Prove2Me
Navigate
DiscoverCollectionsFormalpediaGalaxyBlogsUsersMomentumMy Missions+
Prove2Me
Log in
← Formalpedia

Proof of Theorem 4 — linear growth of Ĵ_t on the event A_t

Proved
KeskinZeevi.IncumbentSufficient.Jhat_linear_growth_on_goodEvent

by mikedeng1 · Oct 3, 2026 · Mathlib 0df444a (Lean v4.33.1)

dynamic-pricingleast-squaresp2o-batch-rm-cafterp2o-gran-per-chapterp2o-plan-paperp2o-v1

Fix a pricing policy, a slope β\betaβ, a constant κ1\kappa_1κ1​, a period t≥3t \ge 3t≥3 and a realization of the shocks. Suppose that condition (ii) of Theorem 4 holds at time t−1t - 1t−1 along this realization:

∑s=1t−1(ψ(bs)−xs+1)2≤κ1log⁡(t−1).\sum_{s=1}^{t-1} (\psi(b_s) - x_{s+1})^2 \le \kappa_1 \log(t-1).s=1∑t−1​(ψ(bs​)−xs+1​)2≤κ1​log(t−1).

If the realization lies in the event AtA_tAt​ (the greedy deviations ψ(bs)\psi(b_s)ψ(bs​) for ⌊(t−1)/2⌋≤s≤t−1\lfloor (t-1)/2\rfloor \le s \le t-1⌊(t−1)/2⌋≤s≤t−1 are all within ∣ψ(β)∣/12|\psi(\beta)|/\sqrt{12}∣ψ(β)∣/12​ of ψ(β)\psi(\beta)ψ(β)), then

J^t≥((ψ(β))28−κ1log⁡(t−1)t−1)(t−1).\hat J_t \ge \Big( \frac{(\psi(\beta))^2}{8} - \frac{\kappa_1 \log(t-1)}{t-1} \Big)(t-1).J^t​≥(8(ψ(β))2​−t−1κ1​log(t−1)​)(t−1).

So the information metric grows linearly in ttt on AtA_tAt​ when ψ(β)≠0\psi(\beta) \ne 0ψ(β)=0: a policy that tracks the greedy price, which is bounded away from the uninformative incumbent price, accumulates information at a constant rate.

Formalization Note. The statement is deterministic (pathwise). The paper applies it for t≥N(β)≥5t \ge N(\beta) \ge 5t≥N(β)≥5; it is stated here for t≥3t \ge 3t≥3, where ⌊(t−1)/2⌋≥1\lfloor (t-1)/2 \rfloor \ge 1⌊(t−1)/2⌋≥1 so that the sum in (ii) starts at s=1s = 1s=1 and covers every index of AtA_tAt​. The sum in condition (ii) starts at s=1s = 1s=1 rather than the paper's s=0s = 0s=0, because b0b_0b0​ is defined only for greedy ILS. log⁡\loglog is the natural logarithm, applied to the real number t−1≥2t - 1 \ge 2t−1≥2.

Preamble
import Mathlib
import Definitions.Def_KeskinZeevi_IncumbentSufficient_LeastSquares
Formal statement
namespace KeskinZeevi.IncumbentSufficient

/-- Keskin–Zeevi 2014, p. 1159, proof of Theorem 4, display after (A16): pathwise, for `t ≥ 3`,
if condition (ii) holds at time `t - 1`, i.e. `Σ_{s=1}^{t-1} (ψ(b_s) - x_{s+1})² ≤ κ₁ log(t - 1)`,
then on the event `A_t` of (A15)
`Ĵ_t ≥ (ψ(β)²/8 - κ₁ log(t - 1)/(t - 1)) (t - 1)`. -/
theorem Jhat_linear_growth_on_goodEvent (M : Model) {Ω : Type*} (π : Policy M) (β : ℝ)
    (ε : ℕ → Ω → ℝ) (κ₁ : ℝ) (t : ℕ) (ht : 3 ≤ t) (ω : Ω)
    (hii : ∑ s ∈ Finset.Icc 1 (t - 1),
        (M.psi (truncEst π β ε s ω) - dev π β ε (s + 1) ω) ^ 2 ≤ κ₁ * Real.log ((t : ℝ) - 1))
    (hA : ω ∈ goodEvent π β ε t) :
    (M.psi β ^ 2 / 8 - κ₁ * Real.log ((t : ℝ) - 1) / ((t : ℝ) - 1)) * ((t : ℝ) - 1) ≤
      Jhat π β ε t ω := by sorry

end KeskinZeevi.IncumbentSufficient
Source
Keskin and Zeevi, Dynamic Pricing with an Unknown Demand Model, Operations Research 62(5), 2014, p. 1159, Appendix A.3, proof of Theorem 4, first display after Eq. (A16) (with the events A_t of Eq. (A15))
Read-back

