Title: Envy-Free Cake Cutting with Graph Constraints

Abstract: We study the classic problem of fairly dividing a heterogeneous and divisible resource -- represented by a cake, $[0,1]$ -- among $n$ agents. This work considers an interesting variant of the problem where agents are embedded on a graph. The graphical constraint entails that each agent evaluates her allocated share only against her neighbor's share. Given a graph, the goal is to efficiently find a locally envy-free allocation where every agent values her share to be at least as much as any of her neighbor's share.
The best known algorithm (by Aziz and Mackenzie) for finding envy-free cake divisions has a hyper-exponential query complexity. One of the key technical contributions of this work is to identify a non-trivial graph structure -- tree graphs with depth at-most two (Depth2Tree) -- on $n$ agents that admits a query efficient cake-cutting protocol (under the Robertson-Webb query model). In particular, we develop a discrete protocol that finds a locally envy-free allocation among $n$ agents on depth-two trees with at-most $O(n^3 \log(n))$ cuts on the cake. For the special case of Depth2Tree where every non-root agent is connected to at-most two agents (2-Star), we show that $O(n^2)$ queries suffice. We complement our algorithmic results with establishing a lower bound of $\Omega(n^2)$ (evaluation) queries for finding a locally envy-free allocation among $n$ agents on a 1-Star graph (under the assumption that the root agent partitions the cake into $n$ connected pieces).