/- Copyright (c) 2026 Michael R. Douglas. All rights reserved. Released under Apache 2.0 license as described in the file LICENSE. # Gaussian Measure Support Theorem Characterizes the support of the Gaussian measure μ_T on the weak dual E'. The measure is supported on configurations with finite Cameron-Martin norm (i.e., Σₙ |ω(eₙ)|² < ∞) if and only if T is Hilbert-Schmidt with respect to the DyninMityaginSpace basis. ## Main definitions - `IsHilbertSchmidt T` — T is Hilbert-Schmidt: Σₙ ‖T(eₙ)‖² < ∞ - `hilbertSchmidtNormSq T` — the HS norm squared: Σₙ ‖T(eₙ)‖² - `cameronMartinInner T` — the Cameron-Martin inner product C(f,g) - `cameronMartinNormSq T` — the Cameron-Martin norm squared ‖T(f)‖² ## Main results - `expected_norm_sq_eq_hs` — E[Σₙ |ω(eₙ)|²] = Σₙ ‖T(eₙ)‖² (HS norm²) - `support_of_hilbertSchmidt` — if T is HS, then a.e. ω has Σₙ |ω(eₙ)|² < ∞ - `weighted_support` — weighted-HS ⟹ a.e. weighted-summable ## References - Glimm-Jaffe, *Quantum Physics: A Functional Integral Point of View*, Ch. 1 - Simon, *The P(φ)₂ Euclidean (Quantum) Field Theory*, Ch. II -/ import GaussianField.Properties import Mathlib.MeasureTheory.Integral.DominatedConvergence import Mathlib.MeasureTheory.Integral.Lebesgue.Markov noncomputable section namespace GaussianField open MeasureTheory ProbabilityTheory TopologicalSpace open scoped BigOperators NNReal open Classical variable {E : Type*} [AddCommGroup E] [Module ℝ E] [TopologicalSpace E] [IsTopologicalAddGroup E] [ContinuousSMul ℝ E] [DyninMityaginSpace E] variable {H : Type*} [NormedAddCommGroup H] [InnerProductSpace ℝ H] [CompleteSpace H] [SeparableSpace H] /-! ## Cameron-Martin inner product and norm -/ /-- The Cameron-Martin inner product: ⟨f, g⟩₀ = ⟨T(f), T(g)⟩_H = C(f,g). This is the reproducing kernel inner product of the Gaussian measure. -/ def cameronMartinInner (T : E →L[ℝ] H) (f g : E) : ℝ := covariance T f g /-- The Cameron-Martin norm squared: ‖f‖₀² = ‖T(f)‖²_H. The Cameron-Martin space H₀ is the completion of E under this norm. -/ def cameronMartinNormSq (T : E →L[ℝ] H) (f : E) : ℝ := ‖T f‖ ^ 2 omit [IsTopologicalAddGroup E] [ContinuousSMul ℝ E] [DyninMityaginSpace E] [CompleteSpace H] [SeparableSpace H] in theorem cameronMartinNormSq_eq_inner (T : E →L[ℝ] H) (f : E) : cameronMartinNormSq T f = cameronMartinInner T f f := by simp [cameronMartinNormSq, cameronMartinInner, covariance] /-! ## Hilbert-Schmidt condition -/ /-- T is Hilbert-Schmidt with respect to the DyninMityaginSpace basis: Σₙ ‖T(eₙ)‖² < ∞, where (eₙ) is the Schauder basis of E. In the QFT application, this is the condition for the Gaussian measure to be supported on a Sobolev space rather than the full distributional dual. -/ def IsHilbertSchmidt (T : E →L[ℝ] H) : Prop := Summable (fun n => ‖T (DyninMityaginSpace.basis n)‖ ^ 2) /-- The Hilbert-Schmidt norm squared: Σₙ ‖T(eₙ)‖². This equals