Entropy chain rule (two-coordinate Han subadditivity)
Provedentropy_chain_rule_general_two_coordinate_han_subadditivityGeneral (non-product) Han subadditivity / chain rule for the entropy functional, two-coordinate exact form. For independent coordinates mu (x) nu and a nonnegative f on the product (NOT required to factor across coordinates), with f, f*log f, and the second-marginal entropy integrand integrable, the entropy functional Ent_m(g) = int g log g dm - (int g dm) * log(int g dm) satisfies the exact chain rule Ent_{mu(x)nu}(f) = Ent_mu(g1) + int_x Ent_nu(f(x, .)) dmu, where g1 x = int_y f(x,y) dnu is the second-coordinate marginal. This is the entropy chain rule underlying the tensorization step of the entropy method (modified log-Sobolev / Massart eq.(4)): iterated over coordinates it decomposes Ent(e^{lambda Z}) into per-coordinate conditional entropies. Unlike product-density tensorization it holds for an arbitrary non-factoring f.
import Mathlib.MeasureTheory.Integral.Prod import Mathlib.MeasureTheory.Measure.Typeclasses.Probability import Mathlib.Analysis.SpecialFunctions.Log.Basic open Real MeasureTheory
theorem entropy_chain_rule_general_two_coordinate_han_subadditivity
{α β : Type*} [mα : MeasurableSpace α] [mβ : MeasurableSpace β]
{μ : Measure α} {ν : Measure β}
[IsProbabilityMeasure μ] [IsProbabilityMeasure ν]
{f : α × β → ℝ}
(hf_int : Integrable f (μ.prod ν))
(hflog_int : Integrable (fun p ↦ f p * Real.log (f p)) (μ.prod ν))
(hg1log_int : Integrable (fun x ↦ (∫ y, f (x, y) ∂ν) * Real.log (∫ y, f (x, y) ∂ν)) μ) :
((∫ p, f p * Real.log (f p) ∂(μ.prod ν))
- (∫ p, f p ∂(μ.prod ν)) * Real.log (∫ p, f p ∂(μ.prod ν)))
= ((∫ x, (∫ y, f (x, y) ∂ν) * Real.log (∫ y, f (x, y) ∂ν) ∂μ)
- (∫ x, (∫ y, f (x, y) ∂ν) ∂μ) * Real.log (∫ x, (∫ y, f (x, y) ∂ν) ∂μ))
+ ∫ x, ((∫ y, f (x, y) * Real.log (f (x, y)) ∂ν)
- (∫ y, f (x, y) ∂ν) * Real.log (∫ y, f (x, y) ∂ν)) ∂μ := by sorry