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

Setting. Take any model MMM, any type Ω\OmegaΩ, any policy π\piπ, any real β\betaβ (not required to lie in [bmin⁡,bmax⁡][b_{\min}, b_{\max}][bmin​,bmax​]), any noise functions εs:Ω→R\varepsilon_s : \Omega \to \mathbb Rεs​:Ω→R, any real κ1\kappa_1κ1​ (no sign assumed), any natural t≥3t \ge 3t≥3 and any ω∈Ω\omega \in \Omegaω∈Ω.

Notation. Everything is evaluated along ω\omegaω for the process that π\piπ generates under β\betaβ and ε\varepsilonε:

  • xs=ps−p^x_s = p_s - \hat pxs​=ps​−p^​ are the price deviations;
  • bs=max⁡(bmin⁡,min⁡(β^s,bmax⁡))b_s = \max(b_{\min}, \min(\hat\beta_s, b_{\max}))bs​=max(bmin​,min(β^​s​,bmax​)) are the truncated least-squares estimates, where β^s\hat\beta_sβ^​s​ is set to 000 when ∑r≤sxr2=0\sum_{r \le s} x_r^2 = 0∑r≤s​xr2​=0;
  • ψ(b)=−p^/2−D^/(2b)\psi(b) = -\hat p/2 - \hat D/(2b)ψ(b)=−p^​/2−D^/(2b).

Hypotheses.

  1. The cumulative pricing error up to period t−1t-1t−1 is small:
∑s=1t−1(ψ(bs)−xs+1)2≤κ1log⁡(t−1).\sum_{s=1}^{t-1} \bigl(\psi(b_s) - x_{s+1}\bigr)^2 \le \kappa_1 \log(t-1).s=1∑t−1​(ψ(bs​)−xs+1​)2≤κ1​log(t−1).
  1. ω∈At\omega \in A_tω∈At​. That is, for every natural sss with ⌊(t−1)/2⌋≤s≤t−1\lfloor (t-1)/2 \rfloor \le s \le t-1⌊(t−1)/2⌋≤s≤t−1:
(ψ(β)−ψ(bs))2≤ψ(β)212.\bigl(\psi(\beta) - \psi(b_s)\bigr)^2 \le \frac{\psi(\beta)^2}{12}.(ψ(β)−ψ(bs​))2≤12ψ(β)2​.

Conclusion.

(ψ(β)28−κ1log⁡(t−1)t−1)(t−1)≤J^t=∑s=1txs2.\left(\frac{\psi(\beta)^2}{8} - \frac{\kappa_1 \log(t-1)}{t-1}\right)(t-1) \le \hat J_t = \sum_{s=1}^t x_s^2.(8ψ(β)2​−t−1κ1​log(t−1)​)(t−1)≤J^t​=s=1∑t​xs2​.

This is a pathwise statement. No measure is involved.

Degenerate cases.

  • Well-definedness: since t≥3t \ge 3t≥3, the real number t−1≥2t - 1 \ge 2t−1≥2. So there is no division by zero, and log⁡(t−1)>0\log(t-1) > 0log(t−1)>0.
  • κ1<0\kappa_1 < 0κ1​<0: hypothesis 1 cannot hold, because its left side is ≥0\ge 0≥0 and its right side is <0< 0<0. The statement is then vacuous.
  • κ1=0\kappa_1 = 0κ1​=0: hypothesis 1 forces ψ(bs)=xs+1\psi(b_s) = x_{s+1}ψ(bs​)=xs+1​ for all 1≤s≤t−11 \le s \le t-11≤s≤t−1.
  • ψ(β)=0\psi(\beta) = 0ψ(β)=0: this is allowed, including the case β=0\beta = 0β=0 with p^=0\hat p = 0p^​=0. Hypothesis 2 then forces ψ(bs)=0\psi(b_s) = 0ψ(bs​)=0 in the range. The conclusion becomes −κ1log⁡(t−1)≤J^t-\kappa_1 \log(t-1) \le \hat J_t−κ1​log(t−1)≤J^t​.
  • Index range for t=3t = 3t=3: the range in AtA_tAt​ is s∈{1,2}s \in \{1, 2\}s∈{1,2}, and it never includes s=0s = 0s=0 here.
  • Empty Ω\OmegaΩ: the statement is vacuous.
Human review
  • Endorsed by Shuze Chen · Oct 21, 2026

    Confirmed by the moderator at approval.

  • Endorsed by mikedeng1 · Oct 21, 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