Introduction
In the field of fair division, researchers typically focus on "subjective" preferences—the principle that different individuals value the same items differently. For instance, one person might prize a vintage typewriter for its aesthetic, while another prefers a modern laptop for its utility. However, real-world scenarios such as divorce settlements, estate liquidations, or corporate dissolutions often introduce a critical third-party constraint: the "Fair Market Value" (FMV). Legal and normative systems frequently demand an "equal" split based on objective market prices, regardless of the agents' personal valuations.
This research, presented at AAAI-2025, introduces a formal model to bridge the gap between these two often-conflicting objectives. The goal is to find allocations that are simultaneously fair to the agents' personal tastes and balanced according to objective market values.
The Challenge of Dual Fairness
Traditional fair division theory, originating with Steinhaus in 1948, focuses on heterogeneous preferences to find envy-free allocations. The introduction of a shared market value complicates this significantly. An allocation that is perfectly balanced by market price might leave an agent feeling unsatisfied because the items they received do not align with their personal needs. Conversely, a subjectively fair allocation might result in one agent receiving a bundle worth significantly more on the open market, potentially leading to legal challenges.
The paper explores whether it is possible to satisfy both criteria. The authors model the problem with $n$ agents and $m$ indivisible goods, where each agent has an additive subjective utility function, and a single additive market value function applies to all.
Methodological Framework
The researchers investigate several fairness notions, primarily focusing on:
- EF1 (Envy-Free up to one good): An agent does not envy another if one item is removed from the other's bundle.
- SD-EF1 (Stochastic Dominance EF1): A stronger variant of EF1 that accounts for the ranking of items.
- Minimal Envied Swap (MES) Algorithm: A novel approach designed for "subadditive" valuations, where the value of a bundle may be less than the sum of its parts.
The study employs a combination of impossibility proofs—constructing specific scenarios where fairness cannot be mathematically achieved—and algorithmic reductions. A key finding is that achieving EF1 for subjective values while maintaining market fairness can be reduced to the known problem of "Fair Division under Cardinality Constraints."
Key Theoretical Findings
The research identifies a "sweet spot" in the trade-off between subjective and objective fairness. While it is not always possible to satisfy the strong SD-EF1 requirement for both subjective and market values simultaneously (a 7-good counter-example proves this impossibility), the authors demonstrate that an allocation providing EF1 for agents and SD-EF1 for the market always exists and can be computed in polynomial time.
As shown in the table above, the compatibility between different fairness notions varies. For example, while EF1 (Agents) and SD-EF1 (Market) are compatible (Corollary 1), Pareto Optimality (PO) for agents is incompatible with SD-EF1 for the market (Theorem 4). This suggests that achieving objective market balance may require accepting a slightly "inefficient" allocation from a subjective standpoint.
Extensions and Limitations
The paper extends its findings to more complex scenarios:
- Subadditive Utilities: For agents with subadditive preferences, the MES algorithm guarantees a "1/2-EF1" allocation alongside market fairness.
- Cake-Cutting: In the context of divisible goods (e.g., land), the authors prove that $2n-2$ cuts are necessary to achieve envy-freeness on both subjective and market fronts.
- Impossibility of Stronger Notions: Stronger fairness guarantees like Maximin Share (MMS) or Envy-Free up to any good (EFX) are shown to be "starkly impossible" to maintain when market values are introduced.
Future Directions
The study opens several avenues for future research. A primary open question is the "Reverse Guarantee": whether an allocation that is SD-EF1 for subjective values and EF1 for market values always exists. Additionally, extending the model to "chores" (items with negative value, like debt) and quantifying the "price of fairness" remains a key challenge for the field.
Summary updated