The Setting
Conditioning on block types records the symbol counts in successive blocks of an affine random walk. The question is how much randomness remains in the translation after these counts are known.
A weighted affine iterated function system on a finite alphabet consists of maps with signed slopes , real translations , and weights satisfying (1): (1)✦ A self-similar probability measure for this system is a probability measure on satisfying (2): (2)✦
For a word , its complete affine map and word probability are the affine map and real number given by (3); the empty word has complete map : (3) The random complete map is obtained by choosing with probability . The attainable set of the random complete map is the set of complete affine maps of words of length , and its mass function assigns each map its total word probability, as in (4); collisions require equality of both signed slope and translation: (4) The map-walk entropy , its rate , and the positive Lyapunov exponent use base-two logarithms, as in (5): (5)
The block type of a word of length is the sequence of symbol-count vectors in its consecutive blocks of length , defined by (6).✦ Let be a random word of length , chosen with word probabilities , and write for its block type and . (6)
Conditional Laws and Entropies
For real weights on a finite set and an observable , the observable entropy is the Shannon entropy in nats of its fiber masses; the weights used for the conditional probability laws below are strictly positive.✦ The conditional observable entropy of an observable given a tag is the difference between the joint observable entropy and the tag entropy, as in (7): (7) The weights need not sum to one. For random variables, uses their joint law, and denotes entropy in bits. For , the tag fiber is the subset of with tag ✦, its mass is the sum of its weights✦, and is the normalized weight function on given in (8): (8) A finite probability law on has finite support with strictly positive point masses summing to one.✦ For a real observable , its conditional observable law and the entropy in bits of a finite law are given by (9), where denotes the probability measure concentrated at : (9) Zero masses contribute zero to every entropy sum. The observable distribution of a real observable on a finite probability space is the finitely supported probability law whose mass at each value is the sum of the weights of all source points taking that value.✦
Let be a finite type alphabet of size , and let be a tag. For a word , its word of types is the sequence defined by .✦ Applying the conditional law in (9) with , weights , tag and observable defines the type-conditioned translation law .✦ For a real function on and a random word of length with word probabilities , write . This conditioning fixes only the first types; future symbols remain unconditioned.
For a measure and a discrete observable , the partition entropy is the entropy in nats of its real fiber masses, as in (10): (10)✦ For real parameters , the shifted grid index is the integer-valued function on given in (11), and the grid entropy is its partition entropy under a measure on ; the entropy bounds use positive scales. Here is measured in bits, whereas and the averaged grid entropy are in nats. (11)
A measure on is compactly supported when some satisfies .✦
For a word and real parameters , the contraction depth and fine scale are the real numbers given in (12). The entropy loss of a word at a scale is the difference between Shannon and grid entropy of in nats. The retained loss is the sum of the entropy losses , weighted by , over words whose contraction depth is at most . (12)
The Claim
Theorem 1. For every weighted affine system, the block translation entropy satisfies (13) for integers : (13)✦ For integers , the entropy-rate bound is (14): (14)✦ For the positive finite weights and observables of (7), conditional entropy is the mean in (15): (15)✦
Suppose the tag has constant signed slope on each type: whenever . For every integer , the mean translation entropy satisfies (16): (16)✦ If is a compactly supported self-similar probability measure, then for every and the mean grid bound is (17): (17)✦
For retained loss, let be a self-similar probability measure and let , , and . Suppose the averaged entropy bound and strict gap in (18) hold; is the entropy coefficient: (18) There are and an integer such that (19) holds: (19)✦
In the paper, under the system, deficit hypothesis, parameters and conditional laws of the preceding setting, with block length , block entropy parameter , retained events and scales , the result reads: for every and , and , where is the first types and the laws condition on the entire infinite type environment , and there are and an integer such that for every .
What Is Checked
The statements concern finite weighted affine systems and their conditional observable laws, with compact support for the self-similar measure in the grid bound and the averaged entropy and gap assumptions for retained loss.
Every word has strictly positive probability✦, and for strictly positive weights on a finite set the normalized weights on each tag fiber sum to one.✦