The Setting
The induction uses pairs of distinct binary words whose complete deletion traces admit finite simulations with a controlled likelihood score. Write for the requested deletion probability, with , so that each bit survives with probability . For requested deletion probability , the working rate and its scale parameter are the positive real numbers in (1); the construction theorems call this working rate . (1)
A finite probability law on a finite set is a nonnegative mass function of total mass one.✦ For a finite law on , the expectation of a function is the weighted sum in (2)✦, and the mixture of laws on a finite set , indexed by , is the law on given there.✦ For an element of a finite set, the point mass at is the probability law on that set that assigns mass one to and zero elsewhere.✦ The pushforward of a finite law on under a map is the law on obtained by mixing the point masses at with weights .✦ (2) For laws on a finite set , their overlap and total variation distance are the real numbers defined in (3). (3)
A binary word of length is a function assigning a bit to each zero-based position .✦ A deletion mask for a binary word of length is a function assigning a retention indicator to each position .✦ The retained subsequence of a binary word under a deletion mask is the finite list of the bits of retained by , in their original order.✦ A binary trace is a finite bit list, including its surviving length.✦ For a nonnegative integer , the complete trace space is the set of binary traces of length at most ✦, whose elements are called bounded traces of bound ✦; the deletion mask probability of a mask is the product in (4).✦ (4) For , the deletion mask law on masks of a fixed word length is the probability law with masses .✦ The complete trace law of a binary word with is the probability law on obtained by pushing the deletion mask law forward under .✦ For a real function on finite binary traces, the complete trace expectation of a binary word is the expectation of under its complete trace law, equivalently the recursion in (5), where is the empty trace and prepends bit to word .✦ (5) For binary words , the complete trace overlap is the overlap of their complete trace laws using the common bound in (6).✦ The repeated word , for a binary word and a nonnegative integer , is the concatenation of copies of .✦ (6)
An -approximate trace simulation of a binary word at rate is a finite law on and a decoder satisfying (7); denote this condition by .✦ (7) For a finite law on , a score , and a nonnegative integer half-order , the squared even moment is the nonnegative real number in (8), taking when .✦ (8)
Given a target length , error allowance , score exponent , overlap lower bound , count multiplier , level , and half-order , a concrete level pair in consists of distinct words of common length , finite laws on a common finite set , a decoder , an error , a score , and an integer , subject to (9).✦ (9)
The Schedule and Level Pairs
For an integer , a slack , and , the induction depth is the truncated floor in (10).✦ Here . (10) For a zero-based block index referring to block and a block dimension , the warmup block count , source block dimensions , block gain , and total score exponents are defined in (11). (11) At depth and size , with zero-based scale index , (12) defines the scale slack , scale exponent , exponent product , and backwards target scale . An empty product is one. (12) The error allowance and half moment orders are defined in (13). The even score exponent is , and . (13) The rank Schur constant and primitive period constant determine the initial overlap lower bound and source overlap constant in (14). (14) For the schedule determined by , a source level pair in at level is a concrete level pair with the scheduled parameters in (15).✦ (15)
The Claim
Theorem 1. Fix an integer and . Let be a real sequence tending to infinity, and let satisfy . Put , and evaluate the schedule at , writing and for its error allowance and half moment order. For every sufficiently large , the simultaneous assertion (16) and final-scale assertion (17) hold. (16)✦ (17)✦
In the paper, the result reads: fixing , the section parameter and the remaining section parameters, along every sequence with and , eventually level pairs satisfying Definition 2.2 exist simultaneously for , with common depending only on and satisfying the stated logarithmic bounds, and one constant depending only on that works at every level.
What Is Checked
The assertions are stated over the concrete finite probability constructions in (9), specialized by the source schedule (15). The deletion mask probabilities sum to one over all masks of a fixed length for every real .✦