)+∑v∈(τa)K∑i=1Vwvlog⁡S(L,s,τσK(a),i)1ι(v)=i,(12)\log\mathcal{S}(L+K,s,\tau^{a}) \ge\log\mathcal{S}(K,s,\tau^{a}) + \sum_{v \in(\tau^{a})_K}\sum_{i=1}^{V}w_v\log\mathcal{S}(L,s,\tau^{\sigma^{K}(a),i})\mathbf{1}_{\iota(v)=i}, \tag*{(12)}

where the second term was achieved by splitting the possible evaluations of ι(v)\iota(v). By the independence of data between the first KK levels and the rest, as well as the invariance of PVμ\mathbb{P}^{\mu}_{V} under permutation of [V][V], we have

Ea∼PVμ(wvlog⁡S(L,s,τσK(a),i)1ι(v)=i)=xL⋅Ea∼PVμ(wv1ι(v)=i).\mathbb{E}_{a \sim\mathbb{P}^{\mu}_{V}}\left(w_v\log\mathcal{S}(L,s,\tau^{\sigma^{K}(a),i})\mathbf{1}_{\iota(v)=i}\right) = x_L\cdot\mathbb{E}_{a \sim\mathbb{P}^{\mu}_{V}}\left(w_v\mathbf{1}_{\iota(v)=i}\right).

Therefore, taking expectations with respect to PVμ\mathbb{P}^{\mu}_{V} of both sides of (12) gives the claim. So by Fekete’s lemma and Corollary 3.3,

Ea∼PVμ(ΦK(τa))≤sup⁡KxKK=lim⁡K→∞xKK≤γV(s).\mathbb{E}_{a\sim\mathbb{P}^{\mu}_{V}}\left(\Phi_{K}(\tau^{a})\right) \leq\sup_{K}\frac{x_{K}}{K}=\lim_{K\to\infty}\frac{x_{K}}{K}\leq\gamma_{V}(s).

In order to transfer convergence of γV\gamma_{V} to its zeros, we require the following lemma.

Lemma 3.9. For all V∈NV\in\mathbb{N} and s,t≥0s,t\geq0,

γV(s+t)≤γV(s)−tlog⁡rmax⁡−1.\gamma_{V}(s+t)\leq\gamma_{V}(s)-t\log r_{\max}^{-1}.

Proof. By observation of (3), for any a∈AVa\in\mathcal{A}_{V} we have (Ma(s+t))ij≤rmax⁡t(Ma(s))ij(M_{a}(s+t))_{ij}\leq r_{\max}^{t}(M_{a}(s))_{ij}, thus

∥Ma∣n(s+t)∥≤rmax⁡tn∥Ma∣n(s)∥.\lVert M_{a|n}(s+t)\rVert\leq r_{\max}^{tn}\lVert M_{a|n}(s)\rVert.

Taking the logarithms and expectations of both sides, dividing by nn, and letting n→∞n\to\infty gives the result.

Corollary 3.10. For any V∈NV\in\mathbb{N}, γV(s)\gamma_{V}(s) is decreasing in ss. Moreover, D(1)≤D(V)≤D(∞)D(1)\leq D(V)\leq D(\infty).

Proof. For the second part, by Proposition 3.8

γV(D(V))≥Ea∼PVμ(Φ1(τa))=Eλ∼μ(log⁡SλD(V))=γ1(D(V)),\gamma_{V}(D(V))\geq\mathbb{E}_{a\sim\mathbb{P}^{\mu}_{V}}\left(\Phi_{1}(\tau^{a})\right)=\mathbb{E}_{\lambda\sim\mu}\left(\log\mathcal{S}_{\lambda}^{D(V)}\right)=\gamma_{1}(D(V)),

and

γV(D(∞))≤log⁡SD(∞)=0.\gamma_{V}(D(\infty))\leq\log\mathcal{S}^{D(\infty)}=0.

Theorem 3.11. Let (F,μ)(F,\mu) be a RIFS. Then D(V)→V→∞D(∞)D(V)\xrightarrow{V\to\infty}D(\infty).

Proof. Let s∗=D(∞)s^{*}=D(\infty). By setting s=D(V)s=D(V) and t=s∗−D(V)t=s^{*}-D(V) (which is non-negative by Corollary 3.10), Lemma 3.9 gives

s∗−D(V)≤−γV(s∗)log⁡rmax⁡−1.s^{*}-D(V)\leq-\frac{\gamma_{V}(s^{*})}{\log r_{\max}^{-1}}.

Finally, by our assumption that S0>1\mathcal{S}^{0}>1 then s∗>0s^{*}>0, so by definition of s∗s^{*} we have S2s∗<Ss∗=1=(Ss∗)2\mathcal{S}^{2s^{*}}<\mathcal{S}^{s^{*}}=1=(\mathcal{S}^{s^{*}})^{2}. Thus by Corollary 3.7 and Proposition 3.8, the Squeeze theorem gives

γV(s∗)→V→∞log⁡Ss∗=0.\gamma_{V}(s^{*})\xrightarrow{V\to\infty}\log\mathcal{S}^{s^{*}}=0.

C. de Wit, School of Mathematics and Statistics, University of St Andrews, St Andrews, KY16 9SS, United Kingdom.

Email address: [email protected]

References

  1. [1]Michael Barnsley. Superfractals. Cambridge University Press, Cambridge, United Kingdom, 2006.
  2. [2]Michael Barnsley, John E. Hutchinson, and Örjan Stenflo. A fractal valued random iteration algorithm and fractal hierarchy. Fractals, 13(2):111–146, 2005.DOI
  3. [3]Michael Barnsley, John E. Hutchinson, and Örjan Stenflo. V-variable fractals: Fractals with partial self similarity. Advances in Mathematics, 218:2051–2088, 2008.arxiv.org/abs/0802.0064
  4. [4]Michael Barnsley, John E. Hutchinson, and Örjan Stenflo. V-variable fractals: dimension results. Forum Math., 24(3):445–470, 2012.
  5. [5]Kenneth J. Falconer. Random fractals. Math. Proc. Cambridge Philos. Soc., 100:559–582, 1986.DOI
  6. [7]Esa Järvenpää, Maarit Järvenpää, Antti Käenmäki, Henna Koivusalo, Örjan Stenflo, and Ville Suomala. Dimensions of random affine code tree fractals. Ergodic Theory and Dynamical Systems, 34(3):854–875, 2014.DOI
  7. [8]R. Daniel Mauldin and S. C. Williams. Random recursive constructions: Asymptotic geometric and topological properties. Trans. Am. Math. Soc., 295(1):325–346, 1986.DOI
  8. [9]Karl R. Stromberg. Probability for Analysts. Chapman & Hall, New York, USA, 1994.
  9. [10]Sascha Troscheit. On the dimensions of attractors of random self-similar graph directed iterated function systems. Journal of Fractal Geometry, 4(3):257–303, 2017.DOI

Paper details

vlnc:W_an8epxtm
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents