The Setting
Take finite index sets and and a real matrix with rows indexed by and columns by whose rows are orthonormal, . Run an independent stationary Gaussian autoregressive chain with coefficient in each column , and read it through : with the chains at step , the filtered state at step is . A filter changes one step at a time, each step writing into a fresh column. The question is how likely it is that every filtered state, at every stage of the filter and every step up to a horizon , stays in a ball of radius a constant times , and how fast this survival probability can decay in .
The survival rate of the filtered states comes from combining constraints at finitely many dyadic scales. The library states the survival bound and, before it, two results on these scales: their choice, and the reconstruction of a bound on , for a vector , from bounds at each scale.
Filters and Their States
For a real matrix and weights , the state matrix of and is the matrix (1)✦ the paper’s . The lifted matrix of a real matrix along is the matrix .✦
A filter step is a triple of a real matrix , a vector and a column ; it acts on real matrices by (2)✦ where is the th standard basis vector of . A filter step is good when is invertible and .✦ A finite sequence of filter steps has fresh columns for when, for every step of , the column it writes into is zero in .✦ The run of a finite sequence of filter steps on is the matrix obtained by applying the steps of to in order, first to last; the empty sequence leaves unchanged.✦ An initial segment of a finite sequence of filter steps is a sequence formed by its first steps, in order, possibly none.
Dyadic Scales
The trapezoid cutoff and the size weight are the real functions (3) For a real matrix , weights and an integer , the state matrix at scale is , and and are the matrices given by the functional calculus.
The Autoregressive Chain
The autoregressive row of a real coefficient is the array , over steps and noise indices in , defined by (4)✦ For and independent standard Gaussian noises , is the weight of in the state of the stationary Gaussian autoregressive chain , .
For coefficients , a horizon and a step , the autoregressive evaluation is the real matrix that sends a noise array in to the states at step of these chains, one with coefficient in each coordinate : (5)✦
The Constants
The bounds use absolute constants built from the exponent : the gain constant , the epoch factor , the increment constant , the moment constant , the survival radius and the state rate , given by (6) (7) and the state radius is ✦.
The Claim
Theorem 1 (Finite set of dyadic scales). Let be a finite set and let , for , be positive definite real matrices with eigenvalues , . Then there is a finite set of integers such that (8)✦ and such that every has for some and .
Theorem 2 (Reconstruction bound). Let be a real matrix with , let have for every , and let be a finite set of integers such that every eigenvalue of the state matrix lies in for some . If a vector satisfies (9) then (10)✦
Theorem 3 (Survival of the filtered states). Let be a real matrix with , let be a finite sequence of good filter steps with fresh columns for and with pairwise distinct columns, let , for in an index set , be initial segments of , and let satisfy , and for every . Then there is a real number such that, for every , a standard Gaussian array in and the autoregressive evaluations with coefficients satisfy (11)✦
What Is Checked
The statements (8), (10) and (11) are stated over finite families of positive definite matrices and their eigenvalues; real matrices with , their state matrices (1), the matrices under the functional calculus of the trapezoid cutoff and the size weight (3), and the lifted matrices ; good filter steps (2) with fresh, pairwise distinct columns and their runs along initial segments; the autoregressive evaluations (5) built from the rows (4), under the standard Gaussian measure; and the constants (6) and (7).
About these constructions the library proves that the state matrix of a real matrix with and weights is positive definite.✦