OAI/Probability/InvariantIsing/Gaussian/GaussianGGBounds.lean
4 declarations, 0 contain sorry, 0 declared as axioms
Gaussian GG estimates retaining the spectral diagonal error.
- lemma
OAI.InvariantIsing.randomReplicaAverage_const_mulline 21 - lemma
OAI.InvariantIsing.randomReplicaAverage_add_boundedline 29 - theorem
OAI.InvariantIsing.randomReplicaEnergy_full_identityline 50Source documentation
Gaussian integration by parts without imposing a constant diagonal.
theorem randomReplicaEnergy_full_identity (hν : Measurable ν) (A : X → ℕ →₀ ℝ) {B : ℝ} (hA : ∀ x, (A x).sum (fun _ c => c ^ 2) ≤ B) (t : ℝ) {r : ℕ} (D : (Fin (r + 1) → X) → ℝ) {c : ℝ} (hc : 0 ≤ c) (hD : ∀ σ, |D σ| ≤ c) : randomReplicaEnergy P ν A t D = t * (randomReplicaAverage P ν A t (fun σ => D σ * (∑ i : Fin (r + 1), cylinderCross (A (σ 0)) (A (σ i)))) - (r + 1 : ℕ) * randomReplicaAverage P ν A t (fun τ : Fin (r + 1 + 1) → X => D (Fin.tail τ) * cylinderCross (A ((Fin.tail τ) 0)) (A (τ 0)))) := by … - theorem
OAI.InvariantIsing.countable_gaussian_gg_bound_variable_diagonalline 88Source documentation
The finite GG error splits into an energy fluctuation and an explicit covariance of the self-overlap. Spectral self-overlaps are not constant before adding their separate deterministic perturbations.
theorem countable_gaussian_gg_bound_variable_diagonal (hν : Measurable ν) (A : X → ℕ →₀ ℝ) {B : ℝ} (hA : ∀ x, (A x).sum (fun _ c => c ^ 2) ≤ B) (t b : ℝ) {r : ℕ} (D : (Fin (r + 1) → X) → ℝ) {c : ℝ} (hc : 0 ≤ c) (hD : ∀ σ, |D σ| ≤ c) (hE : Integrable (fun z : Ω × (ℕ → ℝ) => ∫ x, |cylinderField (A x) z.2 - b| ∂(ν z.1).tilted (fun x => t * cylinderField (A x) z.2)) (P.prod gaussianCoordinates)) : |t| * |(r + 1 : ℕ) * randomReplicaAverage P ν A t (fun τ : Fin (r + 1 + 1) → X => D (Fin.tail τ) * cylinderCross (A ((Fin.tail τ) 0)) (A (τ 0))) - randomReplicaAverage P ν A t D * randomReplicaAverage P ν A t …