production module
AutoSamplingTheory.TechnicalLemmas.Measure.QuadraticOptimalRealMinimality
1 named declarations scanned from AutoSamplingTheory/TechnicalLemmas/Measure/QuadraticOptimalRealMinimality.lean.
Declarations
theorem AutoSamplingTheory.TechnicalLemmas.Measure.QuadraticOptimalRealMinimality.integral_norm_sq_le_of_quadraticOptimal Partial Not mapped
- A quadratic-optimal coupling minimizes the ordinary real squared-displacement integral among all couplings with the same marginals, whenever the compared real costs are integrable.
theorem integral_norm_sq_le_of_quadraticOptimal
{gamma xi : Measure (E × E)} {mu0 mu1 : Measure E}
(hgamma : IsQuadraticOptimalCoupling gamma mu0 mu1)
(hxi : Transport.IsCoupling xi mu0 mu1)
(hgammaInt : Integrable (fun z : E × E => ‖z.1 - z.2‖ ^ 2) gamma)
(hxiInt : Integrable (fun z : E × E => ‖z.1 - z.2‖ ^ 2) xi) :
(∫ z, ‖z.1 - z.2‖ ^ 2 ∂gamma) ≤
∫ z, ‖z.1 - z.2‖ ^ 2 ∂xi := by
have hENN :
(∫⁻ z, WassersteinSpace.quadraticCost (E := E) z ∂gamma) ≤
∫⁻ z, WassersteinSpace.quadraticCost (E := E) z ∂xi := by
rw [hgamma.2]
exact Transport.transportCost_le_lintegral_of_isCoupling
(WassersteinSpace.quadraticCost (E := E)) mu0 mu1 xi hxi
have hOfReal :
ENNReal.ofReal (∫ z, ‖z.1 - z.2‖ ^ 2 ∂gamma) ≤
ENNReal.ofReal (∫ z, ‖z.1 - z.2‖ ^ 2 ∂xi) := by
rw [ofReal_integral_norm_sq_eq_lintegral_quadraticCost gamma hgammaInt,
ofReal_integral_norm_sq_eq_lintegral_quadraticCost xi hxiInt]
exact hENN
have hxiNonneg : 0 ≤ ∫ z, ‖z.1 - z.2‖ ^ 2 ∂xi :=
integral_nonneg fun z => sq_nonneg ‖z.1 - z.2‖
exact (ENNReal.ofReal_le_ofReal_iff hxiNonneg).mp hOfReal
end
end QuadraticOptimalRealMinimality
end Measure
end TechnicalLemmas
end AutoSamplingTheory
AutoSamplingTheory/TechnicalLemmas/Measure/QuadraticOptimalRealMinimality.lean:30published source at ed709a7f320d
Complete Lean source
AutoSamplingTheory/TechnicalLemmas/Measure/QuadraticOptimalRealMinimality.lean — full module with imports and context
import AutoSamplingTheory.TechnicalLemmas.Measure.DisplacementRealQuadraticCost
import AutoSamplingTheory.TechnicalLemmas.Measure.Transport
/-!
# Real quadratic-cost minimality of an optimal coupling
This module isolates the final optimality inequality used by the direct Brenier
perturbation. Once a coupling is known to attain the Kantorovich quadratic
infimum, every other coupling with the same marginals has no smaller real
squared-displacement integral, provided the two real costs are integrable.
-/
namespace AutoSamplingTheory
namespace TechnicalLemmas
namespace Measure
namespace QuadraticOptimalRealMinimality
open MeasureTheory
open DisplacementInterpolation DisplacementRealQuadraticCost
open scoped ENNReal
noncomputable section
variable {E : Type*} [NormedAddCommGroup E] [InnerProductSpace ℝ E]
[FiniteDimensional ℝ E] [MeasurableSpace E] [BorelSpace E]
/-- A quadratic-optimal coupling minimizes the ordinary real
squared-displacement integral among all couplings with the same marginals,
whenever the compared real costs are integrable. -/
theorem integral_norm_sq_le_of_quadraticOptimal
{gamma xi : Measure (E × E)} {mu0 mu1 : Measure E}
(hgamma : IsQuadraticOptimalCoupling gamma mu0 mu1)
(hxi : Transport.IsCoupling xi mu0 mu1)
(hgammaInt : Integrable (fun z : E × E => ‖z.1 - z.2‖ ^ 2) gamma)
(hxiInt : Integrable (fun z : E × E => ‖z.1 - z.2‖ ^ 2) xi) :
(∫ z, ‖z.1 - z.2‖ ^ 2 ∂gamma) ≤
∫ z, ‖z.1 - z.2‖ ^ 2 ∂xi := by
have hENN :
(∫⁻ z, WassersteinSpace.quadraticCost (E := E) z ∂gamma) ≤
∫⁻ z, WassersteinSpace.quadraticCost (E := E) z ∂xi := by
rw [hgamma.2]
exact Transport.transportCost_le_lintegral_of_isCoupling
(WassersteinSpace.quadraticCost (E := E)) mu0 mu1 xi hxi
have hOfReal :
ENNReal.ofReal (∫ z, ‖z.1 - z.2‖ ^ 2 ∂gamma) ≤
ENNReal.ofReal (∫ z, ‖z.1 - z.2‖ ^ 2 ∂xi) := by
rw [ofReal_integral_norm_sq_eq_lintegral_quadraticCost gamma hgammaInt,
ofReal_integral_norm_sq_eq_lintegral_quadraticCost xi hxiInt]
exact hENN
have hxiNonneg : 0 ≤ ∫ z, ‖z.1 - z.2‖ ^ 2 ∂xi :=
integral_nonneg fun z => sq_nonneg ‖z.1 - z.2‖
exact (ENNReal.ofReal_le_ofReal_iff hxiNonneg).mp hOfReal
end
end QuadraticOptimalRealMinimality
end Measure
end TechnicalLemmas
end AutoSamplingTheory