AFTD Auto-Formalizing Theoretical Domains

plackett_luce_surprisal_variance_profile_const_one

theoremverified

For the constant weight profile w(x) = 1 on [0,1], the asymptotic surprisal variance plackett_luce_surprisal_variance_profile equals 0.

Statement

theorem plackett_luce_surprisal_variance_profile_const_one :
    plackett_luce_surprisal_variance_profile (fun _ => 1) = 0
Source
arXiv:2009.09172 (Preference Elicitation as Average-Case Sorting, AAAI 2021)
Verified
15 Nov 2026
Axioms
Classical.choiceQuot.soundpropext
Built on 3
Used by 1

Read back from the Lean

The Plackett-Luce surprisal variance profile of the constant function $x \mapsto 1$ is equal to $0$.

Written by a model that saw only the Lean, never the English above. If the two disagree, that is worth a challenge.

The proof, explained

The proof unfolds the definition of the asymptotic surprisal variance profile for the constant weight profile w(x) = 1 on [0, 1]. Using the previously established formula for the Laplace profile of the constant-one weight, the random variable inside the variance simplifies to log(exp(-t)) + t = -t + t = 0. Since the variance of the constant zero function is 0 under any measure, the integrand vanishes identically, and the integral of zero over [0, 1] is 0.

  1. The surprisal variance profile reduces to the integral over [0, 1] of the variance of the random variable t ↦ -t + t with respect to the exponential measure on ℝ.

    Unfolding the definition of the surprisal variance profile for the constant weight profile w(x) = 1, the Laplace profile simplifies to exp(-t) by `plackett_luce_laplace_profile_const_one`. Taking the logarithm gives log(exp(-t)) = -t by `Real.log_exp`, and the linear term w(x) * t = 1 * t simplifies to t by `one_mul`, leaving the variance of t ↦ -t + t under expMeasure(1).

    lines 4–5

    uses plackett_luce_laplace_profile_const_one, Real.log_exp, one_mul

      unfold plackett_luce_surprisal_variance_profile
      simp only [plackett_luce_laplace_profile_const_one, Real.log_exp, one_mul]
  2. The random variable t ↦ -t + t is equal to the constant zero function.

    For every real number t, -t + t = 0 by `neg_add_cancel`. By function extensionality, the function t ↦ -t + t is identically zero, and rewriting with this identity replaces it by the zero function in the integrand.

    lines 6–8

    uses neg_add_cancel

      have h : (fun t : ℝ => -t + t) = (fun _ => (0:ℝ)) := by
        funext t; exact neg_add_cancel t
      rw [h]
  3. The variance of the zero function is zero, so the integrand is identically 0 on [0, 1].

    After adjusting the zero function to match the type signature of `ProbabilityTheory.variance_zero`, applying this lemma shows that the variance of the zero function under the exponential measure is 0.

    lines 9–11

    uses ProbabilityTheory.variance_zero

      change (∫ (x : ℝ) in Set.Icc 0 1,
        ProbabilityTheory.variance (0 : ℝ → ℝ) (ProbabilityTheory.expMeasure (1:ℝ))) = 0
      rw [ProbabilityTheory.variance_zero]
  4. The integral of 0 over [0, 1] is 0, concluding the proof. automation

    The integral of the constant zero function over the interval [0, 1] evaluates to 0.

    line 12

      simp
plackett_luce_laplace_profile_const_one
For the constant weight profile w = 1, the Laplace profile at t is exp(-t).
Real.log_exp
The natural logarithm of exp(x) is x for any real number x.
one_mul
Multiplying an element by 1 on the left leaves it unchanged (1 * a = a).
neg_add_cancel
In any additive group, adding the negative of an element to itself yields zero (-a + a = 0).
ProbabilityTheory.variance_zero
The variance of the constant zero random variable under any measure is 0.

Written by a model (gemini-3.8-flash) from the Lean proof and the goal states Lean reported at each step. Lean verified the proof; this explanation is a reading aid and is not itself machine-checked.

Lean source view module on GitHub

/-- The asymptotic surprisal variance for the constant weight profile w = 1 is 0. -/theorem plackett_luce_surprisal_variance_profile_const_one :    plackett_luce_surprisal_variance_profile (fun _ => 1) = 0 := by  unfold plackett_luce_surprisal_variance_profile  simp only [plackett_luce_laplace_profile_const_one, Real.log_exp, one_mul]  have h : (fun t : ℝ => -t + t) = (fun _ => (0:ℝ)) := by    funext t; exact neg_add_cancel t  rw [h]  change (∫ (x : ℝ) in Set.Icc 0 1,    ProbabilityTheory.variance (0 : ℝ → ℝ) (ProbabilityTheory.expMeasure (1:ℝ))) = 0  rw [ProbabilityTheory.variance_zero]  simp
See it in the graphDiscuss this resultChallenge it