Lemma 2, p. 12 — a circumcenter always exists, and it is unique when C ∖ {0} ≠ ∅ and α(C) < π/2
ProvedCtxInvOpt.Greedy.lemma2_circumcenter_exists_uniquecircumcenterinverse-optimizationp2o-batch-pfp2ap2o-gran-per-chapterp2o-plan-paperp2o-v1uncertainty-angle
Let and let be nonempty. Then there exists a circumcenter , i.e. a unit vector attaining the infimum in
Furthermore, if is nonempty and , then the circumcenter is unique.
Existence makes the circumcenter policy well defined on every nonempty knowledge set; uniqueness is what pins down the greedy policy's proxy cost in the lower-bound construction of Theorem 2.
Formalization Note The angle uses the paper's convention . The standing assumption of §2 is a hypothesis. Uniqueness is stated as: any two circumcenters are equal.
Preamble
import Mathlib import Definitions.Def_CtxInvOpt_Offline_Geometry
Formal statement
namespace CtxInvOpt.Greedy
/-- **Lemma 2 (Existence and uniqueness of circumcenter), arXiv:2106.14015v3, p. 12.** For every
nonempty `C ⊆ ℝ^d` there is a circumcenter `ĉ(C) ∈ S^d` attaining the infimum in (9); if moreover
`C \ {0}` is nonempty and `α(C) < π/2`, the circumcenter is unique. -/
theorem lemma2_circumcenter_exists_unique {d : ℕ} (hd : 2 ≤ d)
(C : Set (EuclideanSpace ℝ (Fin d))) (hC : C.Nonempty) :
(∃ chat, CtxInvOpt.Offline.IsCircumcenter C chat) ∧
((C \ {0}).Nonempty → CtxInvOpt.Offline.uncAngle C < Real.pi / 2 →
∀ c₁ c₂, CtxInvOpt.Offline.IsCircumcenter C c₁ → CtxInvOpt.Offline.IsCircumcenter C c₂ → c₁ = c₂) := by sorry
end CtxInvOpt.Greedy
Source
Besbes, Fonseca, Lobel, Contextual Inverse Optimization: Offline and Online Learning, arXiv:2106.14015v3, p. 12, Lemma 2
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.