The Setting
An amplification block acts on two distinct binary words together with finite laws that approximately simulate their complete deletion traces. A level pair also controls a likelihood score and the overlap between two consecutive repetition counts of its left word. The deletion probability requested for the channel is ; each bit survives with probability . The working parameters are defined in (1). The construction statements call their working rate ; here it is denoted by . (1)
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 sequence of retention indicators indexed by the positions of .✦ The retained subsequence keeps the selected bits in their original order.✦ A complete trace of a binary word is a finite list of retained bits of in their original order, including its surviving length; the bounded trace space with bound is the set of all binary lists of lengths zero through . The deletion mask law and its pushforward, the complete trace law , use the weights in (2). For a real-valued function on the complete trace space, write for its expectation under the complete trace law. (2)✦
A finite probability law on a finite set is a nonnegative mass function with total mass one.✦ For a finite probability law on , a real-valued function on , and a nonnegative integer half order , the expectation and probability-weighted squared moment quantity are real quantities defined by (3), with the squared moment quantity equal to one when . At level , let be the positive integer half-order and let be the even score exponent; the source-pair schedule below specifies . (3)
For probability laws and on a common finite outcome space, their overlap and total variation distance are the real quantities defined by (4). Complete trace overlap uses the common bounded trace space with bound equal to the larger source length. The word is the concatenation of copies of .✦ (4)
For a finite probability law on , a decoder , and a binary word , an approximate trace simulation of with error is a decoded law satisfying the test bound in (5). (5)✦
A concrete level pair with parameters is a collection of distinct words of common length , finite laws on a common , a common decoder , a likelihood score , a nonnegative simulation error , and an integer repetition count , subject to the conditions below. Write for the family of these pairs. Both decoded laws satisfy (5) for their respective words, and the remaining conditions are (6). (6)
The Global Schedule
Let be the block parameter and the total number of blocks. Block indices are zero-based: index refers to the paper’s block number and dimension . The warmup block count is an integer, the block gain is a real-valued function of the block dimension , and the total score exponent is a real number at level , with their values defined by (7). (7)
For a size parameter , the scale slack , scale exponents , scale exponent products , and backwards target scales are real quantities defined by (8). The scale index is zero-based, and a subscript on an integer difference denotes truncation at zero. The window exponent for block dimension , zero-based nesting index , and slack is given in (8). (8)
The initial overlap lower bound and overlap constant in (9) use the rank Schur constant and primitive period constant . (9)
The source level-pair family consists of concrete level pairs with the parameters in (10). The natural ceiling means the least nonnegative integer at least . The half moment order at level is the nonnegative integer in (10). (10)
For the block following a pair , set and . The degree cutoff , work size , cancellation order , fitting threshold , leaf-score allowance , score weight , and gap are specified in (11). The source gap lower bound is , so . (11)
The integer window sizes are multiples of four. The extended block window sizes are nonnegative integer multiples of four defined by (12), where is the nonnegative integer floor. (12)✦
The block arithmetic schedule starts from the input length and repetition count . Its state at nesting index records a length , count , copy count, and trial count, initialized with copy count one and trial count zero. Write for the root trial count at and for the local trial count at . The scheduled block trials and successive arithmetic states obey (13). (13)
The Block Budgets
The complete window tree has one leaf at depth zero and ordered children above a tree of depth . Its leaf count and source sparsity count are defined in (14). The latter omits the bottom window factor and uses the nesting parameter . (14)
A window collection index at depth is an index for a local or global collection at a nonleaf vertex of a window tree of depth . The index set is empty at depth zero; at depth it contains the two root tags and pairs of a zero-based child index below with a depth- collection index.✦ Local collections at that depth have trials, global collections have trials, and descendant counts are inherited recursively.✦ The scheduled global counts , hidden metadata tail , and fitting tail are defined by (15); denotes the collection trial count. (15)
The hidden binomial cap and hidden smoothing bound are the numerical quantities in (16). (16)
An ordered role shape of depth and branching bound is an ordered rooted tree with at most children per vertex and a role in assigned to each child; at depth zero there is one empty shape.✦ Let count these shapes.✦ For a sequence of level costs, the binary tree budget and interior label budget are real quantities defined by (17). (17)
For and the level costs in (18), the forced contribution budget is the real quantity defined by that display. (18)✦
For smoothing allowance and fitting threshold , the sharp degree budget and sharp energy budget are real quantities defined by (19). (19)
For the probability-weighted score allowance in (11), the centered moment budget and clipping cost are real quantities defined by (20). (20)
The numerical margins use the moment envelope , weight envelope , cap envelope , tail denominator , and tail exponent in (21). (21)
The bad role factor , forced role factor , smoothing prefactor , and skeleton prefactor are real quantities defined by (22). (22)
The numerical schedule margins at scale mean all the inequalities in (23). Their quantified level index is zero-based. (23)✦
The Claim
Theorem 1. Let , , , and . Let . Assume the numerical schedule margins (23) at and the tail condition (24). (24) Then the next-level source pair family is nonempty, as in (25). (25)✦ For the block parameters and trial schedule of , the budget bounds are (26). (26)✦ Independently, let be nonnegative integers with and , and let be real numbers with , , and . The normalized error bound is (27). (27)✦
In the paper, the result reads: for the global parameters and , a level- pair with common admits such that every preceding pair remains valid after enlarging to , and a new level- pair exists with these same enlarged constants, , and , uniformly in the joint regime, with retained at later levels.
What Is Checked
The statements (25) and (26) concern source level pairs built from actual binary words, finite simulation laws, and the prescribed block arithmetic schedule; (27) concerns real errors and integer length and leaf counts. The deletion mask weights in (2) are nonnegative for rates in ✦, and their sum over all masks is one for every real deletion parameter.✦