AFTD Auto-Formalizing Theoretical Domains

pcp_half_two_best_response

lemmaverifiedformalization

For alpha = 1/2 and beta = 2 with positive costs, the closed form of Lemma 3.1, a_j = v_j^4 / (c_j^2 sum_l v_l^4/c_l), solves the researcher's Program (1) (Cauchy-Schwarz / AM-GM).

Statement

open Finset in
lemma pcp_half_two_best_response {k : ℕ} (c v : Fin k → ℝ) (hc : ∀ j, 0 < c j)
    (hD : 0 < ∑ j, v j ^ 4 / c j) :
    pcp_is_best_response (1 / 2) 2 c v (fun j => v j ^ 4 / (c j ^ 2 * ∑ l, v l ^ 4 / c l))
Formalizes
The Publication Choice Problem, AAAI 2026 (arXiv:2511.13678v2), Lemma 3.1 (closed-form best response to Program (1)); special case alpha = 1/2, beta = 2
Verified
06 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 1
  • pcp_is_best_responseProgram (1) of the Publication Choice Problem: a publication vector a is a best response to venue impacts v and costs c if it is…
Used by 2
  • pcp_two_venue_equilibrium_of_fixed_pointTwo venues, alpha = 1/2, beta = 2, any number of types: if the impacts u induced by best responses to the normalised impacts (1, y)…
  • pcx_equilibrium_of_rootEvery positive root x of pcx_poly gives a pure-strategy equilibrium of the binary instance with impacts pcx_impact x, in which every type…

Lean source view module on GitHub

open Finset in/-- For `α = 1/2`, `β = 2` and positive costs, the closed form of Lemma 3.1 of arXiv:2511.13678, `a j = v j ^ 4 / (c j ^ 2 * D)` with `D = ∑ j, v j ^ 4 / c j > 0`, is a best response. -/lemma pcp_half_two_best_response {k : ℕ} (c v : Fin k → ℝ) (hc : ∀ j, 0 < c j)    (hD : 0 < ∑ j, v j ^ 4 / c j) :    pcp_is_best_response (1 / 2) 2 c v (fun j => v j ^ 4 / (c j ^ 2 * ∑ l, v l ^ 4 / c l)) := by  set D := ∑ l, v l ^ 4 / c l with hDdef  set s := Real.sqrt D with hs  have hs0 : 0 < s := Real.sqrt_pos.mpr hD  have hss : s ^ 2 = D := Real.sq_sqrt hD.le  have hrp : ∀ x : ℝ, 0 ≤ x → x ^ (1 / 2 : ℝ) = Real.sqrt x := fun x _ => (Real.sqrt_eq_rpow x).symm  have hv2 : ∀ j, v j ^ (2 : ℝ) = v j ^ 2 := fun j => Real.rpow_two (v j)  have ha_nn : ∀ j, 0 ≤ v j ^ 4 / (c j ^ 2 * D) := fun j => by    have := hc j; positivity  have hsqrt_a : ∀ j, Real.sqrt (v j ^ 4 / (c j ^ 2 * D)) = v j ^ 2 / (c j * s) := fun j => by    have hcj := hc j    have : v j ^ 4 / (c j ^ 2 * D) = (v j ^ 2 / (c j * s)) ^ 2 := by      rw [div_pow, mul_pow, hss]; ring    rw [this, Real.sqrt_sq (by positivity)]  have hval : ∑ j, (v j ^ 4 / (c j ^ 2 * D)) ^ (1 / 2 : ℝ) * v j ^ (2 : ℝ) = s := by    have : ∀ j, (v j ^ 4 / (c j ^ 2 * D)) ^ (1 / 2 : ℝ) * v j ^ (2 : ℝ) = (v j ^ 4 / c j) / s := by      intro j      have hcj := hc j      rw [hrp _ (ha_nn j), hsqrt_a, hv2]      field_simp    rw [Finset.sum_congr rfl (fun j _ => this j), ← Finset.sum_div, ← hDdef, ← hss]    field_simp  refine ⟨ha_nn, ?_, ?_⟩  · have : ∀ j, v j ^ 4 / (c j ^ 2 * D) * c j = (v j ^ 4 / c j) / D := by      intro j      have hcj := hc j      field_simp    rw [Finset.sum_congr rfl (fun j _ => this j), ← Finset.sum_div, ← hDdef, div_self hD.ne']  · intro a' ha' hbud    rw [hval]    have hterm : ∀ j, a' j ^ (1 / 2 : ℝ) * v j ^ (2 : ℝ) ≤        s * (a' j * c j) / 2 + (v j ^ 4 / c j) / (2 * s) := by      intro j      have hcj := hc j      rw [hrp _ (ha' j), hv2]      set r := Real.sqrt (a' j) with hr      have hr2 : r ^ 2 = a' j := Real.sq_sqrt (ha' j)      rw [← hr2]      have key : s * (r ^ 2 * c j) / 2 + v j ^ 4 / c j / (2 * s) - r * v j ^ 2 =          (s * c j * r - v j ^ 2) ^ 2 / (2 * s * c j) := by        field_simp; ring      have : 0 ≤ (s * c j * r - v j ^ 2) ^ 2 / (2 * s * c j) := by positivity      linarith    calc ∑ j, a' j ^ (1 / 2 : ℝ) * v j ^ (2 : ℝ)        ≤ ∑ j, (s * (a' j * c j) / 2 + (v j ^ 4 / c j) / (2 * s)) := Finset.sum_le_sum fun j _ => hterm j      _ = s * (∑ j, a' j * c j) / 2 + D / (2 * s) := by          rw [Finset.sum_add_distrib, ← Finset.sum_div, ← Finset.sum_div, ← Finset.mul_sum, hDdef]      _ ≤ s * 1 / 2 + D / (2 * s) := by gcongr      _ = s := by rw [← hss]; field_simp; ring
See it in the graphDiscuss this resultChallenge it