Lemma EC.2, p. ec3 — strongest bound by regime: sign patterns and unique roots at δL and δM
ProvedValuePersPricing.Deviation.lemma_EC_2lambert-wp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1personalized-pricing
Let and let be the thresholds at :
- The function
is negative for , positive for , and is its unique root in . 2. The function
vanishes at , is non-negative for all , and is its unique root in .
Consequently, in the standardized case, the low bound (EC.5) is the strongest on , the medium bound (EC.6) on , and the high bound (EC.7) on .
Differences from the source:
- encoding: (b) "non-negative for all D ∈ [0, δH]" → for D ∈ (0, δH]; reason: log(S)/D is undefined at D = 0.
- encoding: no hypothesis on S → 1 < S; reason: the lemma sits inside the proof of Theorem 2, whose standing hypothesis is S > 1; for S = 1 all thresholds are 0 and log S = 0.
Preamble
import Mathlib import Definitions.Def_ValuePersPricing_Deviation_Model open MeasureTheory Set Filter Topology
Formal statement
namespace ValuePersPricing.Deviation
/-- Lemma EC.2 (p. ec3), for `S > 1` with `δ_L, δ_M, δ_H` at `M = 1`:
(a) `f(D) = −W₋₁(−(1 − D)/e)/(1 − D) − log(S)/D` is negative on `(0, δ_L)`, positive on
`(δ_L, δ_H]`, and `δ_L` is its unique root in `(0, δ_H]`;
(b) `g(D) = log(S)/D + W₋₁(−1/(e S (1 − D)))` vanishes at `δ_M`, is non-negative on `(0, δ_H]`,
and `δ_M` is its unique root there. -/
theorem lemma_EC_2 (S : ℝ) (hS : 1 < S) :
((∀ D ∈ Ioo 0 (deltaL 1 S),
-lambertWm1 (-(1 - D) / Real.exp 1) / (1 - D) - Real.log S / D < 0) ∧
(∀ D ∈ Ioc (deltaL 1 S) (deltaH 1 S),
0 < -lambertWm1 (-(1 - D) / Real.exp 1) / (1 - D) - Real.log S / D) ∧
-lambertWm1 (-(1 - deltaL 1 S) / Real.exp 1) / (1 - deltaL 1 S)
- Real.log S / deltaL 1 S = 0 ∧
(∀ D ∈ Ioc 0 (deltaH 1 S),
-lambertWm1 (-(1 - D) / Real.exp 1) / (1 - D) - Real.log S / D = 0 →
D = deltaL 1 S)) ∧
(Real.log S / deltaM 1 S
+ lambertWm1 (-1 / (Real.exp 1 * S * (1 - deltaM 1 S))) = 0 ∧
(∀ D ∈ Ioc 0 (deltaH 1 S),
0 ≤ Real.log S / D + lambertWm1 (-1 / (Real.exp 1 * S * (1 - D)))) ∧
(∀ D ∈ Ioc 0 (deltaH 1 S),
Real.log S / D + lambertWm1 (-1 / (Real.exp 1 * S * (1 - D))) = 0 →
D = deltaM 1 S)) := by sorry
end ValuePersPricing.Deviation
Source
Elmachtoub, Gupta & Hamilton, The Value of Personalized Pricing, SSRN 3127719 (review version, PDF sha256 635e22eb…), p. ec3 (PDF 35), Lemma EC.2; proof pp. ec6–ec7 (PDF 38–39)
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.