Chapter Solovay
DefinitionChapterSolovaydefinitionsformalizationtimepiece
Formal definitions for the timepiece Lean 4 formalization (source chapter BookProof/ChapterSolovay.lean): generated def bundle for ChapterSolovay. See BookProof/ChapterSolovay.lean for full context.
Definition code
import Mathlib
import Mathlib.Analysis.InnerProductSpace.Completion
import Mathlib.Topology.UniformSpace.Completion
import Definitions.Def_PhysMehler
/-!
# Chapter S — Solovay-Hilbert Decidability and the Kopperman Tail
This file formalizes the decidability architecture of the decoupled
Kopperman-Solovay framework (see `RandomMap2.md` Phase 1 and Phase 8.3).
Sections:
* S.1 — The Solovay-Hilbert space as a completion (proper construction)
* S.2 — `dependsOnlyOnHead` decidability argument
* S.3 — The uniform Mehler measure on the infinite-dimensional hypersphere
* S.4 — No Gödelian self-reference in the Solovay-Hilbert space
All theorems are `sorry`-free and `axiom`-free (no `EXTERNAL` hypothesis).
-/
open MeasureTheory ProbabilityTheory
open scoped ENNReal Filter Topology
/-! ## Inlined Kopperman–Solovay scaffolding (was `RandomMap.RandomMap2`)
Verbatim from `timepiece331/RandomMap/RandomMap2.lean`. The platform compiles a
bundle against `Mathlib` + `Definitions.*` + `Theorems.*` only, so
`import RandomMap.RandomMap2` resolves nowhere and the server elaborated this
whole payload in an EMPTY import environment (attempt 1: `unknown namespace
`MeasureTheory``, `unknown namespace `ENNReal``, `Unknown identifier `ℕ``).
Everything below needs only the published `Def_PhysMehler` bundle, whose
`Substrate`/`rcpPriorOnSubstrate` live in the root namespace `PhysMehler`. -/
open PhysMehler
noncomputable section
instance substrateMeasurableSpace : MeasurableSpace Substrate := borel _
instance substrateBorelSpace : BorelSpace Substrate := ⟨rfl⟩
/-- The infinite, unknown tail of the inner wave-function. -/
abbrev InnerTail := Substrate
/-- The uniform probability measure over the infinite tail. -/
def tailMeasure : Measure InnerTail := rcpPriorOnSubstrate
instance : IsProbabilityMeasure tailMeasure := rcpPriorOnSubstrate_isProb
/-- The finite known components of the inner wave-function. -/
abbrev InnerHead (N : ℕ) := Fin N → ℝ
/-- The total sample space of inner wave-functions. -/
abbrev InnerSpace (N : ℕ) := InnerHead N × InnerTail
/-- The total probability measure, given an arbitrary law on the head. -/
noncomputable def stateMeasure (N : ℕ) (headDist : Measure (InnerHead N))
[IsProbabilityMeasure headDist] : Measure (InnerSpace N) :=
headDist.prod tailMeasure
instance (N : ℕ) (headDist : Measure (InnerHead N)) [IsProbabilityMeasure headDist] :
IsProbabilityMeasure (stateMeasure N headDist) := by
dsimp [stateMeasure]; infer_instance
/-- A macroscopic observable function depends only on the finite head. -/
def dependsOnlyOnHead {N : ℕ} (f : InnerSpace N → ℂ) : Prop :=
∃ g : InnerHead N → ℂ, f = g ∘ Prod.fst
/-- The Solovay space of outer wave-functions: `L²` over the product measure. -/
abbrev OuterWaveFunction (N : ℕ) (headDist : Measure (InnerHead N))
[IsProbabilityMeasure headDist] := Lp ℂ 2 (stateMeasure N headDist)
end
namespace BookProof.ChapterSolovay
/-! ## S.1 — The Solovay-Hilbert space as a proper completion -/
/-- **S.1.1 — The Solovay-Hilbert space: the completion of `OuterWaveFunction`.
The completion adds the missing metric structure, making this a genuine
Hilbert space. This is the explicit construction that was previously
a placeholder (`RandomMap2.md` Phase 8.3).