the expected squared norm E[Σₙ |ω(eₙ)|²] of the field. -/ noncomputable def hilbertSchmidtNormSq (T : E →L[ℝ] H) : ℝ := ∑' n, ‖T (DyninMityaginSpace.basis n)‖ ^ 2 /-- The squared norm of ω in the basis: Σₙ |ω(eₙ)|². -/ noncomputable def basisNormSq (ω : Configuration E) : ℝ := ∑' n, (ω (DyninMityaginSpace.basis n)) ^ 2 /-! ## Expected squared norm equals HS norm -/ /-- Each term E[|ω(eₙ)|²] = ‖T(eₙ)‖² is the second moment identity applied to the n-th basis element. -/ theorem basis_second_moment (T : E →L[ℝ] H) (n : ℕ) : ∫ ω : Configuration E, (ω (DyninMityaginSpace.basis n)) ^ 2 ∂(measure T) = ‖T (DyninMityaginSpace.basis n)‖ ^ 2 := by rw [second_moment_eq_covariance] exact real_inner_self_eq_norm_sq (T (DyninMityaginSpace.basis n)) /-- Each basis pairing squared is integrable. -/ theorem basis_sq_integrable (T : E →L[ℝ] H) (n : ℕ) : Integrable (fun ω : Configuration E => (ω (DyninMityaginSpace.basis n)) ^ 2) (measure T) := (pairing_memLp T (DyninMityaginSpace.basis n) 2).integrable_sq /-- Each basis pairing squared is AEStronglyMeasurable. -/ theorem basis_sq_aestronglyMeasurable (T : E →L[ℝ] H) (n : ℕ) : AEStronglyMeasurable (fun ω : Configuration E => (ω (DyninMityaginSpace.basis n)) ^ 2) (measure T) := (basis_sq_integrable T n).aestronglyMeasurable /-- The enorm sum condition needed for `integral_tsum`: the tsum of lintegrals of enorms is finite when T is Hilbert-Schmidt. -/ private theorem enorm_tsum_ne_top (T : E →L[ℝ] H) (hHS : IsHilbertSchmidt T) : (∑' n, ∫⁻ ω, ‖(ω (DyninMityaginSpace.basis n)) ^ 2‖ₑ ∂(measure T)) ≠ ⊤ := by -- Convert enorm of nonneg real to ofReal have h_enorm : ∀ n, (∫⁻ ω : Configuration E, ‖(ω (DyninMityaginSpace.basis n)) ^ 2‖ₑ ∂(measure T)) = ENNReal.ofReal (‖T (DyninMityaginSpace.basis n)‖ ^ 2) := by intro n rw [show (fun ω : Configuration E => ‖(ω (DyninMityaginSpace.basis n)) ^ 2‖ₑ) = (fun ω : Configuration E => ENNReal.ofReal ((ω (DyninMityaginSpace.basis n)) ^ 2)) from by ext ω; rw [← ofReal_norm, Real.norm_of_nonneg (sq_nonneg _)]] rw [← ofReal_integral_eq_lintegral_ofReal (basis_sq_integrable T n) (ae_of_all _ (fun ω => sq_nonneg _))] congr 1 exact basis_second_moment T n simp_rw [h_enorm] exact hHS.tsum_ofReal_ne_top /-- The expected squared norm equals the Hilbert-Schmidt norm: E[Σₙ |ω(eₙ)|²] = Σₙ ‖T(eₙ)‖² (= HS norm²). Each term: E[|ω(eₙ)|²] = ‖T(eₙ)‖² by `second_moment_eq_covariance`. The interchange of ∫ and Σ' is justified by `integral_tsum` using the nonneg integrability of each squared term. This is the key identity connecting the a.s. regularity of samples to the Hilbert-Schmidt property of T. -/ theorem expected_norm_sq_eq_hs (T : E →L[ℝ] H) (hHS : IsHilbertSchmidt T) : ∫ ω : Configuration E, basisNormSq ω ∂(measure T) = hilbertSchmidtNormSq T := by simp only [basisNormSq, hilbertSchmidtNormSq] rw [integral_tsum (basis_sq_aestronglyMeasurable