EF ⇒ PROP (Additive Valuations, Complete Allocations)
Theorem. Let $w : N \to G \to \mathbb{R}$ be an additive valuation
([[social_choice.fair_division.indivisible.additive_valuation]]),
allGoods : Finset G the outer good set, and $A$ a complete allocation
of allGoods ([[social_choice.fair_division.indivisible.allocation]]).
If $A$ is envy-free under $w.\mathrm{toValuation}$
([[social_choice.fair_division.indivisible.envy_free]]), then $A$ is
proportional with population size $|N|$
([[social_choice.fair_division.indivisible.proportional]]).
In Lean: SocialChoice.FairDivision.Indivisible.IsEnvyFree.isProportional_additive.
Proof
For each agent $k$, partition $\mathrm{allGoods}$ via $A$ and use additivity: $$ \begin{aligned} v_k(\mathrm{allGoods}) &= \sum_{j} v_k(A(j)) && \text{additivity over the disjoint cover} \\ &\le \sum_{j} v_k(A(k)) && \text{envy-freeness: $v_k(A(j)) \le v_k(A(k))$} \\ &= |N| \cdot v_k(A(k)) && \text{constant sum.} \end{aligned} $$
The first equality combines IsAllocation.complete (the cover) with
AdditiveValuation.toValuation_union extended to a biUnion over the
disjoint family.
Caveats
-
Additivity is essential. For non-additive valuations EF does not imply PROP — additivity is what converts "weakly less than $A(k)$" (a per-bundle inequality) into "total value bounded by $n \cdot v_k(A(k))$" (a sum bound).
-
EF1 does not imply PROP. The counterexample (2 agents, 3 equal-value goods, give 1 to agent A and 2 to agent B; A is EF1 but $v_A < |\mathrm{allGoods}|/2 \cdot v_A(A(A))$) is real even with additive valuations.
References
- [AGT Chapter 11] Nisan, Roughgarden, Tardos, and Vazirani, Algorithmic Game Theory. EF-PROP for additive valuations.