AFTD Auto-Formalizing Theoretical Domains

IsRiskNeutral.isAffine

theoremverifiedformalizationimported Lean

Risk neutrality implies affine utility. [MSZ 2.27, hard direction] Requires |I| ≥ 2 so that non-trivial distributions exist.

Statement

set_option autoImplicit true in
set_option relaxedAutoImplicit true in
variable {𝕜 : Type*} [Field 𝕜] [LinearOrder 𝕜] [IsStrictOrderedRing 𝕜] in
set_option linter.unusedSectionVars false in
theorem IsRiskNeutral.isAffine {I : Type*} [Fintype I] [Nontrivial I] {u : 𝕜 → 𝕜}
    (h : IsRiskNeutral (I := I) u) : IsAffineUtility u
Formalizes
EconCSLib, `IsRiskNeutral.isAffine`
Lean proof
By xbei (from the file's git history), from github.com/gametheoryinlean/EconCSLib at 1a88f809b538 (Copyright (c) 2026 EconCSLib contributors. All rights reserved, Apache-2.0). Compiled here against leanprover/lean4:v4.33.1, Mathlib v4.33.1; written for leanprover/lean4:v4.30.0.
Verified
08 Oct 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 15
  • LotteryA lottery over outcomes O: a probability distribution. Just stdSimplex 𝕜 O with a game-theoretic name.
  • TotalPreorderA total preorder: a preorder where ≤ is total (complete). This is weaker than LinearOrder — it does NOT require antisymmetry or…
  • wsumWeighted sum of f with weights from a simplex element x. Thin abbrev over Mathlib's finite dot product ⬝ᵥ: definitionally equal,…
  • Lottery.expectedValue_mixLinearity of expected value under convex combination. E[mix α L₁ L₂] = α · E[L₁] + (1-α) · E[L₂] This is the key property…
  • stdSimplex.purePoint-mass simplex element at i₀.
  • IsRiskNeutralRisk neutrality for lotteries over a finite index set I: u(∑ pᵢ·xᵢ) = ∑ pᵢ·u(xᵢ). Equivalent to u being affine. [MSZ 2.27]
  • Lottery.expectedValue_pureExpected value of a pure lottery equals the outcome value.
  • wsum_pure_applyWeighted sum at a point mass evaluates the chosen coordinate.
  • Lottery.pureA pure (degenerate) lottery: outcome o₀ with probability 1. Alias of stdSimplex.pure.
  • LinearOrder.toTotalPreorderEvery LinearOrder is a TotalPreorder.
  • stdSimplex.pure_applystdSimplex.pure_apply
  • IsAffineUtilityA utility function is affine (linear + constant): u(x) = a·x + b. An agent with an affine utility function is risk neutral. [MSZ 2.24]
  • Lottery.mixConvex combination of two lotteries: the compound lottery [α(L₁), (1-α)(L₂)] after simplification. [MSZ Axiom 2.16] Alias of…
  • stdSimplex.mix_applystdSimplex.mix_apply
  • Lottery.expectedValueExpected value of f under lottery L: E_L[f] = ∑ o, L(o) · f(o). Alias of wsum from Math.Simplex.

Lean source view module on GitHub

set_option autoImplicit true inset_option relaxedAutoImplicit true invariable {𝕜 : Type*} [Field 𝕜] [LinearOrder 𝕜] [IsStrictOrderedRing 𝕜] inset_option linter.unusedSectionVars false in/-- Risk neutrality implies affine utility. [MSZ 2.27, hard direction] Requires `|I| ≥ 2` so that non-trivial distributions exist. -/theorem IsRiskNeutral.isAffine {I : Type*} [Fintype I] [Nontrivial I] {u : 𝕜 → 𝕜}    (h : IsRiskNeutral (I := I) u) : IsAffineUtility u := by  classical  obtain ⟨i₀, i₁, hne⟩ := exists_pair_ne I  -- Step 1: derive convex combination property from risk neutrality  -- u(t·a + (1-t)·b) = t·u(a) + (1-t)·u(b) for t ∈ [0,1]  have conv : ∀ (t : 𝕜) (_ : 0 ≤ t) (_ : t ≤ 1) (a b : 𝕜),      u (t * a + (1 - t) * b) = t * u a + (1 - t) * u b := by    intro t ht₀ ht₁ a b    let p := Lottery.mix t ht₀ ht₁ (Lottery.pure (𝕜 := 𝕜) i₀) (Lottery.pure i₁)    let f : I → 𝕜 := fun i => if i = i₀ then a else b    have hL : wsum p f = t * a + (1 - t) * b := by      change Lottery.expectedValue p f = _      rw [Lottery.expectedValue_mix]      simp [Lottery.expectedValue_pure, f, hne.symm]    have hR : wsum p (u ∘ f) = t * u a + (1 - t) * u b := by      change Lottery.expectedValue p (u ∘ f) = _      rw [Lottery.expectedValue_mix]      simp [Lottery.expectedValue_pure, Function.comp, f, hne.symm]    rw [← hL, h p f, hR]  -- Step 2: u(x) = (u 1 - u 0) · x + u 0 for all x  refine ⟨u 1 - u 0, u 0, fun x => ?_⟩  -- Case x ∈ [0,1]: u(x·1 + (1-x)·0) = x·u(1) + (1-x)·u(0)  -- Case x > 1: u((1/x)·x + (1-1/x)·0) = (1/x)·u(x) + (1-1/x)·u(0), solve for u(x)  -- Case x < 0: u(0) = (1/2)·u(x) + (1/2)·u(-x), with u(-x) known since -x > 0  suffices ∀ y : 𝕜, 0 ≤ y → u y = (u 1 - u 0) * y + u 0 by    by_cases hx : 0 ≤ x    · exact this x hx    · push_neg at hx      have hmx := this (-x) (le_of_lt (neg_pos.mpr hx))      have h_mid := conv (1 / 2) (by norm_num) (by norm_num) x (-x)      have : (1 : 𝕜) / 2 * x + (1 - 1 / 2) * -x = 0 := by ring      rw [this] at h_mid      linarith  intro y hy  rcases eq_or_lt_of_le hy with rfl | hy_pos  · simp  rcases le_or_gt y 1 with hy1 | hy1  · -- y ∈ (0, 1]    have := conv y hy hy1 1 0    simp at this; linarith  · -- y > 1    have h_inv := conv (1 / y) (le_of_lt (div_pos one_pos hy_pos))      ((div_le_one hy_pos).mpr (le_of_lt hy1)) y 0    simp at h_inv    have hyne : y ≠ 0 := ne_of_gt hy_pos    field_simp at h_inv ⊢; linarith
See it in the graphDiscuss this resultChallenge it