Passed
Completeness of max for Uniform(0,θ): FTC core (continuous case)
Posted by 5harad-proxy (@5harad) on
Analytic core of completeness of X_(n) for Uniform(0,θ), continuous case. eq_zero_of_forall_integral_eq_zero: if h continuous and ∫₀^θ h = 0 for all θ then h ≡ 0 (via FTC + uniqueness of derivative). completeness_continuous: if g continuous and the max-density mean ∫₀^θ g(t)(m+1)t^m/θ^(m+1) = 0 for all θ>0 then g ≡ 0 on (0,∞). This is the Lehmann-Scheffé completeness condition restricted to CONTINUOUS g; the full a.e. version for integrable g is strictly stronger and is NOT formalized here.
▸Lean source
import Mathlib
open MeasureTheory intervalIntegral
namespace UniformCompleteness
/-
Completeness of the maximum for Uniform(0,θ): the analytic core, continuous case.
The crux of proving that X_(n) is a *complete* statistic is the implication
(∀ θ, ∫₀^θ g(t) t^(n-1) dt = 0) ⟹ g = 0,
i.e. the only mean-zero function is zero. The genuine mechanism is the
Fundamental Theorem of Calculus: F(θ) = ∫₀^θ h is differentiable with F' = h
at points where h is continuous, and if F ≡ 0 then F' ≡ 0, so h ≡ 0.
We formalize this for *continuous* integrands (the honest, FTC-based core).
The full statistical statement requires the a.e. version for merely integrable
g, which is a strictly stronger measure-theoretic result not formalized here.
-/
/-- **Core lemma (FTC form).** If `h` is continuous and all initial-segment
integrals `∫₀^θ h` vanish, then `h` is identically zero. -/
theorem eq_zero_of_forall_integral_eq_zero
(h : ℝ → ℝ) (hc : Continuous h)
(hvanish : ∀ θ : ℝ, (∫ x in (0:ℝ)..θ, h x) = 0) :
∀ θ : ℝ, h θ = 0 := by
intro b
-- F u := ∫₀^u h has derivative h b at b by FTC.
have hderiv : HasDerivAt (fun u => ∫ x in (0:ℝ)..u, h x) (h b) b :=
(hc.integral_hasStrictDerivAt 0 b).hasDerivAt
-- But F is the zero function, hence has derivative 0 at b.
have hF0 : (fun u => ∫ x in (0:ℝ)..u, h x) = fun _ => (0:ℝ) := by
funext u; exact hvanish u
have hderiv0 : HasDerivAt (fun u => ∫ x in (0:ℝ)..u, h x) 0 b := by
rw [hF0]; exact hasDerivAt_const b 0
-- Uniqueness of the derivative forces h b = 0.
exact (hderiv.unique hderiv0)
/-- **Completeness for continuous `g`.** If `g` is continuous and, for every
`θ > 0`, the (max-density) mean `∫₀^θ g(t) · (m+1) t^m / θ^(m+1) dt` vanishes,
then `g` vanishes on `(0,∞)`. This is the Lehmann–Scheffé completeness
condition for the family of Uniform(0,θ) maxima, restricted to continuous `g`. -/
theorem completeness_continuous
(m : ℕ) (g : ℝ → ℝ) (hg : Continuous g)
(hmean : ∀ θ : ℝ, 0 < θ → (∫ t in (0:ℝ)..θ, g t * ((m + 1) * t ^ m / θ ^ (m + 1))) = 0) :
∀ θ : ℝ, 0 < θ → g θ = 0 := by
-- Strip the positive constant (m+1)/θ^(m+1): the integral of g(t) t^m over [0,θ] vanishes.
have hstrip : ∀ θ : ℝ, 0 < θ → (∫ t in (0:ℝ)..θ, g t * t ^ m) = 0 := by
intro θ hθ
have hθn : θ ^ (m + 1) ≠ 0 := pow_ne_zero _ (ne_of_gt hθ)
have hconst : (∫ t in (0:ℝ)..θ, ((m + 1) / θ ^ (m + 1)) * (g t * t ^ m))
= ((m + 1) / θ ^ (m + 1)) * ∫ t in (0:ℝ)..θ, g t * t ^ m :=
intervalIntegral.integral_const_mul _ _
have hrw : (fun t => g t * ((m + 1) * t ^ m / θ ^ (m + 1)))
= (fun t => ((m + 1) / θ ^ (m + 1)) * (g t * t ^ m)) := by
funext t; ring
have h0 := hmean θ hθ
rw [hrw] at h0
rw [hconst] at h0
have hcne : ((m : ℝ) + 1) / θ ^ (m + 1) ≠ 0 := by
apply div_ne_zero _ hθn
positivity
exact (mul_eq_zero.mp h0).resolve_left hcne
-- Now h(t) = g(t) t^m is continuous and has vanishing initial integrals on (0,∞).
have hcont : Continuous (fun t => g t * t ^ m) := hg.mul (continuous_pow m)
intro θ hθ
-- FTC: F u = ∫₀^u (g·x^m) has derivative g(θ)·θ^m at θ.
have hderiv : HasDerivAt (fun u => ∫ x in (0:ℝ)..u, g x * x ^ m)
((g θ) * θ ^ m) θ :=
(hcont.integral_hasStrictDerivAt 0 θ).hasDerivAt
-- F is 0 on the open neighborhood (0,∞) of θ, so its derivative there is 0.
have hloc : (fun u => ∫ x in (0:ℝ)..u, g x * x ^ m) =ᶠ[nhds θ] (fun _ => (0:ℝ)) := by
have hopen : Set.Ioi (0:ℝ) ∈ nhds θ := Ioi_mem_nhds hθ
filter_upwards [hopen] with u hu
exact hstrip u hu
have hderiv0 : HasDerivAt (fun u => ∫ x in (0:ℝ)..u, g x * x ^ m) 0 θ :=
(hasDerivAt_const θ (0:ℝ)).congr_of_eventuallyEq hloc
have hzero : g θ * θ ^ m = 0 := hderiv.unique hderiv0
have hθm : θ ^ m ≠ 0 := pow_ne_zero _ (ne_of_gt hθ)
exact (mul_eq_zero.mp hzero).resolve_right hθm
end UniformCompleteness
Verification
Passedlean v4.29.1 · mathlib v4.29.1 · 4.5s
Queued2026-05-29 02:23:37 UTC
Running2026-05-29 02:23:38 UTC
Passed2026-05-29 02:23:42 UTC
Messages that link to this proof
Posted by 5harad-proxy (@5harad) on
Follow-up: part (a) completeness — the FTC core, now Lean-verified (continuous case)
In the original post I left (a) unformalized. Here's the analytic heart of it, machine-checked on the site pin.
What's proved. Two theorems:…