Sign transfer from the log-density difference to the density difference
Provedsign_transfer_loglogarithmreal-analysissiegel-median
Sign-transfer from the log-density difference to the density difference (Siegel 2001, Theorem 2.1 / 2.2, p.5-6). For a strictly positive density , the sign of coincides with the sign of (since is strictly increasing on ). Hence a single-crossing point for — on and on — yields the same sign data on , on . Siegel works the unimodality argument on (its derivative is rational in with a quadratic numerator), then transfers back to ; this lemma is exactly that transfer.
Preamble
import Mathlib.Analysis.SpecialFunctions.Log.Basic import Mathlib.Topology.Order.IntermediateValue set_option autoImplicit false open Set
Formal statement
theorem sign_transfer_log
(f : ℝ → ℝ) (μ c : ℝ)
(hf : ∀ x, 0 < f x)
(hL : ∀ x ∈ Icc (0:ℝ) c, Real.log (f x) - Real.log (f (2*μ - x)) ≤ 0)
(hR : ∀ x ∈ Icc c μ, 0 ≤ Real.log (f x) - Real.log (f (2*μ - x))) :
(∀ x ∈ Icc (0:ℝ) c, f x - f (2*μ - x) ≤ 0) ∧
(∀ x ∈ Icc c μ, 0 ≤ f x - f (2*μ - x)) := by sorrySource
A. Siegel, "Median Bounds and their Application", J. Algorithms 38:184-236, 2001, Theorem 2.1 / Theorem 2.2, p.5-6 (the argument is carried on the log-density and transferred to the density via monotonicity of log).