We use the root-level definitions from `RandomMap2` to avoid shadowing. -/
noncomputable abbrev SolovayHilbertSpace (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist] : Type :=
_root_.UniformSpace.Completion (_root_.OuterWaveFunction N headDist)
noncomputable instance (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist] : InnerProductSpace ℂ (SolovayHilbertSpace N headDist) :=
UniformSpace.Completion.innerProductSpace
noncomputable instance (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist] : CompleteSpace (SolovayHilbertSpace N headDist) :=
UniformSpace.Completion.completeSpace _
/-- The canonical embedding of outer wave-functions into the Solovay-Hilbert space. -/
noncomputable def toSolovay (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist]
(Ψ : _root_.OuterWaveFunction N headDist) : SolovayHilbertSpace N headDist :=
(Ψ : SolovayHilbertSpace N headDist)
/-- The embedding is an isometric linear map. -/
theorem toSolovay_inner (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist]
(Ψ Φ : _root_.OuterWaveFunction N headDist) :
inner ℂ (toSolovay N headDist Ψ) (toSolovay N headDist Φ) = inner ℂ Ψ Φ := by
simp [toSolovay]
/-- The embedding preserves the `L²` norm. -/
theorem toSolovay_norm (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist]
(Ψ : _root_.OuterWaveFunction N headDist) :
‖toSolovay N headDist Ψ‖ = ‖Ψ‖ := by
simp [toSolovay]
/-! ## S.2 — `dependsOnlyOnHead` decidability argument -/
/-- The inner product of two cylindrical (head-only) functions reduces to
a finite-dimensional integral — the fundamental decidability corollary.
Note: `inner ℂ Ψ₁ Ψ₂ = ∫ star(Ψ₁) * Ψ₂` by the L2 inner product convention,
so the RHS uses `star(Ψ₁ (x,0)) * Ψ₂ (x,0)` to match. -/
theorem inner_reduces_to_head (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist]
(Ψ₁ Ψ₂ : _root_.OuterWaveFunction N headDist)
(hcyl₁ : _root_.dependsOnlyOnHead (Ψ₁ : _root_.InnerSpace N → ℂ))
(hcyl₂ : _root_.dependsOnlyOnHead (Ψ₂ : _root_.InnerSpace N → ℂ)) :
inner ℂ Ψ₁ Ψ₂ = ∫ x : _root_.InnerHead N, star (Ψ₁ (x, 0)) * (Ψ₂ (x, 0)) ∂headDist := by
rcases hcyl₁ with ⟨g₁', hg₁⟩
rcases hcyl₂ with ⟨g₂', hg₂⟩
have h_tail_ne_zero : _root_.tailMeasure ≠ 0 := by
have h_univ_one : _root_.tailMeasure Set.univ = 1 := measure_univ
intro h_eq
have h_univ_zero : _root_.tailMeasure Set.univ = 0 := by simp [h_eq]
have h_eq_one_zero : (1 : ENNReal) = 0 := by
rw [← h_univ_one, h_univ_zero]
norm_num at h_eq_one_zero
have h_map_fst : Measure.map Prod.fst (headDist.prod _root_.tailMeasure) = headDist := by
rw [MeasureTheory.Measure.map_fst_prod, measure_univ, one_smul]
have h_ae₁ : AEStronglyMeasurable (Ψ₁ : _root_.InnerSpace N → ℂ)
(headDist.prod _root_.tailMeasure) :=
Lp.aestronglyMeasurable _
have h_ae₂ : AEStronglyMeasurable (Ψ₂ : _root_.InnerSpace N → ℂ)
(headDist.prod _root_.tailMeasure) :=
Lp.aestronglyMeasurable _
have h_ae_comp₁ : AEStronglyMeasurable (g₁' ∘ Prod.fst) (headDist.prod _root_.tailMeasure) := by
rw [← hg₁]; exact h_ae₁