T) (enorm_tsum_ne_top T hHS)] congr 1 ext n exact basis_second_moment T n /-! ## Forward support theorem -/ /-- If T is Hilbert-Schmidt, the Gaussian measure is supported on configurations with finite squared norm Σₙ |ω(eₙ)|² < ∞. Proof: We work in ENNReal. The lintegral of the ENNReal-valued tsum equals the tsum of lintegrals (by `lintegral_tsum`), each of which equals ENNReal.ofReal(‖T eₙ‖²). The tsum is finite by the HS condition. By `ae_lt_top`, the ENNReal tsum is a.e. finite, which gives real summability. -/ theorem support_of_hilbertSchmidt (T : E →L[ℝ] H) (hHS : IsHilbertSchmidt T) : ∀ᵐ ω ∂(measure T), Summable (fun n => (ω (DyninMityaginSpace.basis n)) ^ 2) := by -- Define the ENNReal-valued function set g : Configuration E → ENNReal := fun ω => ∑' n, ENNReal.ofReal ((ω (DyninMityaginSpace.basis n)) ^ 2) with hg_def -- g is measurable (tsum of measurable functions) have hg_meas : Measurable g := by apply Measurable.tsum intro n exact ENNReal.measurable_ofReal.comp ((configuration_eval_measurable (DyninMityaginSpace.basis n)).pow_const 2) -- Each component is measurable have hg_comp_meas : ∀ n, Measurable (fun ω : Configuration E => ENNReal.ofReal ((ω (DyninMityaginSpace.basis n)) ^ 2)) := fun n => ENNReal.measurable_ofReal.comp ((configuration_eval_measurable (DyninMityaginSpace.basis n)).pow_const 2) -- ∫⁻ g < ⊤ have hg_int : ∫⁻ ω, g ω ∂(measure T) ≠ ⊤ := by have h_eq_tsum : ∫⁻ ω, g ω ∂(measure T) = ∑' n, ∫⁻ ω : Configuration E, ENNReal.ofReal ((ω (DyninMityaginSpace.basis n)) ^ 2) ∂(measure T) := lintegral_tsum (fun n => (hg_comp_meas n).aemeasurable) rw [h_eq_tsum] have h_eq : ∀ n, (∫⁻ ω : Configuration E, ENNReal.ofReal ((ω (DyninMityaginSpace.basis n)) ^ 2) ∂(measure T)) = ENNReal.ofReal (‖T (DyninMityaginSpace.basis n)‖ ^ 2) := by intro n rw [← ofReal_integral_eq_lintegral_ofReal (basis_sq_integrable T n) (ae_of_all _ (fun ω => sq_nonneg _))] congr 1 exact basis_second_moment T n simp_rw [h_eq] exact hHS.tsum_ofReal_ne_top -- By ae_lt_top, g(ω) < ⊤ a.e. have hg_ae : ∀ᵐ ω ∂(measure T), g ω < ⊤ := ae_lt_top hg_meas hg_int -- g(ω) < ⊤ means the ENNReal tsum is finite, hence real summable filter_upwards [hg_ae] with ω hω have hω' : ∑' n, ENNReal.ofReal ((ω (DyninMityaginSpace.basis n)) ^ 2) < ⊤ := hω have h_summ := ENNReal.summable_toReal hω'.ne simp only [ENNReal.toReal_ofReal (sq_nonneg _)] at h_summ exact h_summ /-! ## Weighted Hilbert-Schmidt condition and support -/ /-- Weighted Hilbert-Schmidt condition: Σₙ wₙ ‖T(eₙ)‖² < ∞. For Sobolev regularity, take wₙ = (1+λₙ)⁻ˢ where λₙ are eigenvalues of the reference operator (e.g., harmonic oscillator). -/ def IsWeightedHS (T : E →L[ℝ] H) (w : ℕ → ℝ) : Prop := Summable (fun n => w n * ‖T (DyninMityaginSpace.basis n)‖ ^ 2) /-- Weighted basis norm: Σₙ wₙ |ω(eₙ)|². -/ noncomputable def weightedBasisNormSq (w : ℕ → ℝ) (ω : Configuration E) : ℝ := ∑' n, w