Lemma 4.3 — fixed pairs and rounded tails
OpenSublinearChoresMMS.Disposal.Supplement.lemma_4_3choresfair-divisionmaximin-sharep2o-batch-stoc27sp2o-gran-per-chapterp2o-plan-paperp2o-v1
In the normalized Section 4 setting with and , the actual Algorithm 1 law can be rounded to one finite law of integral complete allocations while keeping each agent's assigned opposite pair fixed. The law is ex-ante proportional:
Every positive-weight outcome simultaneously satisfies MMS1 for every agent and, for each agent ,
The joint law retains the Algorithm 1 support index, its exact marginal, and every conditional tail-slot marginal. Tail matchings use positive-mass edges, fill full tail slots, and leave fixed bundles intact. The removable chore must be in the tail. This preserves the precise disjunction needed by Theorem 4.4.
Differences from the source:
- encoding: the constructed opposite pairs and a possibly empty tail in the normalized IDO regime → oppositePairs and oppositeTail with 2n≤m and the published Normalized predicate; reason: zero-based pairing j with 2n-1-j is the same construction; the normalization, positivity and strict order are internal section assumptions
- encoding: Algorithm 1 with an arbitrary maximizing-agent choice, cheapest-bundle choices and two uniform random choices → the explicit recursive weighted list algorithm1 with index tie breaking, plus concluded probability normalization and completeness; reason: every chosen action obeys the algorithm, the full-completion permutations enumerate the residual bijections exactly once, and no cost bound is an input certificate
- encoding: round the actual Algorithm 1 fractional lottery while retaining assigned fixed bundles → a joint law of the Algorithm 1 support index and a tail-slot assignment, preserving the index and conditional slot marginals; reason: the additional witness records the stated rounding construction using the general Section 3.3 argument; it does not assume the frozen specialized Lemma 3.3 applies to new data
- encoding: finite agents, chores, complete partitions and finite lotteries → zero-based Fin labels, owner maps and published finite real-weight laws with positive-weight support; reason: bijections cover exactly the finite source instances, and repeated lottery atoms can be merged
- encoding: cost minimax share over all positive-part-count labelled partitions → the published costShare, the negative of the goods maximin functional on negative additive costs; reason: finite sign duality gives the same attained minimax without imposing any goods nonnegativity predicate
Preamble
import Mathlib import Definitions.Def_SublinearChoresMMS_Disposal_Supplement_TailRounding noncomputable section
Formal statement
namespace SublinearChoresMMS.Disposal.Supplement
attribute [local instance] Classical.propDecidable
/-- Lemma 4.3, printed 16 / PDF 17. The joint law keeps the actual Algorithm 1
outcome and its rounded tail matching. Both fairness clauses and the exact
fixed-pair-or-tail-removal conclusion refer to this same law. -/
theorem lemma_4_3 (n m : ℕ) (hn : 0 < n) (hm : 2 * n ≤ m)
(c : Cost n m) (hc : Normalized c) :
∃ F : FiniteLaw (Completion n),
F.toLaw = algorithm1 c (oppositePairs n m) (oppositeTail n m) ∧
∃ P : FiniteLaw (Fin F.size × SlotAssignment n m),
(∀ b, P.mean (fun o => if o.1 = b then 1 else 0) = F.weight b) ∧
(∀ b e, e ∈ oppositeTail n m → ∀ s : Slots n m,
P.mean (fun o => if o.1 = b ∧ o.2 e = s then 1 else 0) =
F.weight b * tailSlot (oppositeTail n m) (F.atom b) s e) ∧
(∀ i, P.mean (fun o => cost c i (bundle (fun e => (o.2 e).1) i)) ≤ propShare c i) ∧
∀ a, 0 < P.weight a →
RoundsTail (oppositePairs n m) (oppositeTail n m)
(F.atom (P.atom a).1) (P.atom a).2 ∧
∀ i,
MMS1 c i (bundle (fun e => ((P.atom a).2 e).1) i) ∧
(bundle (fun e => ((P.atom a).2 e).1) i =
assignedBundle (oppositePairs n m) (F.atom (P.atom a).1) i ∨
∃ e ∈ bundle (fun e => ((P.atom a).2 e).1) i,
e ∈ oppositeTail n m ∧
cost c i ((bundle (fun e => ((P.atom a).2 e).1) i).erase e) ≤ propShare c i) := by sorry
end SublinearChoresMMS.Disposal.Supplement
Source
Zehan Lin, Xiaowei Wu, Shengwei Zhou, STOC 2027 submission #845 (October 2026), Lemma 4.3, printed p. 16 / PDF p. 17. https://github.com/sigact-stoc/stoc2027/releases/download/papers-1/845.pdf