General reversible Markov operators and slow spectral projections
DefinitionQVolKLS_Supplement_GeneralWarmSpectrumLet be a Markov kernel on an arbitrary measurable space with stationary probability measure . This definition module supplies the kernel action on complex-valued functions, iterated laws, total variation over measurable sets, -warm starts, and mixing-time predicates without using a default-valued infimum. It also defines positivity and ergodicity of the induced Markov operator.
For a vector field taking values in a complex normed space, its homogeneous second- and fourth-moment quantities are
valued in . A slow spectral projection is represented axiomatically as an orthogonal projection commuting with , annihilating constants, and separating the nonstationary spectrum at ; its tensor extension is characterized by all continuous scalar functionals.
These objects provide the general-measure-space interface used by Theorem 3.1.
Formalization Note Nonintegrable moments are , not the default zero of a Bochner integral. The fast complement is centered by subtracting the -mean, so the excluded stationary eigenvalue cannot make the spectral predicate inconsistent.
import Mathlib
set_option autoImplicit false
namespace QVolKLS.Supplement
open MeasureTheory ProbabilityTheory Filter
open scoped ENNReal
/-!
# General reversible Markov operators, warm starts, and slow spectral projections
R. Zhang, *Improved Upper and Lower Bounds for Quantum Convex-Body Volume Estimation*,
arXiv:2610.04987 (held version, sha256 eee7a45e…), §2.3, pp.22–23 and Theorem 3.1,
pp.24–26.
**Formalization Note.** This interface is deliberately not finite-state. A Markov kernel acts on
scalar functions by integration and on measures by kernel composition. Total variation is the
supremum over measurable sets, and `GeneralMixTimeLe` quantifies over all warm probability starts.
Mathlib does not provide the Borel spectral projection of a general bounded self-adjoint operator.
`IsSlowProjection` therefore characterizes the relevant projection by its operator-theoretic
properties: complex linearity, orthogonal projection, commutation with the Markov operator,
annihilation of constants, and the slow/fast quadratic-form split at `1 - 1/(2t)`. This permits any
orthogonal projection realizing that split (including a choice on the threshold eigenspace); the
canonical projection `E_P([1-1/(2t),1))` is included. `IsTensorExtension` characterizes `Π ⊗ I_A` by
commutation with every continuous scalar functional on the auxiliary Hilbert space.
Positivity, ergodicity, and the projection mapping property are imposed on `L²(π)`, exactly as in
the paper; in particular, a purported projection cannot evade the spectral inequalities by sending
an `L²` function to a nonintegrable representative.
`generalJointL2` and `generalJointL4` use nonnegative integrals and take values in `ℝ≥0∞`, so a
nonintegrable fourth moment is `+∞`, never the Bochner integral's default zero.
-/
/-- The action of a Markov kernel on complex-valued functions. -/
noncomputable def kernelApply {X : Type*} [MeasurableSpace X] (P : Kernel X X)
(f : X → ℂ) (x : X) : ℂ :=
∫ y, f y ∂(P x)
/-- The law after `n` applications of a Markov kernel. -/
noncomputable def kernelLawAfter {X : Type*} [MeasurableSpace X] (P : Kernel X X)
(nu : Measure X) : ℕ → Measure X
| 0 => nu
| n + 1 => P ∘ₘ kernelLawAfter P nu n
/-- Total variation on a general measurable space. -/
noncomputable def generalTV {X : Type*} [MeasurableSpace X] (mu nu : Measure X) : ℝ :=
⨆ A : {A : Set X // MeasurableSet A}, |(mu A.1).toReal - (nu A.1).toReal|
/-- `nu` is an `M`-warm probability start for `pi`. -/
