Integer covariance kernels and spectral decompositions
DefinitionPrivateContinualCounting_Kraft_Supplement_CovarianceA covariance kernel is a function . Positive semidefiniteness is the nonnegativity of every finite quadratic form. For a finite integer window , a coordinate and arbitrary real coefficients , the coordinate lower bound is
The stationarised-envelope predicate collects symmetry, finite-window positive semidefiniteness, , the coordinate lower bound and
These predicates are used as conclusions of stationarisation and as the inherited input of interpolation, never as assumptions defining an additive process.
A spectral decomposition specifies finite positive measures on the circle and a nonnegative measurable integrable density such that
Here is the published normalized circle measure on , and is supported there and invariant under circle reflection (the endpoints are identified). The singular part is retained. Positivity almost everywhere and reciprocal integrability of are not part of the definition: Lemma 4.3 must prove them. This separates the covariance and spectral interfaces from privacy and from any special envelope.
import Definitions.Def_PrivateContinualCounting_Kraft_Supplement_Privacy
import Definitions.Def_PrivateContinualCounting_Kraft_SpectralConstruction
open MeasureTheory ProbabilityTheory Filter
set_option autoImplicit false
namespace PrivateContinualCounting.Kraft.Supplement
abbrev Kernel := ℤ → ℤ → ℝ
def KernelPSD (Γ : Kernel) : Prop :=
∀ (W : Finset ℤ) (b : ℤ → ℝ), 0 ≤ ∑ i ∈ W, ∑ k ∈ W, b i * Γ i k * b k
def KernelLowerBound (Γ : Kernel) (c : ℝ) : Prop :=
∀ (W : Finset ℤ) (j : ℤ), j ∈ W → ∀ b : ℤ → ℝ,
c * (b j)^2 ≤ ∑ i ∈ W, ∑ k ∈ W, b i * Γ i k * b k
def StationaryEnvelope (Γ : Kernel) (c : ℝ) (B : ℕ → ℝ) : Prop :=
(∀ i k, Γ i k = Γ k i) ∧ KernelPSD Γ ∧
(∀ i k, Γ (i+1) (k+1) = Γ i k) ∧ KernelLowerBound Γ c ∧
∀ (a : ℤ) (w : ℕ), 0 < w →
(∑ i ∈ Finset.range w, ∑ k ∈ Finset.range w,
Γ (a+(i:ℤ)+1) (a+(k:ℤ)+1)) ≤ B w
/-- Reflection on (-π,π], with the endpoints identified. -/
noncomputable def circleReflect (θ : ℝ) : ℝ := if θ = Real.pi then Real.pi else -θ
/-- Actual Herglotz representation and Lebesgue decomposition; no reciprocal
integrability or positivity is assumed. f is a nonnegative measurable representative. -/
def SpectralDecomposition (Γ : Kernel) (ν νs : Measure ℝ) (f : ℝ → ℝ) : Prop :=
IsFiniteMeasure ν ∧ IsFiniteMeasure νs ∧
ν (Set.Ioc (-Real.pi) Real.pi)ᶜ = 0 ∧
ν.map circleReflect = ν ∧
Measurable f ∧ (∀ θ, 0 ≤ f θ) ∧ Integrable f circleMeasure ∧
ν = circleMeasure.withDensity (fun θ => ENNReal.ofReal (f θ)) + νs ∧
νs.MutuallySingular circleMeasure ∧
∀ i k : ℤ, (Γ i k : ℂ) =
∫ θ, Complex.exp (((i-k : ℤ) : ℂ) * (θ : ℂ) * Complex.I) ∂ν
end PrivateContinualCounting.Kraft.Supplement