We study the problem of allocating a set $M$ of $m$ ${indivisible}$ items among $n$ agents in a fair manner. We consider two well-studied notions of fairness: envy-freeness (EF), and envy-freeness up to any good (EFX). While it is known that complete EF allocations do not always exist, it is not known if complete EFX allocations exist besides a few cases. In this work, we reformulate the problem to allow $M$ to be a multiset. Specifically, we introduce a parameter $t$ for the number of distinct ${types}$ of items, and study allocations of multisets that contain items of these $t$ types. We show the following: 1. For arbitrary $n$, $t$, a complete EF allocation exists when agents have distinct additive valuations, and there are ${enough}$ items of each type. 2. For arbitrary $n$, $m$, $t$, a complete EFX allocation exists when agents have additive valuations with identical ${preferences}$. 3. For arbitrary $n$, $m$, and $t\le2$, a complete EFX allocation exists when agents have additive valuations. For 2 and 3, our approach is constructive; we give a polynomial-time algorithm to find a complete EFX allocation.
Why multisets?
Agents often categorize a large number of items into a smaller number of types, like “genres” of books. With multisets, each agent rates each type (genre) of book with the same value, as opposed to separately valuing each book in the library. Given \(m\) items but only \(t\) types, each agent needs to provide only \(t\) numbers, instead of \(m\) numbers. When \(m=t\), we recover the original problem.
This setting allows us to explore the possibility of EFX for a small number of types, as we don’t know whether it exists in the general setting. This is a new abstraction that has not been studied before. Here we compare it to other abstractions when studying EFX:
“k types” is not the same as “k players”: Restricting the number of types means there can be an arbitrary number of players. Some previous literature studies the existence of EFX with a small number of players. Our setting is natural for allocations among large groups.
“k types” is not the same as “k distinct values”: If there are k types, the valuation function can vary from player to player. This is much less restrictive than “k distinct values” (as previously studied in [1]), a generalization of identical valuations where all valuations across all players must be one of k distinct values. Our setting is natural when agents assign monetary value to each item. Here, the agents do not agree upon a discrete set of distinct values, but still can define a valuation with a small set of numbers.
“k types” is not the same as “approximate EFX” or “EFX with charity”: Even if the number of types is small, we still find complete EFX solutions.
We prove the existence of EFX for 2 types. To build upon this work and abstraction, we suggest that proving whether EFX exists for 3 or more types will assist the eventual goal of proving whether EFX exists in the general setting (as many types as items).