persuasion_sender_utility
def
Sender utility S(q): expected number of coordinates recommended as 1.
Statement
open Finset in noncomputable def persuasion_sender_utility {ι : Type*} [Fintype ι] [DecidableEq ι] (q : (ι → Bool) → (ι → Bool) → ℝ) : ℝ
- Defined in
- Quantifying Theoretical AI Alignment Guarantees: Receiver-Utility Bounds in Bayesian Persuasion, arXiv:2606.22226 v1, Sec. 3 (bit-string model, direct obedient form, R_0, S*, R_max); formalized by the operator (Claude), 2026-10-03
- Verified
- 03 Oct 2026
- Axioms
- none listed
- Used by 8
- persuasion_uniform_two_states_ratio_leFor every finite type ι and every prior μ on ι → Bool that is uniform on 2 distinct states, the maximum receiver utility R_max(μ) is at…
- persuasion_max_receiver_utility_learXiv:2606.22226, Theorem 4.5 (ratio form). For every prior, R_max(μ) ≤ (3/2) R_0(μ).
- persuasion_pair_or_receiver_add_senderA pair-OR scheme never recommends 0 on a coordinate equal to 1, so the trade-off of Lemma 4.2 is an equality for it.
- persuasion_slice_scheme_sender_utilityThe slice scheme gives the sender C(2m+1,m) - C(2m-1,m): a coordinate is recommended 1 iff its m-subset meets the two paired states.
- persuasion_receiver_add_sender_leSender–receiver trade-off (Lemma 4.2). R(q) + S(q) ≤ n + ∑_i p_i(μ).
- is_sender_optimal_schemeq is obedient and maximises the sender's utility among obedient schemes.
- persuasion_slice_sender_leSender value of the slice prior. No obedient scheme gives the sender more than C(2m+1,m) - C(2m-1,m).
- persuasion_pair_or_sender_utilitySender utility of a pair-OR scheme.
Lean source view module on GitHub
open Finset in /-- Sender utility `S(q)`: expected number of coordinates recommended as `1`. -/ noncomputable def persuasion_sender_utility {ι : Type*} [Fintype ι] [DecidableEq ι] (q : (ι → Bool) → (ι → Bool) → ℝ) : ℝ := ∑ x, ∑ a, q x a * ((Finset.univ.filter (fun i => a i = true)).card : ℝ)