OctoInformBeta

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

Corollary 6.2: Equal-length composition injectivity also fixes every map-walk entropy and its infimum rate

Symbol entropy determines 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

The question is how coincidences between affine compositions affect the entropy used to express the dimension of a self-similar measure.

A weighted affine iterated function system S=(Φ,p)S=(\Phi,p) on a finite nonempty alphabet II consists of real similarities Φ=(φi)i∈I\Phi=(\varphi_i)_{i\in I} and strictly positive normalized weights p=(pi)i∈Ip=(p_i)_{i\in I} satisfying (1); the slopes may have either sign and the translations are arbitrary real numbers. φi(x)=rix+ti,0<|ri|<1,pi>0,∑i∈Ipi=1.\varphi_i(x)=r_i x+t_i,\qquad 0<|r_i|<1,\qquad p_i>0,\qquad \sum_{i\in I}p_i=1.(1)✦ A measure μ\mu on ℝ\mathbb R is self-similar for SS when it satisfies (2), where (φi)*μ(\varphi_i)_*\mu is the pushforward of μ\mu under φi\varphi_i. μ=∑i∈Ipi(φi)*μ.\mu=\sum_{i\in I}p_i(\varphi_i)_*\mu.(2) The lower Hausdorff dimension dim⁡Hμ\dim_H\mu of a Borel measure μ\mu on ℝ\mathbb R is the infimum in [0,∞][0,\infty] specified by (3). dim⁡Hμ=inf⁡{dim⁡HE:E⊆ℝ is Borel and μ(E)>0}.\dim_H\mu =\inf\{\dim_H E:E\subseteq\mathbb R\text{ is Borel and }\mu(E)>0\}.(3)✦

Compositions and Entropies

The complete affine map gw=(Rw,Tw)g_w=(R_w,T_w) of a finite address w=(i1,…,in)w=(i_1,\ldots,i_n) is the signed-slope and translation pair of the composition in (4); the empty address represents the identity. φw=φi1∘⋯∘φin,φw(x)=Rwx+Tw,g⌀=(1,0).\varphi_w=\varphi_{i_1}\circ\cdots\circ\varphi_{i_n},\qquad \varphi_w(x)=R_wx+T_w,\qquad g_{\varnothing}=(1,0).(4)✦ The composition map at length nn is the function w↦gww\mapsto g_w on InI^n. A weighted affine system SS has no equal-length exact overlaps when the composition map w↦gww\mapsto g_w on InI^n is injective for every n≥0n\geq0.✦ A system SS has no exact overlaps when the function w↦gww\mapsto g_w is injective on all finite addresses, including the empty address and addresses of different lengths.✦

The word probability pwp_w of an address w=(i1,…,in)∈Inw=(i_1,\ldots,i_n)\in I^n is the product in (5). pw=∏j=1npij.p_w=\prod_{j=1}^{n}p_{i_j}.(5)✦ The attainable affine maps ℳn\mathcal M_n form the finite set of complete maps of length nn addresses✦, and the random walk mass mn(g)m_n(g) is the probability that independent symbol choices with weights pp compose to gg, as in (6). ℳn={gw:w∈In},mn(g)=∑w∈Ingw=gpw.\mathcal M_n=\{g_w:w\in I^n\},\qquad m_n(g)=\sum_{\substack{w\in I^n\\g_w=g}}p_w.(6) The map-walk entropy HnH_n is the base-two Shannon entropy of the masses on ℳn\mathcal M_n✦, and its entropy rate hh is the infimum of the positive-length normalized entropies in (7). Hn=−1ln⁡2∑g∈ℳnmn(g)ln⁡mn(g),h=infn≥1Hnn.H_n=-\frac{1}{\ln2}\sum_{g\in\mathcal M_n}m_n(g)\ln m_n(g), \qquad h=\inf_{n\geq1}\frac{H_n}{n}.(7)

The symbol entropy H(p)H(p) is the Shannon entropy of the weights with natural logarithms✦, and the natural Lyapunov exponent χnat\chi_{\mathrm{nat}} is their mean logarithmic contraction, as defined in (8).✦ The positive Lyapunov exponent χ\chi uses base-two logarithms. H(p)=−∑i∈Ipiln⁡pi,χnat=−∑i∈Ipiln⁡|ri|,χ=χnatln⁡2.H(p)=-\sum_{i\in I}p_i\ln p_i,\qquad \chi_{\mathrm{nat}}=-\sum_{i\in I}p_i\ln|r_i|, \qquad \chi=\frac{\chi_{\mathrm{nat}}}{\ln2}.(8)

The Claim

Theorem 1. Let SS be a weighted affine system on a finite nonempty alphabet with no equal-length exact overlaps. For every integer n≥0n\geq0, Hn=nH(p)ln⁡2.H_n=n\frac{H(p)}{\ln2}.(9)✦ Its entropy rate satisfies h=H(p)ln⁡2.h=\frac{H(p)}{\ln2}.(10)✦ Let SS be a weighted affine system on a finite nonempty alphabet with no exact overlaps, and let μ\mu be a self-similar probability measure for SS. Then dim⁡Hμ=min⁡{1,H(p)χnat}.\dim_H\mu=\min\left\{1,\frac{H(p)}{\chi_{\mathrm{nat}}}\right\}.(11)✦

In the paper, the result reads: for a finite nonempty indexed family Φ\Phi of real similarities with 0<|ri|<10<|r_i|<1 and no exact overlaps, and every strictly positive probability vector pp, the self-similar measure μΦ,p\mu_{\Phi,p} has dim⁡HμΦ,p=min⁡{1,H(p)/χnat}\dim_H\mu_{\Phi,p}=\min\{1,H(p)/\chi_{\mathrm{nat}}\}.

What Is Checked

The identities (9) and (10) concern finite nonempty weighted affine systems with injective composition maps at every fixed length; (11) concerns systems with injective composition on all finite addresses and probability measures satisfying (2).