Logarithmic Laplace transform and cumulant tensors
DefinitionKLSModular_FunctionalInequalities_Analysis_BakryEmery_Cumulant_BasicFunctionalInequalities.Analysis.BakryEmery.logLaplace μ z is , and cumulant μ n is the nth iterated Fréchet derivative of this transform at zero, represented as a continuous n-linear form. For a nonzero finite measure with all exponential radial moments, contDiff_logLaplace asserts that the logarithmic Laplace transform is smooth. LetwinQuadraticBound is an explicit hypothesis requiring, for every self-adjoint continuous operator M, the variance of to be at most eight times the squared Hilbert--Schmidt coordinate norm; it is defined here rather than proved.
Original module imports
import FunctionalInequalities.Analysis.BakryEmery.Tilt.SquareIntegrable
import FunctionalInequalities.Analysis.BakryEmery.Tilt.Local
import Mathlib.Probability.Moments.Variance
Original declaration excerpt
Shown in the original module context.
def logLaplace (μ : Measure E) (z : E) : ℝ :=
Real.log (∫ x, Real.exp ⟪z, x⟫ ∂μ)
Module navigation
Direct project imports: Analysis/BakryEmery/Tilt/SquareIntegrable, Analysis/BakryEmery/Tilt/Local.
Theorems in source order:
tiltAverage_inner_eq_fderiv_logLaplacetiltAverage_inner_eq_fderiv_logLaplace_of_lttaylorTensor_inner_eq_cumulantnorm_coordTensor_taylorTensor_inner_sq_le_of_basiscontDiffOn_logLaplace_ballcumulant_map_linearIsometryEquivcumulant_map_affineletwinQuadraticBound_of_tendsto_moments
Project result: The KLS conjecture.
Original module · FunctionalInequalities.Analysis.BakryEmery.Cumulant.Basic · Frozen source 408915f.
import Definitions.Def_KLSModular_FunctionalInequalities_Analysis_BakryEmery_Tilt_SquareIntegrable
import Definitions.Def_KLSModular_FunctionalInequalities_Analysis_BakryEmery_Tilt_Local
import Mathlib.Probability.Moments.Variance
noncomputable section
open MeasureTheory
open scoped RealInnerProductSpace ContDiff
namespace FunctionalInequalities
namespace Analysis
namespace BakryEmery
universe u
variable {E : Type u} [NormedAddCommGroup E] [InnerProductSpace ℝ E] [CompleteSpace E]
[FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E]
/-- The logarithmic Laplace transform `Λ_μ(z) = log ∫ e^{⟪z, x⟫} dμ(x)` (BKL Section 4). -/
def logLaplace (μ : Measure E) (z : E) : ℝ :=
Real.log (∫ x, Real.exp ⟪z, x⟫ ∂μ)
/-- The cumulant tensor `κ_m^μ = ∇^m Λ_μ(0)` (BKL Section 4), as a continuous `m`-linear form. -/
def cumulant (μ : Measure E) (n : ℕ) : ContinuousMultilinearMap ℝ (fun _ : Fin n => E) ℝ :=
iteratedFDeriv ℝ n (logLaplace μ) 0
section LaplaceCalculus
omit [CompleteSpace E]
/-- The logarithmic Laplace transform is smooth for a nonzero finite measure with all exponential
moments. -/
theorem contDiff_logLaplace (μ : Measure E) [IsFiniteMeasure μ] [NeZero μ]
(hexp : ∀ r : ℝ, Integrable (fun x => Real.exp (r * ‖x‖)) μ) :
ContDiff ℝ ∞ (logLaplace μ) := by
have hZ : ContDiff ℝ ∞ (fun z : E => ∫ x, Real.exp ⟪z, x⟫ ∂μ) := by
simpa [smul_eq_mul, mul_one] using
(contDiff_integral_exp_inner_smul μ hexp
(g := fun _ : E => (1 : ℝ))
(M := 1) (k := 0)
(hg := measurable_const.aestronglyMeasurable)
(hgb := by
intro x
simp))
have hpos : ∀ z : E, 0 < ∫ x, Real.exp ⟪z, x⟫ ∂μ := by
intro z
have hzint : Integrable (fun x : E => Real.exp ⟪z, x⟫) μ := by
refine (hexp ‖z‖).mono' ?_ ?_
· exact
(Real.continuous_exp.comp
(continuous_const.inner continuous_id)).aestronglyMeasurable
· filter_upwards [] with x
rw [Real.norm_eq_abs, abs_of_pos (Real.exp_pos _)]
exact Real.exp_le_exp.mpr (real_inner_le_norm z x)
exact integral_exp_pos hzint
change ContDiff ℝ ∞
(fun z : E => Real.log (∫ x, Real.exp ⟪z, x⟫ ∂μ))
exact hZ.log (fun z => (hpos z).ne')
end LaplaceCalculus
section ThirdCumulant
/-- The quadratic-form estimate of Letwin (2026, Theorem 1.2), stated as a property of the law `μ`:
`Var_μ⟨Mx, x⟩ ≤ 8 |M|²_HS` for every self-adjoint `M`, with `|M|²_HS = ∑_{i,j} ⟪M e_i, e_j⟫²` in an
orthonormal basis. This names the hypothesis; it is not proved here. -/
def LetwinQuadraticBound (μ : Measure E) : Prop :=
∀ M : E →L[ℝ] E, IsSelfAdjoint M →
∫ x, (⟪M x, x⟫ - ∫ y, ⟪M y, y⟫ ∂μ) ^ 2 ∂μ ≤
8 * ∑ i, ∑ j, ⟪M (stdOrthonormalBasis ℝ E i), stdOrthonormalBasis ℝ E j⟫ ^ 2
variable (μ : Measure E) [IsProbabilityMeasure μ]
(hexp : ∀ r : ℝ, Integrable (fun x => Real.exp (r * ‖x‖)) μ)
end ThirdCumulant
end BakryEmery
end Analysis
end FunctionalInequalities