Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proposition 4 — comparison of calibrated regulation mechanisms

Disproved
WangGreenTech.Regulation.proposition_4

by Shuze Chen · Oct 29, 2026 · Mathlib 0df444a (Lean v4.33.1)

global-gamesoperations-researchp2o-gran-per-chapterp2o-plan-paperp2o-v1regulation

Let p=(ϵ,b,q,r,f1,fL,fH,mL,mH)p=(\epsilon,b,q,r,f_1,f_L,f_H,m_L,m_H)p=(ϵ,b,q,r,f1​,fL​,fH​,mL​,mH​) satisfy ϵ>0\epsilon>0ϵ>0, b≥0b\ge0b≥0, 0≤q<10\le q<10≤q<1, r>0r>0r>0, f1,fL>0f_1,f_L>0f1​,fL​>0, fH≥fLf_H\ge f_LfH​≥fL​, and mL≤mH≤0m_L\le m_H\le0mL​≤mH​≤0. Signals and the common fundamental are real. Conditional on signal xxx, the specified posterior is uniform on [x−ϵ,x+ϵ][x-\epsilon,x+\epsilon][x−ϵ,x+ϵ]. Followers form a unit-mass population on [0,1)[0,1)[0,1); the leader has zero population mass. Both choose action 1 at indifference.

The prescribed endogenous adoption threshold is t=b/2+fL−fH+rr+1(1−q)(fH−mL)t=b/2+f_L-f_H+\frac r{r+1}(1-q)(f_H-m_L)t=b/2+fL​−fH​+r+1r​(1−q)(fH​−mL​), with aggregate At(θ)=min⁡{1,max⁡{0,(θ+ϵ−t)/(2ϵ)}}A_t(\theta)=\min\{1,\max\{0,(\theta+\epsilon-t)/(2\epsilon)\}\}At​(θ)=min{1,max{0,(θ+ϵ−t)/(2ϵ)}}. Define

UD(λ,x)=12ϵ∫x−ϵx+ϵ[λ−f1+mL+θ−bAt(θ)+(q+(1−q)At(θ)r)(mH−mL)]dθ.U_D(\lambda,x)=\frac1{2\epsilon}\int_{x-\epsilon}^{x+\epsilon}\left[\lambda-f_1+m_L+\theta-bA_t(\theta)+(q+(1-q)A_t(\theta)^r)(m_H-m_L)\right]d\theta.UD​(λ,x)=2ϵ1​∫x−ϵx+ϵ​[λ−f1​+mL​+θ−bAt​(θ)+(q+(1−q)At​(θ)r)(mH​−mL​)]dθ.

The development set is Dλ={x:UD(λ,x)≥0}D_\lambda=\{x:U_D(\lambda,x)\ge0\}Dλ​={x:UD​(λ,x)≥0} and xaggr(1)(λ)=inf⁡Dλx_{\rm aggr}^{(1)}(\lambda)=\inf D_\lambdaxaggr(1)​(λ)=infDλ​.

For a fixed regulation probability 0<q^<10<\widehat q<10<q​<1, put t^=b/2+fL−(1−q^)mL−q^fH\widehat t=b/2+f_L-(1-\widehat q)m_L-\widehat q f_Ht=b/2+fL​−(1−q​)mL​−q​fH​ and

UF(λ,x)=λ−f1+mL+q^(mH−mL)+x−b2ϵ∫x−ϵx+ϵAt^(θ) dθ.U_F(\lambda,x)=\lambda-f_1+m_L+\widehat q(m_H-m_L)+x-\frac b{2\epsilon}\int_{x-\epsilon}^{x+\epsilon}A_{\widehat t}(\theta)\,d\theta.UF​(λ,x)=λ−f1​+mL​+q​(mH​−mL​)+x−2ϵb​∫x−ϵx+ϵ​At​(θ)dθ.

Set D^λ={x:UF(λ,x)≥0}\widehat D_\lambda=\{x:U_F(\lambda,x)\ge0\}Dλ​={x:UF​(λ,x)≥0} and x^(1)(λ)=inf⁡D^λ\widehat x^{(1)}(\lambda)=\inf\widehat D_\lambdax(1)(λ)=infDλ​.

Suppose b≤2ϵb\le2\epsilonb≤2ϵ, and calibrate the two regulation mechanisms by

q^=q+1−qr+1,\widehat q=q+\frac{1-q}{r+1},q​=q+r+11−q​,

which makes the period-2 adoption cutoffs equal, t=t^t=\widehat tt=t. With all these primitives fixed, there exists a real λ^\widehat\lambdaλ such that for every λ≥0\lambda\ge0λ≥0,

xaggr(1)(λ)≤x^(1)(λ)⟺λ≤λ^.x_{\rm aggr}^{(1)}(\lambda)\le\widehat x^{(1)}(\lambda)\quad\Longleftrightarrow\quad\lambda\le\widehat\lambda.xaggr(1)​(λ)≤x(1)(λ)⟺λ≤λ.

This is the paper's comparison of development incentives under equally effective adoption incentives.

Formalization Note This faithful item retains mH=mLm_H=m_LmH​=mL​ and a finite real comparison threshold, possibly negative. Finding F1 gives a counterexample on that permitted boundary: the cutoffs coincide for every λ\lambdaλ, so no finite λ^\widehat\lambdaλ works. It remains the mission goal; the corrected item adds a strict payoff gap. The existential comparison threshold precedes the quantifier over λ\lambdaλ. The Lean statement expands the posterior integrals and development-set infima for a self-contained blind audit; a separate, fully proved verification lemma checks that this expansion is definitionally identical to the canonical cutoff comparison.

Preamble
import Definitions.Def_WangGreenTech_Regulation_Policies

set_option autoImplicit false
Formal statement
namespace WangGreenTech.Regulation

open MeasureTheory Set

theorem proposition_4 (ε b q r f₁ fL fH mL mH qhat : ℝ)
    (hε : 0 < ε) (hb0 : 0 ≤ b) (hq0 : 0 ≤ q) (hq1 : q < 1) (hr : 0 < r)
    (hf₁ : 0 < f₁) (hfL : 0 < fL) (hfH : fL ≤ fH) (hm : mL ≤ mH) (hmH : mH ≤ 0)
    (hqh0 : 0 < qhat) (hqh1 : qhat < 1) (hb : b ≤ 2 * ε)
    (hcal : qhat = q + (1 - q) / (r + 1)) :
  let t : ℝ := b / 2 + fL - fH + r / (r + 1) * (1 - q) * (fH - mL)
  let that : ℝ := b / 2 + fL - (1 - qhat) * mL - qhat * fH
  let A : ℝ → ℝ → ℝ := fun cutoff θ => min 1 (max 0 ((θ + ε - cutoff) / (2 * ε)))
  let E : ℝ → (ℝ → ℝ) → ℝ := fun x h =>
    ∫ θ, h θ ∂((ENNReal.ofReal (2 * ε))⁻¹ • volume.restrict (Icc (x - ε) (x + ε)))
  let UD : ℝ → ℝ → ℝ := fun lam x => E x (fun θ =>
    lam - f₁ + mL + θ - b * A t θ + (q + (1 - q) * (A t θ) ^ r) * (mH - mL))
  let UF : ℝ → ℝ → ℝ := fun lam x => E x (fun θ =>
    lam - f₁ + θ - b * A that θ + qhat * mH + (1 - qhat) * mL)
  ∃ lamhat : ℝ, ∀ lam : ℝ, 0 ≤ lam →
    (sInf {x : ℝ | 0 ≤ UD lam x} ≤ sInf {x : ℝ | 0 ≤ UF lam x} ↔ lam ≤ lamhat) := by sorry

end WangGreenTech.Regulation
Source
Wang, Cho and Scheller-Wolf, Green Technology Development and Adoption: Competition, Regulation, and Uncertainty—A Global Game Approach, Management Science forthcoming artifact, https://doi.org/10.1287/mnsc.2019.3538, body p. 25 (PDF 25), Proposition 4; proof Online Appendix A pp. 13–14 (PDF 46–47)
Read-back

