OctoInformBeta

OpenAI preprint, September 2026 · §6. Overlap conventions and dimension of the attractor

Corollary 6.3: Compactness and the similarity-root formula for finite families, including a single similarity

Attractor dimension without exact overlaps.

Hover over a blue or green term in the text, or a node in the graph, to find it in the other. Select it to read its declaration both in Lean and in words.

The Setting

Let II be a finite nonempty alphabet of cardinality NN, and let ri,ti∈ℝr_i,t_i\in\mathbb R for i∈Ii\in I. Consider the similarities φi(x)=rix+ti,0<|ri|<1.\varphi_i(x)=r_i x+t_i, \qquad 0<|r_i|<1.(1) A weighted affine iterated function system consists of the maps (1) and real weights pi>0p_i>0 with ∑i∈Ipi=1\sum_{i\in I}p_i=1.✦ The uniformly weighted system for this family has pi=1/Np_i=1/N.✦

For a finite word ww over II, the affine word coefficients (aw,bw)∈ℝ2(a_w,b_w)\in\mathbb R^2 are the signed slope and translation of its composition in address order; write ⌀\varnothing for the empty word and iwiw for a symbol prefixed to ww: (a⌀,b⌀)=(1,0),(aiw,biw)=(riaw,ribw+ti),φi1⋯in=φi1∘⋯∘φin.\begin{aligned} (a_{\varnothing},b_{\varnothing})&=(1,0),\\ (a_{iw},b_{iw})&=(r_i a_w,r_i b_w+t_i),\\ \varphi_{i_1\cdots i_n} &=\varphi_{i_1}\circ\cdots\circ\varphi_{i_n}. \end{aligned}(2) A family has no exact overlaps if any two nonempty words with the same coefficient pair in (2) are equal.

For an infinite address ω∈Iℕ\omega\in I^{\mathbb N}, the prefix contraction product an(ω)a_n(\omega) and scaled translation term bn(ω)b_n(\omega) give the coding map π:Iℕ→ℝ\pi:I^{\mathbb N}\to\mathbb R by an(ω)=∏j=0n−1rωj,bn(ω)=an(ω)tωn,π(ω)=∑n=0∞bn(ω)=∑n=0∞(∏j=0n−1rωj)tωn.\begin{aligned} a_n(\omega)&=\prod_{j=0}^{n-1}r_{\omega_j}, & b_n(\omega)&=a_n(\omega)t_{\omega_n},\\ \pi(\omega)&=\sum_{n=0}^{\infty}b_n(\omega) =\sum_{n=0}^{\infty} \left(\prod_{j=0}^{n-1}r_{\omega_j}\right)t_{\omega_n}. \end{aligned}(3) The coding attractor K⊂ℝK\subset\mathbb R is the range of the map (3); it is also the unweighted coding attractor of the uniformly weighted system: K={π(ω):ω∈Iℕ}.K=\{\pi(\omega):\omega\in I^{\mathbb N}\}.(4) The Hausdorff dimension dim⁡HK\dim_H K of the set (4) takes values in [0,∞][0,\infty].

The similarity sum P:ℝ→ℝP:\mathbb R\to\mathbb R of the family is P(s)=∑i∈I|ri|s.P(s)=\sum_{i\in I}|r_i|^s.(5)✦ A nonnegative similarity root ss of the family is a real number with s≥0s\ge0 and P(s)=1P(s)=1 in (5). The similarity dimension s*s_* of the family is the nonnegative root of P(s)=1P(s)=1 in (5).✦

The Claim

Theorem 1. Suppose the family has no exact overlaps. For every nonnegative similarity root ss, dim⁡HK=min⁡{1,s}.\dim_H K=\min\{1,s\}.(6)✦ For the same family with no exact overlaps, the uniquely selected similarity exponent satisfies dim⁡HK=min⁡{1,s*}.\dim_H K=\min\{1,s_*\}.(7)✦ For the same finite nonempty family with 0<|ri|<10<|r_i|<1 and no exact overlaps, K is compact and nonempty,∃!s∈ℝ:s≥0andP(s)=1,∀s∈ℝ,(s≥0andP(s)=1)⇒dim⁡HK=min⁡{1,s}.\begin{aligned} &K\text{ is compact and nonempty},\\ &\exists!\,s\in\mathbb R:\quad s\ge0\ \text{and}\ P(s)=1,\\ &\forall s\in\mathbb R,\quad (s\ge0\ \text{and}\ P(s)=1) \Longrightarrow\dim_H K=\min\{1,s\}. \end{aligned}(8)✦

In the paper, the result reads: for a finite nonempty family of similarities of ℝ\mathbb R with arbitrary real translations and 0<|ri|<10<|r_i|<1, absence of exact overlaps implies dim⁡HK=min⁡{1,s*}\dim_H K=\min\{1,s_*\}.

What Is Checked

The statements (6), (7) and (8) are over finite nonempty alphabets, nonzero strictly contractive signed slopes, arbitrary real translations and injectivity of the map sending each nonempty word to its affine coefficient pair.