Simultaneously Efficient Allocation of Indivisible Items Across Multiple Dimensions

Abstract: Many allocation problems are intrinsically multidimensional, since an item may contribute differently to several criteria, and optimizing a single aggregate objective can hide severe losses in other dimensions. We study how much efficiency can be guaranteed simultaneously when indivisible items have multiple attributes. To this end, we introduce the \emph{multidimensional efficient allocation} (MDEA) model, where each agent has an additive valuation in each dimension, and investigate simultaneous efficiency under utilitarian social welfare (USW) and egalitarian social welfare (ESW). Our results reveal a sharp worst-case frontier. For exact efficiency, maximizing the number of dimensions attaining the USW optimum admits a $c/\ell$-approximation for every fixed constant $c$, and this dependence on the number $\ell$ of dimensions is essentially unavoidable; for ESW, even deciding whether two dimensions can be optimized simultaneously is NP-hard with binary valuations. For approximate simultaneous efficiency in every dimension, we identify a tight threshold of order $1/\ell$, showing that such guarantees always exist for both USW and ESW, while any asymptotically better dependence on $\ell$ is impossible, even for binary valuations. Finally, we introduce three natural multidimensional Pareto notions and characterize both their relationships and their computational complexity.

• The paper introduces the MDEA model for simultaneous allocation efficiency, revealing that maximizing USW and ESW dimensions exactly is often infeasible.
• It establishes a tight 1/ℓ threshold for approximate efficiency, providing polynomial-time algorithms with dimension-dependent guarantees.
• The study proves NP-hardness and coNP-completeness for various Pareto optimality notions, highlighting fundamental tradeoffs in multidimensional allocation.

Simultaneous Efficiency in Multidimensional Indivisible Item Allocation

Model and Motivation

The paper introduces the Multidimensional Efficient Allocation (MDEA) model, which abstracts resource allocation problems where each indivisible item possesses multiple attributes and each agent has additive valuations per attribute (dimension). This model generalizes classical allocation frameworks by incorporating multidimensional evaluation criteria, directly addressing the inefficiencies resulting from aggregating multiple objectives into a single score. The motivation spans practical domains such as cloud computing (CPU, memory, bandwidth) and personnel assignment (skills, experience, communication), where optimizing any one criterion in isolation can substantially degrade performance in others.

Notions of Efficiency and Welfare

The study examines simultaneous efficiency regarding two welfare measures: utilitarian social welfare (USW), defined as the sum of agents' utilities per dimension, and egalitarian social welfare (ESW), defined as the minimum utility across agents per dimension. The goal is to attain maximum USW or ESW in all dimensions simultaneously (sUmax, sEmax). The work further introduces relaxed versions that allow approximations with controlled additive losses stemming from items not optimally contributing to a given dimension, formalized as $\alpha$-sUmax$\,c$ and $\alpha$-sEmax$\,c$ allocations.

Theoretical Results: Frontiers and Tradeoffs

Exact Simultaneous Efficiency

• Impossibility of Exact Solutions: Even simple instances preclude sUmax or sEmax allocations for both criteria; maximizing total welfare in every dimension simultaneously is infeasible despite optimal values existing per dimension in isolation.
• Maximizing Efficient Dimensions: For USW, maximizing the number of dimensions achieving the optimum is polynomial-time tractable when either the number of items or dimensions is constant, with a $c/\ell$-approximation achievable for fixed $c$. Tight NP-hardness results prohibit approximations better than $1/\ell^{1-\epsilon}$ for any $\epsilon > 0$, even with two agents and binary valuations. For ESW, maximizing the number of dimensions is NP-hard already for two dimensions with binary valuations.

Approximate Simultaneous Efficiency

• Sharp Dimension-Dependent Thresholds: For both USW and ESW, a tight $1/\ell$ threshold is established: a $(1/\ell)$-sUmax$\,c$0 or $\,c$1-sEmax$\,c$2 allocation always exists and is computable in polynomial time, but any $\,c$3 is infeasible even with binary valuations. These results quantify the necessary tradeoff as the number of dimensions increases.
• Aggregate Objectives: Maximizing the sum of USWs across dimensions is efficiently solvable, whereas maximizing the minimum USW or ESW across dimensions is NP-hard to approximate, reflecting the intrinsic difficulty of achieving strong simultaneous guarantees in every dimension.

Pareto Optimality in Multidimensional Allocation

Three multidimensional Pareto notions are considered:

• PO-agent: No allocation weakly improves every agent-dimension pair and strictly improves some.
• PO-USW: No allocation weakly improves the USW vector and strictly improves some entry.
• PO-ESW: No allocation weakly improves the ESW vector and strictly improves some entry.

Key structural results include:

• PO-USW implies PO-agent, but not vice versa.
• PO-USW and PO-ESW, as well as PO-agent and PO-ESW, are generally incomparable.
• There always exists an allocation that satisfies both PO-agent and PO-ESW.

Verifying PO-USW or PO-ESW is coNP-complete even for two agents and two dimensions; checking PO-agent is similarly hard and computing PO-ESW allocations is NP-hard even with binary two-dimensional instances.

Algorithmic Contributions

• Round-Robin and Iterative Rounding: The paper adapts round-robin selection (dimensions as virtual agents) and iterative rounding techniques to construct allocations that attain the $\,c$4 threshold for approximate simultaneous efficiency. These algorithms deliver allocations with dimension-dependent guarantees and controlled additive losses, and run in polynomial time.
• Lower Bound Constructions: Hardness and impossibility results are established via reductions from classical NP-complete problems such as Partition, 3-dimensional matching, and Hitting Set, as well as via explicit worst-case instance constructions.

Implications and Future Directions

The findings establish the fundamental limits of simultaneous welfare optimization in multidimensional settings. Guarantees linearly degrade with the number of dimensions, and this loss is intrinsic to general models absent further structure. Practical implications include the necessity of careful dimensionality reduction or aggregation before applying worst-case guarantees; the choice and representation of dimensions directly affect the achievable efficiency frontier. Future work may consider richer objective functions, allocation of chores, empirical studies, or structural assumptions that ameliorate the dimensionality tradeoff.

The theoretical results underscore the intractability of combining efficiency across multiple conflicting objectives in indivisible allocation, motivating both domain-specific modeling (dimension compression/selection) and the development of structurally guided algorithms. The separation between exact and approximate guarantees, as well as between various Pareto notions, points to nuanced tradeoffs in algorithm design and mechanism selection for fair division and resource allocation in increasingly multidimensional environments.

Conclusion

The paper provides a rigorous characterization of the simultaneous efficiency frontier for multidimensional indivisible item allocation, delineating the computational and structural limitations inherent to the general MDEA model. The tight $\,c$5 threshold for approximate simultaneous efficiency, the strong computational hardness for higher-dimensional Pareto notions, and the exponential paucity of exact efficient dimensions collectively establish both the baseline and the challenges for future algorithmic and economic research in multidimensional fair division and welfare optimization (2606.21346).