have h_ae_comp₂ : AEStronglyMeasurable (g₂' ∘ Prod.fst) (headDist.prod _root_.tailMeasure) := by
rw [← hg₂]; exact h_ae₂
have h_ae_g₁ : AEStronglyMeasurable g₁' headDist :=
h_ae_comp₁.of_comp_fst h_tail_ne_zero
have h_ae_g₂ : AEStronglyMeasurable g₂' headDist :=
h_ae_comp₂.of_comp_fst h_tail_ne_zero
have h_mem_comp₁ : MemLp (g₁' ∘ Prod.fst) 2 (headDist.prod _root_.tailMeasure) := by
rw [← hg₁]; exact Lp.memLp _
have h_mem_comp₂ : MemLp (g₂' ∘ Prod.fst) 2 (headDist.prod _root_.tailMeasure) := by
rw [← hg₂]; exact Lp.memLp _
have h_mem_g₁ : MemLp g₁' 2 headDist := by
have h_ae_map : AEStronglyMeasurable g₁' (Measure.map Prod.fst
(headDist.prod _root_.tailMeasure)) := by
rw [h_map_fst]; exact h_ae_g₁
have h_meas_fst : AEMeasurable Prod.fst
(headDist.prod _root_.tailMeasure) :=
measurable_fst.aemeasurable
have h_equiv := MeasureTheory.memLp_map_measure_iff (p := 2) h_ae_map h_meas_fst
rw [h_map_fst] at h_equiv
exact h_equiv.mpr h_mem_comp₁
have h_mem_g₂ : MemLp g₂' 2 headDist := by
have h_ae_map : AEStronglyMeasurable g₂' (Measure.map Prod.fst
(headDist.prod _root_.tailMeasure)) := by
rw [h_map_fst]; exact h_ae_g₂
have h_meas_fst : AEMeasurable Prod.fst
(headDist.prod _root_.tailMeasure) :=
measurable_fst.aemeasurable
have h_equiv := MeasureTheory.memLp_map_measure_iff (p := 2) h_ae_map h_meas_fst
rw [h_map_fst] at h_equiv
exact h_equiv.mpr h_mem_comp₂
let g₁ : Lp ℂ 2 headDist := h_mem_g₂.toLp g₂'
let g₂ : Lp ℂ 2 headDist := h_mem_g₁.toLp g₁'
have h_inner_eq : inner ℂ Ψ₁ Ψ₂ = ∫ z : _root_.InnerSpace N,
(g₂' z.1) * star (g₁' z.1) ∂(headDist.prod _root_.tailMeasure) := by
rw [MeasureTheory.L2.inner_def (𝕜 := ℂ) Ψ₁ Ψ₂]
simp_rw [RCLike.inner_apply]
refine integral_congr_ae ?_
filter_upwards with z
simp [hg₁, hg₂]
have h_fubini_eq : ∫ z : _root_.InnerSpace N,
(g₂' z.1) * star (g₁' z.1) ∂(headDist.prod _root_.tailMeasure) =
∫ x, (g₂' x) * star (g₁' x) ∂headDist := by
have h_int_comp : Integrable (fun z : _root_.InnerSpace N => (g₂' z.1) * star (g₁' z.1))
(headDist.prod _root_.tailMeasure) := by
have h_int_inner : Integrable (fun z : _root_.InnerSpace N =>
((Ψ₂ : _root_.InnerSpace N → ℂ) z) * star ((Ψ₁ : _root_.InnerSpace N → ℂ) z))
(headDist.prod _root_.tailMeasure) := by
have h := MeasureTheory.L2.integrable_inner (𝕜 := ℂ) Ψ₁ Ψ₂
simpa [RCLike.inner_apply, _root_.stateMeasure] using h
have h_eq : (fun z : _root_.InnerSpace N => (g₂' z.1) * star (g₁' z.1))
=ᵐ[headDist.prod _root_.tailMeasure]
(fun z =>
((Ψ₂ : _root_.InnerSpace N → ℂ) z) * star ((Ψ₁ : _root_.InnerSpace N → ℂ) z)) := by
filter_upwards with z
simp [hg₁, hg₂]
exact (integrable_congr h_eq).mpr h_int_inner
have h_fubini : ∫ z : _root_.InnerSpace N,
(g₂' z.1) * star (g₁' z.1) ∂(headDist.prod _root_.tailMeasure) =
∫ y, ∫ x, (g₂' x) * star (g₁' x) ∂headDist ∂_root_.tailMeasure := by
exact integral_prod_symm
(fun z : _root_.InnerHead N × _root_.InnerTail => (g₂' z.1) * star (g₁' z.1)) h_int_comp
rw [h_fubini]
simp [integral_const]
have h_g_eq : ∫ x, (g₂' x) * star (g₁' x) ∂headDist = ∫ x, g₁ x * star (g₂ x) ∂headDist := by
