OctoInformBeta

The entropy-rate dimension formula for self-similar measures

Consider a finite nonempty indexed family Φ=(φi)i∈Λ\Phi=(\varphi_i)_{i\in\Lambda} of affine contractions φi(x)=rix+ti\varphi_i(x)=r_ix+t_i on ℝ\mathbb R, with 0<|ri|<10<|r_i|<1. Fix a strictly positive probability vector p=(pi)i∈Λp=(p_i)_{i\in\Lambda}. A Borel probability measure μ\mu is self-similar if μ=∑i∈Λpi(φi)*μ\mu=\sum_{i\in\Lambda}p_i(\varphi_i)_*\mu. An exact overlap is an equality between the complete affine maps obtained by composing two distinct words of the same positive length. For independent symbols IjI_j with law pp, let Gn=φI1∘⋯∘φInG_n=\varphi_{I_1}\circ\cdots\circ\varphi_{I_n} be the random composed affine map. Write HH for Shannon entropy with logarithms to base two. The random-walk entropy rate is h=inf⁡n≥1H(Gn)/nh=\inf_{n\ge1}H(G_n)/n. The Lyapunov exponent is χ=−∑ipilog⁡2|ri|\chi=-\sum_i p_i\log_2|r_i|. The lower Hausdorff dimension of a Borel probability measure ν\nu is dim⁡Hν=inf⁡{dim⁡HE:E⊆ℝ Borel,ν(E)>0}\dim_H\nu=\inf\{\dim_H E:E\subseteq\mathbb R\text{ Borel},\ \nu(E)>0\}. The dimension question is whether h/χh/\chi determines dim⁡Hμ\dim_H\mu without separation assumptions.

Feng and Hu proved that lim⁡s↓0log⁡μ([x−s,x+s])/log⁡s\lim_{s\downarrow0}\log\mu([x-s,x+s])/\log s is constant for μ\mu-almost every xx, without a separation assumption. Rapaport proved dim⁡Hμ=min⁡{1,H(p)/χ}\dim_H\mu=\min\{1,H(p)/\chi\} for algebraic contraction ratios when there are no exact overlaps. Rapaport and Varjú proved dim⁡Hμ=min⁡{1,h/χ}\dim_H\mu=\min\{1,h/\chi\} for rational translations and a positive common contraction ratio, including exact overlaps.

The preprint proves dim⁡Hμ=min⁡{1,h/χ}\dim_H\mu=\min\{1,h/\chi\} for every such finite system and positive probability vector, allowing unequal contraction ratios of either sign, exact overlaps and repeated generators.

The Lean library SelfSimilar takes the self-similarity equation for the probability measure as a hypothesis of its main theorem. Lean has checked every result of the Lean library. The Lean library is therefore the reference for the statements on these pages. Each result is stated as in the Lean library and linked to the Lean declaration behind it. Where the preprint states a result differently, the statement of the Lean library is given with the preprint’s beside it. The preprint serves only as a guide to the order and names of the results.

The Lean library: SelfSimilar

SelfSimilar is a Lean library: 25 items, in six groups. Each has a page written by OctoInform, listed below.

§1. Introduction 3 items
  1. §1 (setting): affine systems, the map walk and measure dimension Affine systems and their entropy rates.
  2. Theorem 1.1 Dimension from the entropy of affine compositions.
  3. §1 (setting): exact dimensionality and the upper bound used in the proof Self-similar measures have constant local dimension.
§2. Entropy between two scales 5 items
  1. §2 (setting): grid entropy, averaged entropy and deficits Entropy from grids to scale windows.
  2. Lemma 2.1 Averaged entropy respects affine scale changes.
  3. Lemma 2.2 Radial dimension bounds entropy growth.
  4. Lemma 2.3 Entropy slopes below one force uniform deficits.
  5. Lemma 2.4 Intermediate deficits control two-point entropy gains.
§3. Pair mass below a fixed scale 2 items
  1. Definition 3.1 Maximum pair mass at a dyadic scale.
  2. Lemma 3.2 Pair capacities bound fixed-grid entropy loss.
§4. Conditioning on block types 3 items
  1. §4 (setting): block types and conditional translation laws Conditioning translations on block types.
  2. Lemma 4.1 Block types control the loss of translation entropy.
  3. Proposition 4.2 Retained entropy loss forces pair capacity.
§5. Disjoint windows of entropy gain 5 items
  1. §5 (setting): suffix laws and entropy windows Finite conditional suffixes give decreasing entropy windows.
  2. Lemma 5.1 Entropy drops control dyadic pair capacity.
  3. §5 (setting): separated bands and retention Separated bands from an eventual pair budget.
  4. Lemma 5.2 Retained windows share a finite entropy budget.
  5. §5 (proof of Theorem 1.1): summing targets and completing the contradiction Entropy budgets bound the coding exponent below.
§6. Overlap conventions and dimension of the attractor 7 items
  1. §6 (setting): overlap conventions The overlap conventions are equivalent.
  2. Lemma 6.1 Exact overlaps occur at equal word lengths.
  3. Corollary 6.2 Symbol entropy determines dimension without exact overlaps.
  4. §6 (setting): the attractor and similarity dimension Coding the attractor and defining similarity dimension.
  5. Corollary 6.3 Attractor dimension without exact overlaps.
  6. Corollary 6.4 Dimension for homogeneous systems without exact overlaps.
  7. Corollary 6.5 Dimension for three similarities without exact overlaps.

No item matches.