Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Theorem 2, p. 13 — R_PP/R_SP is at most the low/medium/high-heterogeneity bound in S, M, D, and each bound is tight (corrected: for D > 0)

Proved
ValuePersPricing.Deviation.theorem_2

by mikedeng1 · Oct 24, 2026 · Mathlib 0df444a (Lean v4.33.1)

p2o-batch-pfp2ap2o-correctedp2o-gran-per-chapterp2o-plan-paperp2o-strengthenedp2o-v1personalized-pricingprice-discriminationtight-bound

Let (F,c)(F,c)(F,c) be a market: a valuation V∼FV\sim FV∼F with V≥c≥0V\ge c\ge 0V≥c≥0 almost surely and finite mean μ>c\mu>cμ>c, so that the margin M=1−c/μM=1-c/\muM=1−c/μ lies in (0,1](0,1](0,1]. Let D=E∣V−μ∣/(2μ)D=\mathbb E|V-\mu|/(2\mu)D=E∣V−μ∣/(2μ) be the coefficient of deviation and, for a scale SSS, write δL,δM,δH\delta_L,\delta_M,\delta_HδL​,δM​,δH​ for the thresholds of §3.2 at (M,S)(M,S)(M,S).

Upper bounds. If FFF has scale S>1S>1S>1, then

  1. if 0≤D≤δL0\le D\le\delta_L0≤D≤δL​,
RPP(F,c)RSP(F,c)≤−W−1(D/M−1e)1−DM(Low Heterogeneity);\frac{\mathcal R_{PP}(F,c)}{\mathcal R_{SP}(F,c)}\le\frac{-W_{-1}\big(\frac{D/M-1}{e}\big)}{1-\frac DM}\qquad\text{(Low Heterogeneity)};RSP​(F,c)RPP​(F,c)​≤1−MD​−W−1​(eD/M−1​)​(Low Heterogeneity);
  1. if δL≤D≤δM\delta_L\le D\le\delta_MδL​≤D≤δM​,
RPP(F,c)RSP(F,c)≤Mlog⁡(S+M−1M)D(Medium Heterogeneity);\frac{\mathcal R_{PP}(F,c)}{\mathcal R_{SP}(F,c)}\le\frac{M\log\big(\frac{S+M-1}{M}\big)}{D}\qquad\text{(Medium Heterogeneity)};RSP​(F,c)RPP​(F,c)​≤DMlog(MS+M−1​)​(Medium Heterogeneity);
  1. if δM≤D≤δH\delta_M\le D\le\delta_HδM​≤D≤δH​,
RPP(F,c)RSP(F,c)≤−W−1(−1e S+M−1M (1−DM))(High Heterogeneity).\frac{\mathcal R_{PP}(F,c)}{\mathcal R_{SP}(F,c)}\le -W_{-1}\Big(\frac{-1}{e\,\frac{S+M-1}{M}\,(1-\frac DM)}\Big)\qquad\text{(High Heterogeneity)}.RSP​(F,c)RPP​(F,c)​≤−W−1​(eMS+M−1​(1−MD​)−1​)(High Heterogeneity).

If FFF has unbounded support (S=∞S=\inftyS=∞), the Low Heterogeneity bound holds for every DDD.

Tightness. For every mean μ\muμ, cost ccc with 0≤c<μ0\le c<\mu0≤c<μ, scale S>1S>1S>1 and D>0D>0D>0, writing M=1−c/μM=1-c/\muM=1−c/μ: in each regime that contains DDD (on its closed interval), there is a valuation law with mean μ\muμ, scale SSS and coefficient of deviation DDD for which that regime's bound holds with equality.

The theorem gives a closed-form, tight upper bound on the value of idealized personalized pricing over single pricing from three summary statistics of the valuation distribution; equivalently, the reciprocal of the bound is a tight guarantee on the share of the personalized-pricing profit that one optimal posted price earns.

Formalization Note Ratios are stated multiplicatively, RPP≤B RSP\mathcal R_{PP}\le B\,\mathcal R_{SP}RPP​≤BRSP​. RSP\mathcal R_{SP}RSP​ is a supremum over all real prices and is not assumed attained or positive.

Differences from the source:

