Owner choices encode all large modeled balls within the capacity bound
ProvedStableMSR.Recovery.Supplement.BinaryRecovery.Regional.bounded_owner_ball_encodingalgorithmsclusteringdynamic-programming
Choose an input center and farthest point for each modeled large block and place the code with an owner within 2R of its center. All blocks assigned to one owner are contained in its 3R-ball, hence the per-owner tuple length is at most h. The selected positive balls recover exactly the large subfamily.
Preamble
import Definitions.Def_StableMSR_Recovery_Supplement_PriorityRegions import Definitions.Def_StableMSR_Recovery_Supplement_RegionalModel open StableMSR.Recovery StableMSR.Recovery.Supplement StableMSR.Recovery.Supplement.BinaryRecovery open StableMSR.Recovery.Supplement.BinaryRecovery.Regional
Formal statement
theorem StableMSR.Recovery.Supplement.BinaryRecovery.Regional.bounded_owner_ball_encoding (γ : ℚ) (I : RationalInput) (hv : I.Valid) (R : ℚ) (w : Block) (rp : 0<R)
(h : ℕ) (P : Partition I.n) (md : LocalModel I γ (R:ℝ) (rawBlock I.n w) h P) :
let z : RegionInput := (fieldsOf I,R,w)
∃ L⊆P, ∃ s : Plan, AllowedPlan (netKeys z) (tuples (ballCodes z.1) h) s ∧
(∀ e∈ballCodes z.1, let B := (Finset.univ : Cluster I.n).filter (fun x => strikes γ z s (true,e) x.val=true)
B=∅ ∨ B∈L) ∧
(∀ C∈L, ∃e∈ballCodes z.1, (Finset.univ : Cluster I.n).filter (fun x => strikes γ z s (true,e) x.val=true)=C) ∧
∀ C∈P, C∉L → C∈components I.dist (2*(((γ:ℝ)-1)*R/4)) Finset.univ ∧ radius I.dist C<((γ:ℝ)-1)*R/4 := by
sorry
Source
S. Ravichandran and L. Reyzin, Exact Sum-of-Radii Clustering under Any Constant Stability, Proposition 4.2 and Appendices A.4 and A.6, https://arxiv.org/abs/2611.03547. The priority-ball construction is an alternative implementation of the regional optimization.