gpa_lhs_attain
lemmaverified
The bound 1 - 1/α + 1/k is attained on the simplex, by the uniform distribution on a maximum independent set I (1 ≤ |I| ≤ k).
Statement
open Finset in lemma gpa_lhs_attain {k : ℕ} (G : SimpleGraph (Fin k)) [DecidableRel G.Adj] (I : Finset (Fin k)) (hI : gpa_indep G I) (hne : 1 ≤ I.card) (hk : I.card ≤ k) : ∃ σ ∈ gpa_simplex k, gpa_uStar (gpa_adjMat G) σ - gpa_cost σ = 1 - 1 / (I.card : ℝ) + 1 / (k : ℝ)
- Source
- step towards gpa_eq54_false / gpa_eq54_fails_of_indep (Generalized Principal-Agency: Contracts, Information, Games and Beyond, WINE 2024 (arXiv:2209.01146v2), Sec. 5.3, Eq. (5.4))
- Verified
- 06 Oct 2026
- Axioms
Classical.choiceQuot.soundpropext- Built on 6
- gpa_adjMatThe 0/1 matrix e of arXiv:2209.01146, Section 5.3 (taken from the maximum-independent-set instances of the cited convex-maximisation…
- gpa_uStarThe principal's utility of arXiv:2209.01146, Section 5.3: u*(σ) = Σ_i max{σ_i - Σ_j e_{ij} σ_j, 0} for a 0/1 matrix e.
- gpa_costThe convex signal cost of arXiv:2209.01146, Section 5.3: h(σ) = max{‖σ‖_∞ - 1/k, 0} (the norm on Fin k → ℝ is the sup norm).
- gpa_simplexThe probability simplex Δ_k.
- gpa_indepAn independent set of the graph G.
- gpa_nbr_nonnegNeighbour sums are nonnegative for nonnegative vectors.
- Used by 1
- gpa_lhs_isGreatestExact value of the left side of Eq. (5.4) of arXiv:2209.01146 on the independent-set instances: max_{σ ∈ Δ_k} u*(σ) - h(σ) = 1 - 1/α(G) +…
Lean source view module on GitHub
Choose a tactic — a marked line — to see the goals Lean had before and after it, as in Lean’s infoview. Step with ← → or j k.
open Finset in/-- The bound `1 - 1/α + 1/k` is attained on the simplex, by the uniform distribution on a maximum independent set `I` (`1 ≤ |I| ≤ k`). -/lemma gpa_lhs_attain {k : ℕ} (G : SimpleGraph (Fin k)) [DecidableRel G.Adj] (I : Finset (Fin k)) (hI : gpa_indep G I) (hne : 1 ≤ I.card) (hk : I.card ≤ k) : ∃ σ ∈ gpa_simplex k, gpa_uStar (gpa_adjMat G) σ - gpa_cost σ = 1 - 1 / (I.card : ℝ) + 1 / (k : ℝ) := by classical set a : ℝ := (I.card : ℝ) with ha_def have haR : (1 : ℝ) ≤ a := by rw [ha_def]; exact_mod_cast hne have hapos : (0 : ℝ) < a := by linarith have hkR : a ≤ (k : ℝ) := by rw [ha_def]; exact_mod_cast hk let σ : Fin k → ℝ := fun i => if i ∈ I then 1 / a else 0 have hσnn : ∀ i, 0 ≤ σ i := fun i => by simp only [σ]; split_ifs · exact div_nonneg zero_le_one hapos.le · exact le_refl 0 have hσsum : ∑ i, σ i = 1 := by simp only [σ] rw [Finset.sum_ite_mem, Finset.univ_inter, Finset.sum_const, nsmul_eq_mul] rw [← ha_def]; field_simp refine ⟨σ, ⟨hσnn, hσsum⟩, ?_⟩ -- u*(σ) = 1 have hu : gpa_uStar (gpa_adjMat G) σ = 1 := by unfold gpa_uStar have hterm : ∀ i, max (σ i - ∑ j, gpa_adjMat G i j * σ j) 0 = σ i := by intro i by_cases hi : i ∈ I · have hz : ∑ j, gpa_adjMat G i j * σ j = 0 := by apply Finset.sum_eq_zero intro j _ by_cases hj : j ∈ I · have : ¬ G.Adj i j := hI i hi j hj simp [gpa_adjMat, this] · simp [σ, hj] rw [hz, sub_zero]; exact max_eq_left (hσnn i) · have hnn := gpa_nbr_nonneg G σ hσnn i have h0 : σ i = 0 := by simp [σ, hi] rw [h0] exact max_eq_right (by linarith) simp_rw [hterm] exact hσsum -- ‖σ‖ = 1/a have hnorm : ‖σ‖ = 1 / a := by apply le_antisymm · apply (pi_norm_le_iff_of_nonneg (div_nonneg zero_le_one hapos.le)).2 intro i rw [Real.norm_eq_abs, abs_of_nonneg (hσnn i)] simp only [σ]; split_ifs · exact le_refl _ · exact div_nonneg zero_le_one hapos.le · obtain ⟨i0, hi0⟩ : I.Nonempty := Finset.card_pos.1 (by omega) have := norm_le_pi_norm σ i0 rw [Real.norm_eq_abs, abs_of_nonneg (hσnn i0)] at this simpa [σ, hi0] using this have hcost : gpa_cost σ = 1 / a - 1 / (k : ℝ) := by unfold gpa_cost rw [hnorm] apply max_eq_left have hkpos : (0 : ℝ) < k := by linarith rw [sub_nonneg] exact one_div_le_one_div_of_le hapos hkR rw [hu, hcost] ring