[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Probing EFX via PMMS: (Non-)Existence Results in Discrete Fair Division

Created by
  • Haebom

Author

Jaros{\l}aw Byrka, Franciszek Malinka, Tomasz Ponitka

Outline

This paper studies the fair distribution problem of indivisible items, and provides new insights into the EFX problem, which is considered a central unsolved problem in the field of fair distribution, and the PMMS problem, a more rigorous variant of EFX. First, we establish a formal separation between EFX and PMMS by showing that PMMS allocation does not exist for three agent instances with two monotonic cost functions and one additive cost function. Then, we prove the existence of fair allocation for three important special cases: personalized bivalued valuations, PMMS allocation when $a_i$ is divisible by $b_i$, and binary-valued MMS-feasible valuations. Finally, we show that PMMS allocation exists for pair-demand valuations, which are extensions of single-demand cost functions to at most two items. We provide polynomial-time algorithms for all existence results.

Takeaways, Limitations

Takeaways:
We deepen our understanding of the process distribution problem by clarifying the formal differences between EFX and PMMS problems.
We prove the existence of fair allocation for three important special cases: personalized binary cost functions, PMMS allocation under certain conditions, and binary cost functions (without monotonicity assumption), and provide polynomial-time algorithms for them.
We prove the existence of PMMS allocation for pair-demand valuations.
Limitations:
The proven existence results are limited to special cases. The existence of fair allocation in the general case remains an open question.
Only counterexamples for three agent instances are presented, and studies involving more agents are needed.
👍