Compact log-concavity from a Borell density
DefinitionKLSModular_FunctionalInequalities_KLS_LogConcave_Density_FromBorellfunctional-inequalitiesguedonmilman2011klskls-conjecturekls-modularlog-concavemodule-context
IsBorellDensity.compactLogConcave states that a finite-dimensional probability measure on an inner-product space with a pointwise log-concave Borell density relative to volume satisfies LogConcave on compact sets. The density may vanish, so positivity everywhere is not assumed; the result is an application of the finite-dimensional Prékopa–Leindler inequality.
Original module imports
import FunctionalInequalities.Probability.BrunnMinkowski
import FunctionalInequalities.KLS.LogConcave.Density.BorellRepresentation.Basic
Original declaration excerpt
Shown in the original module context. Body omitted.
theorem compactLogConcave {ν : ProbabilityMeasure V} {ρ : V → ℝ}
(hρ : IsBorellDensity (ν : Measure V) volume ρ) : LogConcave ν
Module navigation
Direct project imports: Probability/BrunnMinkowski, KLS/LogConcave/Density/BorellRepresentation/Basic.
Project result: The KLS conjecture.
Original module · FunctionalInequalities.KLS.LogConcave.Density.FromBorell · Frozen source 408915f.
Definition code
import Definitions.Def_KLSModular_FunctionalInequalities_Probability_BrunnMinkowski
import Definitions.Def_KLSModular_FunctionalInequalities_KLS_LogConcave_Density_BorellRepresentation_Basic
set_option autoImplicit false
noncomputable section
open MeasureTheory Set
namespace FunctionalInequalities
namespace KLS
namespace IsBorellDensity
variable {V : Type*} [NormedAddCommGroup V] [InnerProductSpace ℝ V]
[FiniteDimensional ℝ V] [MeasurableSpace V] [BorelSpace V]
/-- A probability law with a pointwise log-concave Borell density is log-concave
on compact sets. This does not require the density to be positive everywhere. -/
theorem compactLogConcave {ν : ProbabilityMeasure V} {ρ : V → ℝ}
(hρ : IsBorellDensity (ν : Measure V) volume ρ) : LogConcave ν := by
intro K₀ K₁ hK₀ hK₁ t ht0 ht1
have ht1' : (0 : ℝ) < 1 - t := by linarith
set C := {z : V | ∃ x ∈ K₀, ∃ y ∈ K₁, z = t • x + (1 - t) • y} with hC
have hCc : IsCompact C := isCompact_combination hK₀ hK₁ t
have hmass : ∀ S : Set V, MeasurableSet S →
(ν : Measure V) S = ∫⁻ x, ENNReal.ofReal (S.indicator ρ x) := by
intro S hS
have hind : ∀ x, ENNReal.ofReal (S.indicator ρ x) =
S.indicator (fun y => ENNReal.ofReal (ρ y)) x := by
intro x
by_cases hx : x ∈ S <;> simp [hx]
simp_rw [hind]
rw [lintegral_indicator hS, ← hρ.withDensity_eq, withDensity_apply _ hS]
rw [hmass K₀ hK₀.measurableSet, hmass K₁ hK₁.measurableSet,
hmass C hCc.measurableSet]
refine Analysis.prekopaLeindler (hρ.measurable.indicator hK₀.measurableSet)
(hρ.measurable.indicator hK₁.measurableSet)
(hρ.measurable.indicator hCc.measurableSet)
(fun x => Set.indicator_nonneg (fun y _ => hρ.nonneg y) x)
(fun y => Set.indicator_nonneg (fun z _ => hρ.nonneg z) y) ht0 ht1 ?_
intro x y
have hh0 : ∀ z, 0 ≤ C.indicator ρ z := fun z =>
Set.indicator_nonneg (fun w _ => hρ.nonneg w) z
by_cases hx : x ∈ K₀
· by_cases hy : y ∈ K₁
· have hz : t • x + (1 - t) • y ∈ C := ⟨x, hx, y, hy, rfl⟩
rw [Set.indicator_of_mem hx, Set.indicator_of_mem hy, Set.indicator_of_mem hz]
exact hρ.logConcave x y t ht0 ht1
· rw [Set.indicator_of_notMem hy, Real.zero_rpow ht1'.ne', mul_zero]
exact hh0 _
· rw [Set.indicator_of_notMem hx, Real.zero_rpow ht0.ne', zero_mul]
exact hh0 _
end IsBorellDensity
end KLS
end FunctionalInequalities
Source
Source: Guédon–Milman, Interpolating thin-shell and sharp large-deviation estimates for isotropic log-concave measures (arXiv:1011.0943v3), item DLC. Lean source: https://github.com/human-ai-mathematics/kls-lean/blob/408915f/formalization/FunctionalInequalities/KLS/LogConcave/Density/FromBorell.lean#L22-L57 (repository human-ai-mathematics/kls-lean, commit 408915f, module `FunctionalInequalities.KLS.LogConcave.Density.FromBorell`, definition bundle `KLSModular_FunctionalInequalities_KLS_LogConcave_Density_FromBorell`; the repository is private at upload time).