AFTD Auto-Formalizing Theoretical Domains

spi_two_point_core

lemmaverified

Real-arithmetic core of Proposition 3.1 for two-point rewards: accepted masses with posterior mean at least T have total at most spi_two_point_max_prob, and attaining it pins the accepted value to spi_two_point_max_value.

Statement

lemma spi_two_point_core (h l q T a b : ℝ) (hlh : l ≤ h) (hq0 : 0 ≤ q) (hq1 : q ≤ 1)
    (ha0 : 0 ≤ a) (ha1 : a ≤ q) (hb0 : 0 ≤ b) (hb1 : b ≤ 1 - q) (hc : T * (a + b) ≤ a * h + b * l) :
    a + b ≤ spi_two_point_max_prob h l q T ∧
      (spi_two_point_max_prob h l q T ≤ a + b → a * h + b * l = spi_two_point_max_value h l q T)
Source
step towards strategic_prophet_no_half_robust_static_threshold (no static threshold is 1/2-robust, open in Intrinsic Robustness of Prophet Inequality to Strategic Reward Signaling, NeurIPS 2024 (arXiv:2409.18269), Remark 4.1 and Sec. 6)
Verified
06 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 2
  • spi_two_point_max_probProposition 3.1 for a two-point reward (h with probability q, else l): the largest acceptance probability achievable under threshold T (1…
  • spi_two_point_max_valueThe searcher's expected reward from a two-point player using an optimal scheme (E[X] if T <= E[X], T times the pooling probability if E[X]…
Used by 1
  • spi_two_point_boundFor a two-point reward, every scheme's acceptance probability is at most spi_two_point_max_prob, and if it attains it the accepted value…

Lean source view module on GitHub

/-- Real-arithmetic core of Prop. 3.1 for a two-point reward: accepted masses `a ≤ q` (on `h`) and `b ≤ 1 - q` (on `l`) with posterior mean at least `T` have total at most `spi_two_point_max_prob`, and attaining it pins the accepted value to `spi_two_point_max_value`. -/lemma spi_two_point_core (h l q T a b : ℝ) (hlh : l ≤ h) (hq0 : 0 ≤ q) (hq1 : q ≤ 1)    (ha0 : 0 ≤ a) (ha1 : a ≤ q) (hb0 : 0 ≤ b) (hb1 : b ≤ 1 - q) (hc : T * (a + b) ≤ a * h + b * l) :    a + b ≤ spi_two_point_max_prob h l q T ∧      (spi_two_point_max_prob h l q T ≤ a + b → a * h + b * l = spi_two_point_max_value h l q T) := by  unfold spi_two_point_max_prob spi_two_point_max_value  split_ifs with h1 h2  · refine ⟨by linarith, fun hP => ?_⟩    have : a = q := by linarith    have : b = 1 - q := by linarith    subst_vars; ring  · have hTl : l < T := by nlinarith    have hTl' : 0 < T - l := by linarith    have key : b * (T - l) ≤ a * (h - T) := by nlinarith    refine ⟨?_, fun hP => ?_⟩    · rw [le_div_iff₀ hTl']; nlinarith    · rw [div_le_iff₀ hTl'] at hP      have : a * (h - T) ≤ b * (T - l) := by nlinarith      have hab : a + b = q * (h - l) / (T - l) := by        rw [eq_div_iff hTl'.ne']; nlinarith      rw [← hab]; nlinarith  · have hab : (a + b) * (T - h) ≤ 0 := by nlinarith    have : a + b ≤ 0 := by      by_contra hneg; push Not at hneg      have := mul_pos hneg (show 0 < T - h by linarith); linarith    refine ⟨this, fun _ => ?_⟩    have : a = 0 := by linarith    have : b = 0 := by linarith    subst_vars; ring
See it in the graphDiscuss this resultChallenge it