Existence and Computation of Fair Allocations under Constraints

A team of computer scientists from the Indian Institute of Science and the University of Patras has published groundbreaking research on fair division algorithms under realistic constraints. Their paper 'Existence and Computation of Fair Allocations under Constraints' proves that Feasible Envy-Free (FEF) allocations—where agents don't envy each other's budget-feasible bundles—can always be achieved alongside Pareto-optimality, even when goods have agent-specific values and sizes. This addresses a fundamental challenge in algorithmic fairness: how to allocate resources fairly when agents have different budgets and valuations, with leftover goods handled through a 'charity' construct.

The research reveals the non-convex structure of FEF allocation spaces and uses sophisticated fixed-point arguments to establish their main existential result. Crucially, they demonstrate the impossibility of achieving truthfulness alongside both fairness and Pareto-optimality—a significant limitation for mechanism design. However, they show truthfulness can be paired individually with either FEF or Pareto-optimality. These findings provide essential mathematical foundations for developing AI systems that allocate computational resources, cloud credits, or shared infrastructure fairly among users with different needs and constraints.