dsimp [g₁, g₂]
refine integral_congr_ae ?_
filter_upwards [MemLp.coeFn_toLp h_mem_g₂, MemLp.coeFn_toLp h_mem_g₁] with x h₂ h₁
simp [h₂, h₁]
calc
inner ℂ Ψ₁ Ψ₂ = ∫ x, g₁ x * star (g₂ x) ∂headDist := by
rw [h_inner_eq, h_fubini_eq, h_g_eq]
_ = ∫ x, (g₂' x) * star (g₁' x) ∂headDist := by rw [h_g_eq]
_ = ∫ x, star (g₁' x) * g₂' x ∂headDist := by
refine integral_congr_ae ?_
filter_upwards with x
ring
_ = ∫ x : _root_.InnerHead N, star (Ψ₁ (x, 0)) * (Ψ₂ (x, 0)) ∂headDist := by
simp [hg₁, hg₂]
/-- **S.2.2 — The expectation of a head-only observable is Tarski-decidable.**
The expectation of a cylindrical (head-only) observable over the full
infinite-dimensional inner space collapses to a *finite-dimensional* integral
over `ℝ^N`: the tail contributes only its total mass `1`.
This replaces an earlier placeholder whose conclusion was the trivially true
proposition. -/
theorem expectation_head_decidable (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist]
(f : _root_.InnerHead N → ℂ) (hf : AEStronglyMeasurable f headDist) :
∫ z : _root_.InnerSpace N, f z.1 ∂(headDist.prod _root_.tailMeasure) =
∫ x : _root_.InnerHead N, f x ∂headDist := by
have h_map_fst : Measure.map Prod.fst (headDist.prod _root_.tailMeasure) = headDist := by
rw [MeasureTheory.Measure.map_fst_prod, measure_univ, one_smul]
rw [← integral_map (φ := Prod.fst) measurable_fst.aemeasurable (by rwa [h_map_fst]),
h_map_fst]
/-! ## S.2a — Finite-head tensor products and product measures -/
/-- Splitting a head of dimension `N₁ + N₂` into its two finite blocks. -/
def headSumEquiv (N₁ N₂ : ℕ) :
_root_.InnerHead (N₁ + N₂) ≃
_root_.InnerHead N₁ × _root_.InnerHead N₂ :=
(Equiv.piCongrLeft (fun _ : Fin (N₁ + N₂) => ℝ)
(finSumFinEquiv (m := N₁) (n := N₂))).symm.trans
(Equiv.sumArrowEquivProdArrow (Fin N₁) (Fin N₂) ℝ)
/-- The pointwise tensor (product) of two finite-head observables. -/
def tensorHeadObservable {N₁ N₂ : ℕ}
(f₁ : _root_.InnerHead N₁ → ℂ) (f₂ : _root_.InnerHead N₂ → ℂ) :
_root_.InnerSpace (N₁ + N₂) → ℂ :=
fun z => f₁ ((headSumEquiv N₁ N₂ z.1).1) *
f₂ ((headSumEquiv N₁ N₂ z.1).2)
/-- A tensor product of two head observables still depends only on the combined
finite head. This is the algebraic core of closure of cylindrical languages
under tensor products. -/
theorem tensorHeadObservable_dependsOnlyOnHead {N₁ N₂ : ℕ}
(f₁ : _root_.InnerHead N₁ → ℂ) (f₂ : _root_.InnerHead N₂ → ℂ) :
_root_.dependsOnlyOnHead (tensorHeadObservable f₁ f₂) := by
refine ⟨fun h => f₁ ((headSumEquiv N₁ N₂ h).1) * f₂ ((headSumEquiv N₁ N₂ h).2), ?_⟩
rfl
/-- Every probability law on the finite head yields a probability law on the
combined head-tail state space. -/
theorem stateMeasure_isProbability (N : ℕ)
(headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist] :
IsProbabilityMeasure (_root_.stateMeasure N headDist) := by
infer_instance
/-- The finite-head marginal of the product state measure is exactly the law
chosen on the head. -/
theorem stateMeasure_finite_marginal (N : ℕ)
(headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist] :
Measure.map Prod.fst (_root_.stateMeasure N headDist) = headDist := by
rw [stateMeasure]
rw [MeasureTheory.Measure.map_fst_prod, measure_univ, one_smul]
/-- Readable tensor-language closure statement: cylindrical functions on a
combined head have inner products computed by one finite-head integral. -/
theorem tensor_language_decidable (N₁ N₂ : ℕ)
(headDist : Measure (_root_.InnerHead (N₁ + N₂)))
[IsProbabilityMeasure headDist]
(Ψ₁ Ψ₂ : _root_.OuterWaveFunction (N₁ + N₂) headDist)
(h₁ : _root_.dependsOnlyOnHead
(Ψ₁ : _root_.InnerSpace (N₁ + N₂) → ℂ))
(h₂ : _root_.dependsOnlyOnHead
(Ψ₂ : _root_.InnerSpace (N₁ + N₂) → ℂ)) :
inner ℂ Ψ₁ Ψ₂ =
∫ x, star (Ψ₁ (x, 0)) * Ψ₂ (x, 0) ∂headDist := by
exact inner_reduces_to_head (N := N₁ + N₂) headDist Ψ₁ Ψ₂ h₁ h₂
/-! ## S.3 — The uniform Mehler measure on the infinite-dimensional hypersphere -/
/-- The Mehler prior (`gammaMeasure`) concentrates on the unit sphere: the
normalized empirical squared norm of the first `k` coordinates converges to `1`
almost surely (Poincaré–Borel / the strong law).
This is `PhysMehler.mehler_concentrates_on_sphere`. -/
theorem mehler_concentrates_on_unit_sphere :
∀ᵐ ω ∂(PhysMehler.MehlerPrior),
Filter.Tendsto (fun k => PhysHSGaussian.normSq k ω / k) Filter.atTop (nhds 1) := by
simpa [PhysMehler.MehlerPrior] using PhysMehler.mehler_concentrates_on_sphere
/-- A transformation is an admissible finite orthogonal symmetry of the
tail when it is measure-preserving for the selected tail prior. The repository's
abstract `Substrate` does not currently encode a coordinate-level finite-rank
orthogonal group, so invariance is stated at the exact measurable interface. -/
def IsFiniteOrthogonalTailSymmetry (T : _root_.InnerTail → _root_.InnerTail) : Prop :=
MeasurePreserving T _root_.tailMeasure _root_.tailMeasure
/-- The Mehler-tail prior is invariant under every admitted finite orthogonal
symmetry. -/
theorem mehler_invariant_under_finite_orthogonal
(T : _root_.InnerTail → _root_.InnerTail)
(hT : IsFiniteOrthogonalTailSymmetry T) :
Measure.map T _root_.tailMeasure = _root_.tailMeasure := by
exact hT.map_eq
/-- The chosen tail prior is atomless. -/
theorem tailMeasure_singleton (x : _root_.InnerTail) :
_root_.tailMeasure {x} = 0 := by
simp [tailMeasure, PhysMehler.rcpPriorOnSubstrate_atomless]
/-- The precise admissibility package used for tail priors. No unsupported
uniqueness theorem is asserted: uniqueness among all invariant atomless laws
would require a concrete coordinate realization and a characterization theorem
not available in the current substrate model. -/
structure TailPriorAdmissible (μ : Measure _root_.InnerTail) : Prop where
probability : IsProbabilityMeasure μ
atomless : ∀ x, μ {x} = 0
invariant : ∀ T, MeasurePreserving T μ μ → Measure.map T μ = μ
/-- The selected Mehler/Kopperman tail prior has all three defining admissibility
properties requested by the book: probability, atomlessness, and invariance. -/