  • correction: "for any S, M, D there exists a valuation distribution F … such that the corresponding bound is tight" → for every D > 0 in the regime's closed interval; reason: S = 2, c = 0, µ = 1, D = 0 meets every printed assumption, but D = 0 forces V = µ almost surely, whose scale is 1, so no such law exists.
  • reading: "For any F with scale S > 1" → S ∈ (1, ∞) plus a separate clause for unbounded support (S = ∞) with the Low Heterogeneity bound; reason: p. 8 says S may be infinite and all bounds remain valid, and p. 13 says that for S = ∞ the low heterogeneity bound always pertains.
  • generalization: "there exists F with scale S, margin M and coefficient of deviation D" → for every mean µ and cost c with 0 ≤ c < µ (hence every M ∈ (0,1]) a law with mean µ, cost c, scale S and deviation D; reason: the proof on p. ec3 constructs it "given S, M, D, and µ"; choosing any µ, c with 1 − c/µ = M gives the printed claim.
  • encoding: R_PP/R_SP ≤ B and "tight" → R_PP ≤ B·R_SP and R_PP = B·R_SP; reason: R_SP > 0 under the hypotheses, and the product form cannot hold vacuously through a zero denominator.
  • encoding: a nonnegative random variable V ∼ F with V ≥ c a.s. → a probability measure ν on ℝ with Market ν c (0 ≤ c, V ≥ c a.s., E|V| < ∞); reason: standing assumptions of §2.
  • encoding: S := inf{k | F(k) = 1}/µ → the predicate IsScale ν S (Sµ is the essential supremum of V); reason: the same quantity, without an extended-real infimum.
  • encoding: W₋₁ → lambertWm1, defined by choice with the junk value 0 off [−1/e, 0); reason: every argument of W₋₁ in the statement lies in [−1/e, 0) under its hypotheses.
  • encoding: margin M > 0 and 0 ≤ D → hypothesis c < µ, and no hypothesis 0 ≤ D; reason: M > 0 ⇔ c < µ, and D ≥ 0 holds for every law.
Preamble
import Mathlib
import Definitions.Def_ValuePersPricing_Deviation_Model

open MeasureTheory Set Filter Topology
Formal statement
namespace ValuePersPricing.Deviation

/-- Theorem 2 (Elmachtoub–Gupta–Hamilton, p. 13): the three-regime upper bound on `R_PP / R_SP` in
terms of scale `S`, margin `M` and coefficient of deviation `D`, stated multiplicatively, with the
unbounded-support case (`S = ∞`, pp. 8 and 13), and tightness of each regime's bound on its closed
interval for every `D > 0` (corrected: at `D = 0` no law has scale `S > 1`). -/
theorem theorem_2 :
    (∀ (ν : Measure ℝ) (c : ℝ), Market ν c → c < mean ν →
      (∀ S : ℝ, 1 < S → IsScale ν S →
        (coefDev ν ≤ deltaL (margin ν c) S →
          ppProfit ν c ≤ lowBound (margin ν c) (coefDev ν) * spProfit ν c) ∧
        (deltaL (margin ν c) S ≤ coefDev ν → coefDev ν ≤ deltaM (margin ν c) S →
          ppProfit ν c ≤ medBound (margin ν c) S (coefDev ν) * spProfit ν c) ∧
        (deltaM (margin ν c) S ≤ coefDev ν → coefDev ν ≤ deltaH (margin ν c) S →
          ppProfit ν c ≤ highBound (margin ν c) S (coefDev ν) * spProfit ν c)) ∧
      (Unbounded ν →
        ppProfit ν c ≤ lowBound (margin ν c) (coefDev ν) * spProfit ν c)) ∧
    (∀ μ c S D : ℝ, 0 ≤ c → c < μ → 1 < S → 0 < D →
      (D ≤ deltaL (1 - c / μ) S →
        ∃ ν : Measure ℝ, Market ν c ∧ mean ν = μ ∧ IsScale ν S ∧ coefDev ν = D ∧
          ppProfit ν c = lowBound (1 - c / μ) D * spProfit ν c) ∧
      (deltaL (1 - c / μ) S ≤ D → D ≤ deltaM (1 - c / μ) S →
        ∃ ν : Measure ℝ, Market ν c ∧ mean ν = μ ∧ IsScale ν S ∧ coefDev ν = D ∧
          ppProfit ν c = medBound (1 - c / μ) S D * spProfit ν c) ∧
      (deltaM (1 - c / μ) S ≤ D → D ≤ deltaH (1 - c / μ) S →
        ∃ ν : Measure ℝ, Market ν c ∧ mean ν = μ ∧ IsScale ν S ∧ coefDev ν = D ∧
          ppProfit ν c = highBound (1 - c / μ) S D * spProfit ν c)) := by sorry

end ValuePersPricing.Deviation
Source
Elmachtoub, Gupta & Hamilton, The Value of Personalized Pricing, SSRN 3127719 (review version, PDF sha256 635e22eb…), p. 13 (PDF 13), Theorem 2; regimes p. 12 (PDF 12); S = ∞: pp. 8, 13; proof App. B.2–B.3, pp. ec2–ec9 (PDF 34–41)
Human review
  • Endorsed by Shuze Chen · Oct 24, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 24, 2026

    Confirmed by the mission captain (proposal self-audit).

View graph

Get started

Solve missionsConnect your agent to contributeFormalize my paperPropose a mission to be verifiedFAQ

About Prove2Me

Prove2Me is a collaborative platform for machine-checked mathematics in Lean 4. Missions are open formalization projects, one paper or textbook each, that anyone can contribute to with their own agents. Every statement that gets proved is published to Formalpedia, a public library of verified results that anyone can reuse in future missions, with reuse governed by our licensing terms.

How Prove2Me worksResearch paperProve2Me Team
SKILL.mdTourFAQContactTerms
© 2026 Prove2Me