Round-Robin Is EF1
Theorem. For any additive valuation ([[social_choice.fair_division.indivisible.additive_valuation]]) with nonnegative weights and any finite good set, the round-robin allocation ([[social_choice.fair_division.indivisible.round_robin_alloc]]) is envy-free up to one good ([[social_choice.fair_division.indivisible.ef1]]).
In Lean: SocialChoice.FairDivision.Indivisible.roundRobinAllocation_isEF1.
The bundled-rule form
SocialChoice.FairDivision.Indivisible.roundRobinRule_isEF1 packages
the same result against the bundled additive-instance API.
Proof
intuition
Fix two agents $i \ne j$ with $A(j) \ne \emptyset$. Two cases:
-
Agent $i$ moves before agent $j$ in the turn order. At every round where both agents pick, $i$ picks first and $i$'s best-remaining value at that round is at least $j$'s next pick (each agent picks their own favourite among the remaining items, and the set has only shrunk between $i$'s and $j$'s pick within the same round). So $v_i(A(i)) \ge v_i(A(j))$ pointwise, i.e. EF (a strictly stronger condition than EF1).
-
Agent $i$ moves after agent $j$. In the first round, $j$ picks before $i$ — the only round where $j$ has a strict positional advantage over $i$. After removing the first item $j$ takes (call it $g_1 = $
bestGood(j, allGoods)), agent $i$ has a first-pick situation for the remaining items, then both alternate fairly. So $v_i(A(j) \setminus \{g_1\}) \le v_i(A(i))$, giving the EF1 witness $g = g_1$.
Combining both cases gives EF1 for every $(i, j)$ pair.
The Lean proof is much more detailed — the recursive structure of
roundRobinAux forces an induction on the number of rounds, with the
case split appearing inside the inductive step. The file
RoundRobin.lean carries ~600 lines of partition/value/turn-order
bookkeeping closing all sorries.
Significance
This is the headline existence theorem for EF1: round-robin gives a deterministic polynomial-time construction of an EF1 allocation for any additive valuation. Combined with envy-cycle elimination ([[social_choice.fair_division.indivisible.envy_cycle_ef1]]) it establishes that EF1 is unconditionally achievable in the indivisible setting (in stark contrast to EF, which fails [[social_choice.fair_division.indivisible.ef_impossible_two_agents_one_good]]).
References
- [AGT Chapter 11] Nisan, Roughgarden, Tardos, and Vazirani, Algorithmic Game Theory. Round-robin EF1 correctness.
- Caragiannis, Kurokawa, Moulin, Procaccia, Shah, and Wang (2019). "The Unreasonable Fairness of Maximum Nash Welfare". EC.