What the Lean code literally says, in plain math · claude-opus-5-5

The statement fixes ten real parameters ε,b,q,r,f1,fL,fH,mL,mH,q^\varepsilon, b, q, r, f_1, f_L, f_H, m_L, m_H, \hat qε,b,q,r,f1​,fL​,fH​,mL​,mH​,q^​ and assumes all of the following:

  • ε>0\varepsilon > 0ε>0;
  • 0≤b≤2ε0 \le b \le 2\varepsilon0≤b≤2ε;
  • 0≤q<10 \le q < 10≤q<1;
  • r>0r > 0r>0;
  • f1>0f_1 > 0f1​>0;
  • fL>0f_L > 0fL​>0 and fL≤fHf_L \le f_HfL​≤fH​;
  • mL≤mH≤0m_L \le m_H \le 0mL​≤mH​≤0;
  • 0<q^<10 < \hat q < 10<q^​<1;
  • the calibration identity
q^=q+1−qr+1.\hat q = q + \frac{1-q}{r+1}.q^​=q+r+11−q​.

Given 0≤q<10 \le q < 10≤q<1 and r>0r > 0r>0, this identity already forces 0<q^<10 < \hat q < 10<q^​<1. Everything else in the statement is defined locally from these parameters; no named definition from the imported bundle appears.

Two thresholds are defined:

t=b2+fL−fH+rr+1(1−q)(fH−mL),t^=b2+fL−(1−q^) mL−q^ fH.t = \frac b2 + f_L - f_H + \frac{r}{r+1}(1-q)(f_H - m_L), \qquad \hat t = \frac b2 + f_L - (1-\hat q)\,m_L - \hat q\, f_H.t=2b​+fL​−fH​+r+1r​(1−q)(fH​−mL​),t^=2b​+fL​−(1−q^​)mL​−q^​fH​.

For a cutoff ccc and a real θ\thetaθ, a clamped linear ramp is defined:

A(c,θ)=min⁡ ⁣(1, max⁡ ⁣(0, θ+ε−c2ε)).A(c,\theta) = \min\!\Big(1,\ \max\!\Big(0,\ \frac{\theta + \varepsilon - c}{2\varepsilon}\Big)\Big).A(c,θ)=min(1, max(0, 2εθ+ε−c​)).

It is 000 for θ≤c−ε\theta \le c - \varepsilonθ≤c−ε, rises linearly to 111 on [c−ε,c+ε][c-\varepsilon, c+\varepsilon][c−ε,c+ε], and is 111 for θ≥c+ε\theta \ge c+\varepsilonθ≥c+ε. This is the value at θ\thetaθ of the distribution function of the uniform law on [c−ε,c+ε][c-\varepsilon, c+\varepsilon][c−ε,c+ε].

For a real xxx and a function h:R→Rh:\mathbb R \to \mathbb Rh:R→R, an averaging operator is defined:

Ex[h]=∫h(θ) dμx(θ),μx=12ε Leb∣[x−ε, x+ε].E_x[h] = \int h(\theta)\, d\mu_x(\theta), \qquad \mu_x = \frac{1}{2\varepsilon}\,\mathrm{Leb}\big|_{[x-\varepsilon,\,x+\varepsilon]}.Ex​[h]=∫h(θ)dμx​(θ),μx​=2ε1​Leb​[x−ε,x+ε]​.

This is the expectation of h(θ)h(\theta)h(θ) for θ\thetaθ uniform on [x−ε,x+ε][x-\varepsilon, x+\varepsilon][x−ε,x+ε], computed as a Bochner integral.

Two payoff functions of λ\lambdaλ and xxx are then defined:

UD(λ,x)=Ex[λ−f1+mL+θ−b A(t,θ)+(q+(1−q) A(t,θ)r)(mH−mL)],U_D(\lambda,x) = E_x\Big[\lambda - f_1 + m_L + \theta - b\,A(t,\theta) + \big(q + (1-q)\,A(t,\theta)^{r}\big)(m_H - m_L)\Big],UD​(λ,x)=Ex​[λ−f1​+mL​+θ−bA(t,θ)+(q+(1−q)A(t,θ)r)(mH​−mL​)], UF(λ,x)=Ex[λ−f1+θ−b A(t^,θ)+q^ mH+(1−q^) mL].U_F(\lambda,x) = E_x\Big[\lambda - f_1 + \theta - b\,A(\hat t,\theta) + \hat q\, m_H + (1-\hat q)\, m_L\Big].UF​(λ,x)=Ex​[λ−f1​+θ−bA(t^,θ)+q^​mH​+(1−q^​)mL​].

