The Setting
A linear filter processes the vectors one at a time, updating a coefficient matrix at each step; a cutoff applied to the state and pulled back to the coefficients gives a cutoff matrix at each time. Lemma 4.1 asks how far the cutoff matrix can move over a stretch of the filter that costs little energy, the change being measured in a Hilbert-Schmidt norm weighted by the coefficients.
The estimate holds in two forms: for a run of filter steps from a coisometry, a matrix with , and for two coisometries joined by a run of moves, which are updates of a state matrix, the identities tying the two ends together being assumed.
Filter Steps and Cutoff Matrices
Fix finite index sets , for the coordinates of the state, and , for the coefficients. A filter step is a triple of a real matrix , a vector and a column , acting on real matrices, with the th standard basis vector of , by (1)✦ A filter step is good when is invertible and .✦ The run of a list of filter steps on a real matrix is the matrix obtained from by applying the steps of by (1), first to last.✦ A list of filter steps is fresh for a real matrix when, for each step of , the column of is zero.✦
For weights on the columns, the state matrix of a real matrix is the matrix (2)✦
The trapezoid cutoff is the continuous piecewise linear function on given by (3)✦ and applied to symmetric matrices by the functional calculus.
The lift of a real matrix along a real matrix is the matrix .✦ The cutoff matrix of under the weights is the lift of along , (4)
The weighted square of a real matrix with columns indexed by against a real matrix is , which for positive semidefinite is the squared Hilbert-Schmidt norm .✦
Moves and Their Energy Cost
A move on is a pair of real matrices, its contraction and its gain , acting on a state by .✦ A move is admissible when in the Loewner order and is positive semidefinite.✦ The final state of a list of moves run from a real matrix is the state reached by applying its moves in order.✦ The pair obtained by following a list of moves from a pair of real matrices is the last pair reached by applying each move of in turn to both, by and .✦ The energy cost of a list of moves run from a real matrix is the sum of the costs of the moves at the states , that the run passes through; the empty list costs .✦
For weights on , the move of a filter step is ; for a list of filter steps, is the list of the moves of its steps, in order.✦
The exponent is ✦, and the gain constant , the increment constant and the energy threshold are (5) The size weight is the function on . The size of a symmetric matrix is .
The Claim
Theorem 1. (a) Let be an coisometry, , let be positive weights on , and let be a list of good filter steps that is fresh for and whose steps write into pairwise distinct columns. Put , and . If for a real , then (6)✦
(b) Let , , be positive definite matrices and an invertible matrix with and . Let and be coisometries with , and let and be positive semidefinite matrices with , , , and . Let be a list of admissible moves with , such that following from ends at a pair whose first matrix is , and with for a real . Then, with , (7)✦
In the paper, the result reads: there are absolute constants and such that, for every recursion of its Proposition 2.2, every positive definite diagonal , every dyadic and all , if then , where , , , and stand for its , , , and .
What Is Checked
Both parts are stated over filter steps with their goodness, runs and freshness, the state matrix, the trapezoid cutoff, the lift and the weighted square, over moves with their admissibility, final states, following and energy cost, the move of a filter step, the size weight, and the exponent with the constants and .