Theorem 3, p. 18 — under Assumption 1, and weights ( solving (23)) give the optimal central mechanism
ProvedPrivAcq.Central.theorem_3Consider the central-privacy data acquisition model under the standing assumptions, and suppose Assumption 1 holds: every virtual cost is increasing. For every type profile let be an optimal solution of
chosen measurably in . Let the mechanism use privacy losses , the estimator weights and Laplace parameter , and the payments (19) with . Then
- is an admissible mechanism of class (22), incentive compatible (17) and individually rational (18); and
- its expected platform cost is at most that of every admissible mechanism of class (22) that is incentive compatible and individually rational.
So the optimal privacy loss level is , and user 's data receives weight in the platform's estimator.
Formalization Note The theorem is proved for every measurable selection of minimizers (the minimizer need not be unique); it does not claim that every optimal mechanism coincides with . The feasible set of (23) adds (at the objective is undefined). Model conventions (compact type support with , no differentiability or sign constraint on payments, strictly increasing ) are listed in PrivAcq.Central.Model.
import Mathlib import Definitions.Def_PrivAcq_Central_Model
namespace PrivAcq.Central
open MeasureTheory
theorem theorem_3 {n : ℕ} (S : Setting n) (hS : S.Standing) (hA1 : S.Assumption1)
(ystar : (Fin n → ℝ) → Fin n → ℝ) (hmeas : Measurable ystar)
(hopt : ∀ c ∈ S.types, IsOpt23 S c (ystar c)) :
Admissible S (optMech S ystar) ∧ IC S (optMech S ystar) ∧ IR S (optMech S ystar) ∧
∀ M : Mech n, Admissible S M → IC S M → IR S M →
objective S (optMech S ystar) ≤ objective S M := by sorry
end PrivAcq.Central
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.