The Setting
Let be a probabilistic machine that runs in logarithmic space within a polynomial clock, and let be an input. In the configuration graph of on each coin flip is an edge of weight , and the acceptance probability is a value at the initial configuration. The derandomization builds a hierarchy of levels over this graph, in which every vertex has copies at the next level, and its estimators depend on a random environment of three channels, each an affine line in for a prime . The size of a function of environment and vertex is measured by its fourth-root norm, a fourth moment averaged over the vertices that a random channel retains, each weighted by the inverse of the retention rate. The question is what this norm says at a single vertex, the copy of the start at level among the vertices of that level, given that this vertex is retained, and what follows there for the estimator and for the truncation of the hierarchy at level .
The Machine and Its Start
A source machine is a probabilistic Turing machine with states, binary work tapes and input heads, whose transition reads one coin bit per step, together with a space coefficient and a polynomial clock on inputs .✦ A source machine keeps its space bound when, on every input , with every coin tape and up to every time , the number of distinct work cells visited, summed over the tapes, is at most .✦
On an input , the configurations of are numbered as nodes below the node count of on , each node encoding a time, a state, the input head positions and the work tapes; the initial node is the number of the start configuration. The halved edge weight is the probability that one fair coin leads from node to node within the clock.✦ The accepting reward is when node is a legal encoding of a configuration with time equal to the clock and output accept, and otherwise.✦ The acceptance potential is the potential of at depth equal to the clock, where the potential at depth is and each further unit of depth adds to the coin average, over the successors of , of the potential one unit shallower.✦
A stage consists of parameter constants (an iteration count , rank parameters and and a number of copies of each vertex), a level, a finite domain of nodes and a matrix of transition weights.✦ A prize triple consists of a stage, a real vector of rewards paid at the nodes and a real vector meant to be the true acceptance probability at the nodes.✦ The parameter constants satisfy , and , and and are bounded below in terms of and . The raw prize of on under is the prize triple whose stage has level , domain and matrix , whose rewards are and whose values are .✦
A matter record over a prime consists of a source machine , analytic constants , an input word and a bit budget ✦; the analytic constants are parameter constants together with two natural numbers and .✦ The copies of a vertex at the next level are for , and its child is the copy of index . The start of is the child of the initial node at depth , (1)✦
The Hierarchy and Readiness
The matter scene of is the structure on the raw prize , whose recursive rolls compute the ticks used below, with levels, height , roof , gap and walk length , where, with the matter top, (2) The frame stage of the matter scene of at level is the iterate , obtained from by steps of the scene’s recursion on prize triples; denotes the reward vector of and the domain of its stage.
A wares proof for is a certificate that the analytic constants meet a list of technical conditions: bounds on the parameter constants , a constant bounding the test shapes and the Lipschitz constants of the algorithm’s ramp functions, a budget for these bounds that meets the budget requirement at , and the walk length condition for and .✦ The tonic bank of a wares proof is the number , with .✦ The matter record is ready for when (3)✦
Environments and Retention
Write .✦ A rank channel over is a pair of a nonzero direction and an offset ✦; its evaluation at is the point at parameter of its line.✦ An environment is a triple of rank channels over , the pool is the set of all environments, and denotes the average over the pool.✦ The box of level is with side , a vector lying in it when the least nonnegative representatives of its coordinates do; in particular .✦ A vertex of a finite domain is retained at level by a channel when , with read modulo , and the retention rate of level is .✦ An environment polls the start when its first channel retains at level .✦
Means, Norms and the Estimator
A paint is a real function of an environment of the pool and a vertex .✦ The view of is a request for the reward vector at level and source level ; it is a vertex request, that is, a request for a paint, and it asks for no detector.✦ The vertex axis of the view of is the estimation axis with domain , size , and source and destination levels .✦ The conditional mean vector of a paint along has at the conditional average (4)✦
A monotone linear meter on a set is a functional on the real functions on that set that is monotone, additive and homogeneous.✦ The vertex meter of is the meter on pairs of an environment and a vertex that sums over the vertices of retained at level by the first channel, averages over environments and scales by .✦ The vertex norm of a paint is the fourth root of the vertex meter of , (5)✦
A table request asks for a family of tables and a vertex request asks for a paint, and a tick of the scene answers every request of either kind. The forecast tick at a time is the tick computed at by the scene’s recursive rolls from the ticks at earlier times.✦ The estimator of , also called its oracle view, is the paint that the forecast tick at the matter top assigns to the view.✦ For a real function on the pool, the jury value is the conditional mean (4) at of the paint equal to at every vertex.✦
The Claim
Theorem 1. Let be a matter record over a prime , with source machine , analytic constants , input , bit budget and start , and let be a wares proof for . If is ready for and keeps its space bound, then every paint satisfies (6)✦ and the estimator satisfies (7)✦ If is ready for , then (8)✦ Every real function on the pool with at every environment that polls the start satisfies (9)✦
In the paper, the result reads: for with , every specified and every satisfy and , with , and in the roles of , and , in the role of the start and in that of its retention. Here is the paper’s retention indicator, and , , and are the paper’s own symbols, not those of the page.
What Is Checked
The claim is stated over a matter record with its source machine and the space bound of , a wares proof with the readiness conditions (3), the start (1), the conditional mean (4) and the vertex norm (5) along the vertex axis of the view, the reward vector of the iterate , the acceptance potential of the raw prize, the estimator and the jury value . The library also proves that the finite average is monotone, additive and homogeneous, which makes the average over environments a monotone linear meter.