The Setting
Conditioning on the tags of a finite word gives affine mixtures of a terminal measure. Removing successive letters leaves a family of suffixes, each viewed at the scale inherited from its removed prefix. Their entropy windows compare the entropy at two spatial scales.
A weighted affine iterated function system on a finite alphabet consists of maps and weights satisfying (1), with arbitrary real translations and signed slopes . (1)✦ The complete affine composition of a finite word is the pair , and its word similarity is , defined by (2); the first letter is outermost. (2)
For a tagging map and a symbol , the tag fiber is the set of symbols with the tag of .✦ The fiber mass for a tagging map and a symbol is the total weight of the tag fiber of ; the conditional symbol probability for is the normalized weight specified in (3). (3) For reference and sampled words , the conditional word probability is the product of the conditional symbol probabilities.✦ The conditional tail law associated with a reference word and a measure on is the affine mixture of with weights over , as in (4), where denotes pushforward. (4) Only the specified tags are conditioned on; the future represented by remains unconditioned.
Entropy Between Scales
For a measure on a measurable space , a labeling map and a label , the fiber mass is the mass of the set labeled , and the partition entropy is the negative sum of the fiber masses times their logarithms, as in (5), with . (5) At scale and offset , the shifted grid cell index is the integer-valued function on given in (6). For a measure on , the shifted grid entropy is the partition entropy of , and the averaged grid entropy is its integral over offsets , as in (6). (6) For a measure on and positive scales , the entropy window is the difference in (7). These entropies use natural logarithms; division by gives bits. (7)✦ A measure on is compactly supported if there is a real such that .✦ For a measure on and , the signed dilation is the pushforward of under .
The Claim
Let be a finite word of length . For , write for its first letters and for the remaining suffix.
Definition 1. For a finite reference word , a terminal measure on and , the scaled conditional tail law is the measure in (8). (8)✦ The scaled suffix law associated with a finite reference word and a terminal measure on is given for by (9). (9)✦ For a scaled suffix law , a scale and a positive integer , the suffix entropy window is the quantity in (10). (10)✦
Theorem 1. Let be a compactly supported probability measure, let , and let be a positive integer. For every finite reference word , its suffix windows satisfy (11). (11)✦ If the signed slope is constant on every tag fiber, the windows are nonincreasing as specified in (12). (12)✦
In the paper, the result reads: retaining the environment and constants of §4, with the type shift, the infinite type environment, its conditional tail law, its conditional block law, and the signed prefix dilation, put , so that conditional independence gives ; choose with for every , where , and an integer with , and for an integer put and , obtaining and .
What Is Checked
The bounds concern the scaled suffix laws of a finite weighted affine system with any tagging map, a compactly supported terminal probability measure, a finite reference word, a positive scale and a positive integer scale ratio; monotonicity additionally assumes that symbols with the same tag have the same signed slope.