def IsWarmStart {X : Type*} [MeasurableSpace X] (M : ℝ) (nu pi : Measure X) : Prop :=
nu Set.univ = 1 ∧ nu ≤ ENNReal.ofReal M • pi
/-- `τ_mix(η,ν,π;P) ≤ t`, without an `sInf` default-value convention. -/
def GeneralMixTimeLe {X : Type*} [MeasurableSpace X] (eta : ℝ) (nu pi : Measure X)
(P : Kernel X X) (t : ℕ) : Prop :=
∃ j ≤ t, generalTV (kernelLawAfter P nu j) pi ≤ eta
/-- Uniform mixing by time `t` from every two-warm start. -/
def GeneralMixTimeTwoLe {X : Type*} [MeasurableSpace X] (eta : ℝ) (pi : Measure X)
(P : Kernel X X) (t : ℕ) : Prop :=
∀ nu : Measure X, IsWarmStart 2 nu pi → GeneralMixTimeLe eta nu pi P t
/-- Positivity of the Markov operator on `L²(π)`, expressed by its quadratic form. -/
def IsPositiveMarkovOperator {X : Type*} [MeasurableSpace X] (pi : Measure X)
(P : Kernel X X) : Prop :=
∀ f : X → ℂ, MemLp f 2 pi → MemLp (kernelApply P f) 2 pi →
Integrable (fun x ↦ star (f x) * kernelApply P f x) pi →
0 ≤ (∫ x, star (f x) * kernelApply P f x ∂pi).re
/-- Ergodicity: the integrable fixed functions are almost everywhere constant. -/
def IsErgodicMarkovOperator {X : Type*} [MeasurableSpace X] (pi : Measure X)
(P : Kernel X X) : Prop :=
∀ f : X → ℂ, MemLp f 2 pi →
(∀ᵐ x ∂pi, kernelApply P f x = f x) → ∃ c : ℂ, ∀ᵐ x ∂pi, f x = c
/-- The nonnegative `L²(π;H)` norm of a vector field. -/
noncomputable def generalJointL2 {X H : Type*} [MeasurableSpace X] [NormedAddCommGroup H]
(pi : Measure X) (Psi : X → H) : ℝ≥0∞ :=
(∫⁻ x, ENNReal.ofReal ‖Psi x‖ ^ 2 ∂pi) ^ (1 / 2 : ℝ)
/-- The homogeneous vector-valued fourth-moment norm `( ∫ ‖Ψ(x)‖⁴ dπ )¹ᐟ⁴`. -/
noncomputable def generalJointL4 {X H : Type*} [MeasurableSpace X] [NormedAddCommGroup H]
(pi : Measure X) (Psi : X → H) : ℝ≥0∞ :=
(∫⁻ x, ENNReal.ofReal ‖Psi x‖ ^ 4 ∂pi) ^ (1 / 4 : ℝ)
/-- A scalar projection realizing a (possibly smaller) slow spectral subspace above
`1 - 1/(2t)` and below the stationary eigenvalue `1`. -/
def IsSlowProjection {X : Type*} [MeasurableSpace X] (pi : Measure X) (P : Kernel X X)
(t : ℕ) (S : (X → ℂ) → X → ℂ) : Prop :=
(∀ a b : ℂ, ∀ f g, ∀ᵐ x ∂pi,
S (fun y ↦ a * f y + b * g y) x = a * S f x + b * S g x) ∧
(∀ f, MemLp f 2 pi → MemLp (S f) 2 pi) ∧
(∀ f, ∀ᵐ x ∂pi, S (S f) x = S f x) ∧
(∀ f g, Integrable f pi → Integrable g pi → Integrable (S f) pi → Integrable (S g) pi →
∫ x, star (S f x) * g x ∂pi = ∫ x, star (f x) * S g x ∂pi) ∧
(∀ f, ∀ᵐ x ∂pi, S (kernelApply P f) x = kernelApply P (S f) x) ∧
(∀ c : ℂ, ∀ᵐ x ∂pi, S (fun _ ↦ c) x = 0) ∧
(∀ f, Integrable f pi → Integrable (S f) pi → Integrable (kernelApply P (S f)) pi →
Integrable (fun x ↦ star (S f x) * kernelApply P (S f) x) pi →
(1 - 1 / (2 * (t : ℝ))) * ∫ x, ‖S f x‖ ^ 2 ∂pi ≤
(∫ x, star (S f x) * kernelApply P (S f) x ∂pi).re) ∧
(∀ f, Integrable f pi →
let g := fun x ↦ f x - S f x - ∫ y, f y ∂pi
Integrable g pi → Integrable (kernelApply P g) pi →
Integrable (fun x ↦ star (g x) * kernelApply P g x) pi →
(∫ x, star (g x) * kernelApply P g x ∂pi).re ≤
(1 - 1 / (2 * (t : ℝ))) * ∫ x, ‖g x‖ ^ 2 ∂pi)
/-- `SA` is the tensor extension `S ⊗ I_H`, characterized by scalar continuous functionals. -/
def IsTensorExtension {X H : Type*} [NormedAddCommGroup H] [NormedSpace ℂ H]
[MeasurableSpace X] (pi : Measure X)
(S : (X → ℂ) → X → ℂ) (SA : (X → H) → X → H) : Prop :=
∀ (Psi : X → H) (ell : H →L[ℂ] ℂ), ∀ᵐ x ∂pi,
ell (SA Psi x) = S (fun y ↦ ell (Psi y)) x
/-- The general-space Markov data to which the slow-spectrum theorem applies. -/
structure GeneralWarmSpectrum (X : Type*) [MeasurableSpace X] where
pi : Measure X
P : Kernel X X
invariant : Kernel.Invariant P pi
reversible : Kernel.IsReversible P pi
positive : IsPositiveMarkovOperator pi P
ergodic : IsErgodicMarkovOperator pi P
end QVolKLS.Supplement