n * (ω (DyninMityaginSpace.basis n)) ^ 2 /-- Weighted second moment: E[wₙ |ω(eₙ)|²] = wₙ ‖T(eₙ)‖². -/ theorem weighted_basis_second_moment (T : E →L[ℝ] H) (w : ℕ → ℝ) (_hw : ∀ n, 0 ≤ w n) (n : ℕ) : ∫ ω : Configuration E, w n * (ω (DyninMityaginSpace.basis n)) ^ 2 ∂(measure T) = w n * ‖T (DyninMityaginSpace.basis n)‖ ^ 2 := by rw [MeasureTheory.integral_const_mul, basis_second_moment] /-- Weighted basis pairing is integrable. -/ theorem weighted_basis_sq_integrable (T : E →L[ℝ] H) (w : ℕ → ℝ) (n : ℕ) : Integrable (fun ω : Configuration E => w n * (ω (DyninMityaginSpace.basis n)) ^ 2) (measure T) := (basis_sq_integrable T n).const_mul (w n) /-- If T is weighted-HS with nonneg weights, the Gaussian measure is supported on configurations with finite weighted norm Σₙ wₙ|ω(eₙ)|² < ∞. Proof: same strategy as `support_of_hilbertSchmidt` — the weight wₙ is a constant that passes through the integral, so E[wₙ(ω eₙ)²] = wₙ‖T eₙ‖². Summability of these gives finite lintegral, then `ae_lt_top`. -/ theorem weighted_support (T : E →L[ℝ] H) (w : ℕ → ℝ) (hw : ∀ n, 0 ≤ w n) (hWHS : IsWeightedHS T w) : ∀ᵐ ω ∂(measure T), Summable (fun n => w n * (ω (DyninMityaginSpace.basis n)) ^ 2) := by set g : Configuration E → ENNReal := fun ω => ∑' n, ENNReal.ofReal (w n * (ω (DyninMityaginSpace.basis n)) ^ 2) have hg_comp_meas : ∀ n, Measurable (fun ω : Configuration E => ENNReal.ofReal (w n * (ω (DyninMityaginSpace.basis n)) ^ 2)) := fun n => ENNReal.measurable_ofReal.comp (measurable_const.mul ((configuration_eval_measurable (DyninMityaginSpace.basis n)).pow_const 2)) have hg_meas : Measurable g := Measurable.tsum hg_comp_meas have hg_int : ∫⁻ ω, g ω ∂(measure T) ≠ ⊤ := by have h_eq_tsum : ∫⁻ ω, g ω ∂(measure T) = ∑' n, ∫⁻ ω : Configuration E, ENNReal.ofReal (w n * (ω (DyninMityaginSpace.basis n)) ^ 2) ∂(measure T) := lintegral_tsum (fun n => (hg_comp_meas n).aemeasurable) rw [h_eq_tsum] have h_eq : ∀ n, (∫⁻ ω : Configuration E, ENNReal.ofReal (w n * (ω (DyninMityaginSpace.basis n)) ^ 2) ∂(measure T)) = ENNReal.ofReal (w n * ‖T (DyninMityaginSpace.basis n)‖ ^ 2) := by intro n rw [← ofReal_integral_eq_lintegral_ofReal (weighted_basis_sq_integrable T w n) (ae_of_all _ (fun ω => mul_nonneg (hw n) (sq_nonneg _)))] congr 1 exact weighted_basis_second_moment T w hw n simp_rw [h_eq] exact hWHS.tsum_ofReal_ne_top have hg_ae : ∀ᵐ ω ∂(measure T), g ω < ⊤ := ae_lt_top hg_meas hg_int filter_upwards [hg_ae] with ω hω have h_summ := ENNReal.summable_toReal hω.ne simp only [ENNReal.toReal_ofReal (mul_nonneg (hw _) (sq_nonneg _))] at h_summ exact h_summ omit [CompleteSpace H] [SeparableSpace H] in /-- Unweighted HS is the special case of weighted HS with w = 1. -/ theorem isHilbertSchmidt_eq_weightedHS_one (T : E →L[ℝ] H) : IsHilbertSchmidt T ↔ IsWeightedHS T (fun _ => 1) := by simp [IsHilbertSchmidt, IsWeightedHS, one_mul] end GaussianField