The Setting
At a cut in a finite symbolic address, the remaining similarities act on the self-similar measure. The quantity of interest is the change in entropy between two fixed spatial scales when the cut moves across a block and the future types are averaged.
A weighted affine iterated function system (IFS) on a finite alphabet consists of similarities with signed slopes, real translations and strictly positive probability weights, as in (1): (1) A probability measure on is self-similar for when it satisfies (2): (2)✦ A probability measure on is compactly supported when for some .
For a finite word , its word similarity and complete affine map are specified in (3); composition follows address order: (3) A collision of complete affine maps is equality of their signed slopes and translations. The word probability , symbol depth and word contraction depth are given by (4); denotes the accumulated depth of the first at most letters of a finite word : (4)
Conditional Tails
A finite probability law on consists of a finite support and masses for , with .✦ For a finite source , real weights and a map , the fiber masses are given by (5): (5)✦ For a real-valued map on a finite source with strictly positive normalized weights , the observable distribution of is the finite probability law with fiber masses given by (5).
For a tag map on a finite source , strictly positive weights on and a source point , the tag fiber , its mass and the conditional symbol probability for are defined in (6): (6) For a real-valued observable on a finite source with strictly positive weights , a tag map and a source point , the conditional observable distribution of given is the finitely supported probability law on obtained from the values of on with weights .
Choose a type map for the IFS and use . The word tag sequence of a finite word over is the sequence specified by .✦ For a finite word over , the type-conditioned translation law is the conditional observable distribution of on words of the same length as , given and using the word weights . For a word over , the conditional word probability of a word of the same length is the product in (7), the conditional tail is the probability measure given by its weighted sum, and is the probability measure obtained by dilating by the signed real factor : (7) Only the types in are conditioned; the continuation with law remains unconditioned.
Entropy Windows and Pair Capacity
Let be the real mass of the fiber for a measure on a measurable space and a map .✦ The partition entropy of a map with respect to a measure on is the quantity in (8), with . For measures on , with spatial scale and grid shift , is the shifted grid cell index, its grid entropy and its averaged grid entropy: (8) For a measure on and spatial scales and , the entropy window and one-step dyadic deficit are the quantities defined by (9): (9) These entropies use natural logarithms; division by gives their values in bits.
The set of fair pairs and the incident pair mass at an atom are given in (10) for a finite law , an integer and weights assigned to its fair pairs. A feasible pair weighting for a finite law in band is a family of weights on such that for every pair and for every atom .✦ For a finite law on and an integer , the fair-pair capacity is the supremum of total pair weight over feasible pair weightings in band , as in (10): (10)
Retained Windows
Write for the first at most letters of , and for its remaining suffix. For a finite word , a cut , a scale and a natural factor , the scaled suffix law is the probability measure specified in (11), and its suffix entropy window is the quantity specified there: (11)
Fix , and integers , and put . The retained cuts form the set in (12): (12)✦ For a positive integer , put . Let be the first at most letters of . The retained capacity and retained window drop are defined by (13): (13) The Bernoulli word average of a real function on words of length is the weighted sum in (14): (14)
The Claim
Theorem 1. Let be a weighted affine IFS, a type map and a compactly supported self-similar probability measure for . Suppose symbols with the same type have the same signed slope. Let satisfy for every , and let satisfy and .
For , , and , the conditional gain satisfies (15), with and with denoting the concatenation of and a word of length : (15)✦ For with , and , the expected retained gain satisfies (16): (16)✦
In the paper, the result reads: with the preceding suffix laws conditioned on an entire infinite type environment , with denoting its shift, the sigma algebra of conditioning and the corresponding type-conditioned translation law for a block of length , positive , and suffix windows , for every integer , and , .
What Is Checked
The statements concern finite type partitions of weighted affine systems with constant signed slope on each type, compactly supported self-similar probability measures and finite Bernoulli word averages. Every word has positive Bernoulli weight✦, and the normalized weights on the tag fiber of every source point sum to one.✦