In classical fair division, we assume every participant has an equal claim to the resources at hand. But real-world scenarios are rarely so symmetric. A business partner with a 60% stake expects more than one with 40%; a large city deserves more medical supplies than a small town; and a senior employee might be entitled to fewer unpleasant shifts than a trainee.
When resources are "divisible"—like money or land—we simply scale the portions. However, when resources are "indivisible"—like houses, specialized equipment, or specific work shifts—the math breaks. You cannot give someone 60% of a car. This research direction explores how to distribute discrete items among agents with unequal entitlements (or weights) while maintaining rigorous standards of fairness.
Scaling the Definition of Fairness
To handle asymmetry, researchers must first redefine what "fair" means. In a symmetric world, Envy-Freeness (EF) means agent $i$ prefers their own bundle $A_i$ over anyone else's bundle $A_j$. In a weighted world, we use Weighted Envy-Freeness (WEF).
The intuition is to normalize an agent's value by their weight $w_i$. As Warut Suksompong and Pasin Manurangsi describe in Asymptotic Analysis of Weighted Fair Division, fairness is achieved when the "value-per-unit-of-entitlement" for an agent's own bundle is at least as high as what they perceive for others. Mathematically, agent $i$ is not "weighted-envious" if: $$\frac{v_i(A_i)}{w_i} \geq \frac{v_i(A_j)}{w_j}$$
Because indivisible items make perfect envy-freeness impossible (one person always gets the "better" single item), researchers focus on relaxations like Weighted EF1 (envy-free up to one item). Warut Suksompong notes in Expanding the Reach of Social Choice Theory that WEF1 allocations are guaranteed to exist and can be found efficiently.
The Geometry of Picking Sequences
One of the most intuitive ways to divide items is a "picking sequence": agents take turns selecting their favorite remaining item. In a symmetric setting, a simple round-robin ($1, 2, \dots, n, 1, 2, \dots$) works well.
In the weighted setting, the sequence must reflect the entitlements. Mithun Chakraborty, Ulrike Schmidt-Kraepelin, and Warut Suksompong show in Picking Sequences and Monotonicity in Weighted Fair Division that we can borrow techniques from political science—specifically apportionment methods used to assign parliamentary seats to states. By treating each "turn" in the sequence as a seat to be won, these algorithms ensure that agents with higher weights get to pick more often and earlier, naturally satisfying weighted fairness notions.
Weighted Maximin Shares
Another pillar of fairness is the Maximin Share (MMS): the value an agent can guarantee themselves if they were the one to partition the items into $n$ bundles, but were forced to take the worst one.
When agents have weights, this "worst-case" guarantee must be adjusted. Haris Aziz and Bo Li pioneered the Weighted Maximin Share (WMMS) in Weighted Maxmin Fair Share Allocation of Indivisible Chores. Here, the agent imagines partitioning items into many small pieces and grouping them into bundles proportional to the agents' weights.
While finding a perfect WMMS allocation is often impossible, recent breakthroughs have provided strong approximations. For example, Saeed Seddighin proved in Fair Assignment of Indivisible Chores to Asymmetric Agents that even for "chores" (items with negative value), we can achieve a constant-factor guarantee for WMMS, ensuring no agent is stuck with an unfairly large burden relative to their obligation.
The Challenge of Chores and Costs
Allocating "chores"—tasks that agents want to avoid—is fundamentally different from allocating "goods." In Almost (Weighted) Proportional Allocations for Indivisible Chores, Bo Li and Xiaowei Wu demonstrate that while some fairness notions for goods translate easily, others do not.
For chores, the goal shifts to Weighted Proportionality up to any item (PROPX). The intuition here is that an agent's cost should not exceed their "fair share" by more than the cost of any single task they were assigned. This is particularly relevant in logistics; as Piotr Matl and Richard F. Hartl argue in Equity Objectives in Vehicle Routing, ignoring driver weights (like seniority or contract hours) leads to "artificial equity" where everyone is equally overworked, rather than fairly balanced.
Balancing Efficiency and Fairness
Finally, researchers look at Weighted Nash Social Welfare (WNSW)—maximizing the product of agents' utilities, each raised to the power of their weight ($\prod v_i^{w_i}$). This objective is the "gold standard" because it naturally balances efficiency (total utility) and fairness (no one gets zero).
Computing the maximum WNSW is NP-hard, but Yuda Feng and Ruilong Zhang recently provided the first constant-factor approximation for this problem under complex submodular preferences in Constant Approximation for Weighted Nash Social Welfare with Submodular Valuations. This ensures that even in highly asymmetric systems, we can find allocations that are both globally productive and individually fair.