In UDU_DUD​, the power A(t,θ)rA(t,\theta)^rA(t,θ)r is the real power with real exponent rrr, applied to a base in [0,1][0,1][0,1].

The conclusion is that there exists a real number λ^\hat\lambdaλ^ such that for every λ≥0\lambda \ge 0λ≥0,

inf⁡{x∈R:UD(λ,x)≥0}  ≤  inf⁡{x∈R:UF(λ,x)≥0}⟺λ≤λ^.\inf\{x \in \mathbb R : U_D(\lambda,x) \ge 0\} \;\le\; \inf\{x \in \mathbb R : U_F(\lambda,x) \ge 0\} \quad\Longleftrightarrow\quad \lambda \le \hat\lambda .inf{x∈R:UD​(λ,x)≥0}≤inf{x∈R:UF​(λ,x)≥0}⟺λ≤λ^.

Some consequences are immediate from the form of this claim:

  • λ^\hat\lambdaλ^ is not required to be nonnegative.
  • The equivalence is only asserted for λ≥0\lambda \ge 0λ≥0; nothing is said about negative λ\lambdaλ.
  • No real λ^\hat\lambdaλ^ satisfies λ≤λ^\lambda \le \hat\lambdaλ≤λ^ for every λ≥0\lambda \ge 0λ≥0. The statement therefore implies that the set of λ≥0\lambda \ge 0λ≥0 at which the left-hand comparison holds is either empty or exactly [0,λ^][0,\hat\lambda][0,λ^]. In particular, the comparison fails for all sufficiently large λ\lambdaλ.
  • If that set is empty, any negative λ^\hat\lambdaλ^ witnesses the claim, so λ^\hat\lambdaλ^ is then not unique.

Degenerate and default-value cases:

  • No junk divisions. The hypotheses permit b=0b = 0b=0, which removes the ramp penalty terms, as well as q=0q = 0q=0 and mL=mHm_L = m_HmL​=mH​, which makes the factor (mH−mL)(m_H - m_L)(mH​−mL​) vanish. They also permit any r>0r > 0r>0, including non-integers, and fH=fLf_H = f_LfH​=fL​. Since ε>0\varepsilon > 0ε>0 and r+1>1r + 1 > 1r+1>1, the divisions by 2ε2\varepsilon2ε and by r+1r+1r+1 are genuine, and the averaging interval is nondegenerate with μx\mu_xμx​ a probability measure. There is no natural-number subtraction and there are no empty or one-element types.
  • No junk integrals. Each integrand is continuous and bounded in θ\thetaθ on the bounded interval [x−ε,x+ε][x-\varepsilon, x+\varepsilon][x−ε,x+ε]; the power ArA^rAr is continuous on [0,1][0,1][0,1], with 0r=00^r = 00r=0, because r>0r > 0r>0. So none of the integrals falls back to the default value 000 used for non-integrable functions.
  • No junk infima. The real infimum would return the default value 000 if a set were empty or unbounded below. This does not happen here. In each payoff, θ\thetaθ averages to xxx and every other term is bounded, so UD(λ,x)U_D(\lambda,x)UD​(λ,x) and UF(λ,x)U_F(\lambda,x)UF​(λ,x) equal xxx plus a bounded function of xxx. Hence each set {x:U(λ,x)≥0}\{x : U(\lambda,x) \ge 0\}{x:U(λ,x)≥0} is nonempty and bounded below, and both infima are genuine.
Human review
  • Endorsed by Community (Bot) · Oct 29, 2026

    Confirmed by the moderator at approval.

  • Endorsed by Shuze Chen · Oct 29, 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