Proof of Theorem 2, p. 7627 — for f = xEy, g = wEz with the slope in the window, the unique optimal mixture is α* = 1/(1 + (u(x)−u(y))/(u(z)−u(w)))
ProvedSurelyBetter.CLB.theorem2_interior_optimumLet be a finite set of states and let be a convex set containing two distinct points; stands for , the range of a nonconstant affine utility on simple lotteries. Acts are represented by their utility profiles (see the definitions item for why this is exact). Let be a c-linearly biseparable preference with capacity : is normalized, monotone and continuous on , and for every event the functional , , represents on all binary acts. Fix an event and write for its complement.
Suppose . Let (utilities of prizes) with , , , , and let be the bet paying on and off , the bet paying on and off . Suppose the slope of the segment between the two profiles lies strictly inside the window given by the two linear pieces:
Then the set of optimal mixtures — all such that is optimal in — is the single point
the interior weight that equalizes utility across and .
Formalization Note Acts are utility profiles S → ℝ with values in U; the hypotheses Convex ℝ U and ∃ a ∈ U, ∃ b ∈ U, a ≠ b encode that for a nonconstant affine , and IsCLinBisep U V ρ bundles the standing MBA restriction of §4 with the representation (7) as an order equivalence over binary acts of all events. The slope window is stated cross-multiplied, and , because is allowed; then the left fraction is and the second inequality reduces to . The paper's existence remark (such exist) is not part of this statement.
import Mathlib import Definitions.Def_Supermodularity_Lattices_InducedSetOrder import Definitions.Def_SurelyBetter_CLB_Model
namespace SurelyBetter.CLB
theorem theorem2_interior_optimum {S : Type*} [Fintype S] (U : Set ℝ)
(hU : Convex ℝ U) (hU2 : ∃ a ∈ U, ∃ b ∈ U, a ≠ b) (V : (S → ℝ) → ℝ) (ρ : Set S → ℝ)
(hV : IsCLinBisep U V ρ) (E : Set S) (hρ : ρ E + ρ Eᶜ < 1)
(x y w z : ℝ) (hx : x ∈ U) (hy : y ∈ U) (hw : w ∈ U) (hz : z ∈ U)
(hxy : y < x) (hzw : w < z) (hxw : w < x) (hzy : y < z)
(hslope_lo : ρ E * (x - w) < (1 - ρ E) * (z - y))
(hslope_hi : ρ Eᶜ * (z - y) < (1 - ρ Eᶜ) * (x - w)) :
optMix V (binProf E x y) (binProf E w z) = {1 / (1 + (x - y) / (z - w))} := by sorry
end SurelyBetter.CLB
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.