Exact source context

import AutoSamplingTheory.TechnicalLemmas.StochasticProcesses.CarreDuChamp

/-!
# The algebraic Bakry--Émery interpolation bound

For a linear generator, the definition of the iterated carré du champ gives

`2 Γ₂(f) = L Γ(f) - 2 Γ(f, Lf)`.

Combining this identity with `CD(κ, ∞)` supplies the lower derivative bound
used in the standard backward semigroup interpolation.  This file proves that
algebraic implication without assuming a concrete semigroup or its derivative
formula.
-/

namespace AutoSamplingTheory
namespace TechnicalLemmas
namespace StochasticProcesses

open CarreDuChamp

variable {X : Type*}

/-- The fully expanded diagonal `CD(alpha, infinity)` inequality gives the
generator expression used in backward interpolation.

To keep the public interface directly inspectable, the statement exposes
positivity of `alpha`, every curvature quantifier, and the exact one-half
normalization of the iterated carré du champ in its signature. -/
theorem interpolationDerivative_lower_bound_of_expandedBakryEmery
    (generator : (X → ℝ) →ₗ[ℝ] (X → ℝ))
    {alpha : ℝ} (_halpha : 0 < alpha)
    (hcurvature : ∀ (g : X → ℝ) (y : X),
      alpha * carreDuChamp generator g g y ≤
        (2 : ℝ)⁻¹ *
          (generator (carreDuChamp generator g g) y -
            carreDuChamp generator g (generator g) y -
            carreDuChamp generator g (generator g) y))
    (f : X → ℝ) (x : X) :
    2 * alpha * carreDuChamp generator f f x ≤
      generator (carreDuChamp generator f f) x -
        2 * carreDuChamp generator f (generator f) x := by
  have h := hcurvature f x
  nlinarith

end StochasticProcesses
end TechnicalLemmas
end AutoSamplingTheory