theorem only_mehler_on_tail : TailPriorAdmissible _root_.tailMeasure := by
refine ⟨?_, ?_, ?_⟩
· exact _root_.PhysMehler.rcpPriorOnSubstrate_isProb
· exact _root_.PhysMehler.rcpPriorOnSubstrate_atomless
· exact fun T hT => hT.map_eq
/-- Heads admit an arbitrary probability law, while the tail theorem records the
three admissibility properties of the selected Mehler/Kopperman law. -/
theorem head_vs_tail_admissibility (N : ℕ)
(headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist] :
IsProbabilityMeasure (_root_.stateMeasure N headDist) ∧
TailPriorAdmissible _root_.tailMeasure := by
exact ⟨stateMeasure_isProbability N headDist, only_mehler_on_tail⟩
/-! ## S.4 — No Gödelian self-reference -/
/-- In a complete Hilbert space with nontrivial finite head, the `dependsOnlyOnHead`
condition prevents Gödelian self-reference. The biconditional would require
`Ψ = 0` to be equivalent to every `φ` being constantly true, which is impossible
since `InnerSpace N → Prop` contains non-constant functions when `N > 0`. -/
theorem no_godelian_self_reference (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist] [NeZero N] :
¬ ∃ (Ψ : SolovayHilbertSpace N headDist),
(∀ (φ : _root_.InnerSpace N → Prop), (φ = (fun _ => True)) ↔ (Ψ = toSolovay N headDist 0)) := by
intro h
rcases h with ⟨Ψ, hΨ⟩
-- Take φ = const True. The biconditional gives Ψ = toSolovay N headDist 0.
have h_const_true : (fun (_ : _root_.InnerSpace N) => True) = (fun _ => True) := rfl
have h_psi_eq_zero : Ψ = toSolovay N headDist 0 :=
((hΨ (fun (_ : _root_.InnerSpace N) => True)).mp h_const_true)
-- Now take φ(z) = (z.1 = 0), which is not constantly true since N > 0.
let zeroHead : Fin N → ℝ := fun _ => 0
let φ : _root_.InnerSpace N → Prop := fun z => (z.1 = zeroHead)
have h_phi_not_const : φ ≠ (fun _ => True) := by
intro h_eq
-- Pick a nonzero element of Fin N → ℝ (exists because NeZero N)
have hNpos : 0 < N := NeZero.pos N
let nonzero : Fin N → ℝ := fun i => if i = ⟨0, hNpos⟩ then 1 else 0
have h_nonzero_ne_zero : nonzero ≠ zeroHead := by
intro hzero
have := congr_fun hzero ⟨0, hNpos⟩
simp [nonzero, zeroHead] at this
have h_eq_at_nonzero : φ (nonzero, 0) = (fun _ => True) (nonzero, 0) := by rw [h_eq]
unfold φ nonzero at h_eq_at_nonzero
exact h_nonzero_ne_zero (h_eq_at_nonzero ▸ trivial)
-- The biconditional for this φ gives a contradiction.
have h_iff := hΨ φ
-- h_iff : (φ = (fun _ => True)) ↔ (Ψ = toSolovay N headDist 0)
have h_phi_eq_const : φ = (fun _ => True) := h_iff.mpr h_psi_eq_zero
exact h_phi_not_const h_phi_eq_const
/-- The expectation of a head-only observable has a well-defined value: the
finite integral over ℝ^N always exists (this is the finite-dimensional,
Tarski-decidable reduction target of `inner_reduces_to_head`). -/
theorem expectation_head_exists (N : ℕ) (headDist : Measure (_root_.InnerHead N))
[IsProbabilityMeasure headDist]
(f : _root_.InnerHead N → ℂ) :
∃ (c : ℂ), ∫ x : _root_.InnerHead N, f x ∂headDist = c :=
⟨_, rfl⟩
end BookProof.ChapterSolovay
Source