The Query Complexity of Envy-Free Division: Cake, Chores, and Mixed Resources
The paper studies how many Robertson–Webb Eval/Cut queries are needed to compute exactly envy-free allocations of cake, of chores, and of mixed resources with signed valuations. It asks how far existing bounds can be lowered, what randomization achieves for chores, and whether bounded-query guarantees extend to signed valuations.
For chores, a deterministic protocol uses deferred compensation. Packages of partial allocations are combined through Offset SubCore calls, and a capacity-based accounting argument bounds package weights. A Las Vegas protocol handles both cake and chores, and the lower bounds use measurable randomized-protocol arguments. The four main results are formalized in Lean 4.35.0 with Mathlib and a fixed-point library.
The deterministic chore protocol uses 2^{n log2 n + O(n)} queries. The Las Vegas protocol uses O(n^12 log(2n)) expected queries and O(n^11 log(2n)) expected cuts, improving on the prior Õ(n^30) for cake. Any exact protocol with fixed positive success probability needs Ω(n^2) expected queries, and for mixed resources no uniform finite expected-query bound exists.
| Setting | Prior benchmark | This paper |
|---|---|---|
| Deterministic exact chores | Bounded function of n | Q,C ≤ 2^{n log2 n + O(n)} |
| Randomized exact cake | E[Q] = Õ(n^30) | E[Q] = O(n^12 log(2n)), E[C] = O(n^11 log(2n)) |
| Randomized exact chores | Bounded expected queries | Same polynomial bounds as cake |
| Exact lower bounds | Ω(n^2) deterministic cake queries | E[Q] ≥ max{n(np−1)/2, n^2 p/4 − 1} |
