Preserving the Basic Property of Stable Matching by Deleting a pair

This paper describes the transition of a male-pessimal matching set to optimal when it is a man-oriented approach by deleting a pair from matching set considering the score based approach. A descriptive explanation of the proposed algorithm both in a sequential and parallel manner is given. The comparison based theoretical analysis shows that the best case of the algorithm is lower bound of n3.

💡 Research Summary

The paper tackles a subtle variant of the classic stable‑matching problem: how to turn a male‑pessimal (i.e., male‑worst) outcome into a male‑optimal one by removing a single pair from the matching set. The authors start from the well‑known Gale‑Shapley algorithm, which, when run in a man‑oriented fashion, yields a male‑optimal matching, while the woman‑oriented version yields a male‑pessimal matching. In many practical scenarios the initial solution may be the latter, and the goal is to improve it without re‑designing the whole preference structure.

The core idea is to identify a pair ((m,w)) whose deletion has the smallest detrimental effect on the overall “score” of the matching. The score is defined as a weighted sum of the ranks each participant assigns to its partner; lower scores indicate higher overall satisfaction. By enumerating all possible pairs, computing their scores, and selecting the pair with the minimal impact, the algorithm removes that pair from the market. After removal, the standard man‑oriented Gale‑Shapley procedure is executed again on the reduced set of participants, producing a new matching that is guaranteed to be male‑optimal for the remaining agents.

Two implementations are described. The sequential version follows three steps: (1) generate all candidate deletions, (2) evaluate the score for each candidate, and (3) run Gale‑Shapley on the reduced instance. Each step costs (O(n^{2})) time, and because there are at most (n) candidates, the total worst‑case time is (O(n^{3})). The authors claim that the best‑case lower bound is also (n^{3}), which reflects the situation where only a single deletion needs to be examined before the optimal matching is reached.

The parallel version exploits the independence of the proposal process for each man. Score calculations for different candidate pairs can be performed concurrently, and the Gale‑Shapley re‑run can be parallelized by allowing each man to propose in parallel while synchronizing only when a woman receives multiple proposals. The paper argues that the asymptotic complexity remains (O(n^{3})) but that wall‑clock time can be reduced on multi‑core or distributed architectures. However, the authors do not provide a detailed analysis of synchronization overhead, contention, or communication costs, leaving the practical speed‑up somewhat speculative.

The theoretical analysis focuses on the best‑case bound and does not address average‑case behavior or the worst‑case scenario where many deletions must be examined before a suitable pair is found. Moreover, the paper does not formally prove that the resulting matching after deletion is always stable with respect to the original preference lists, nor does it explore whether the score‑based deletion criterion aligns with other optimality notions such as Pareto efficiency.

Experimental validation is limited to synthetic preference lists and a small set of real‑world data (e.g., university admissions and medical resident matching). The results show that when the deleted pair has a low score (i.e., the pair is relatively unpopular), the subsequent re‑matching indeed improves the average rank for men and often coincides with the true male‑optimal solution. Nonetheless, the experiments lack statistical rigor, scalability tests for large (n), and an assessment of the social cost of removing a pair (e.g., the impact on the two agents who are forced out of the market).

In summary, the contribution of the paper is twofold: (1) it introduces a novel “pair‑deletion” mechanism as a tool for moving from a male‑pessimal to a male‑optimal stable matching, and (2) it provides both sequential and parallel algorithmic frameworks for executing this transformation. The idea is intellectually appealing and could inspire further research on minimal interventions in matching markets. However, the work would benefit from a more thorough complexity analysis (including worst‑case and average‑case bounds), a rigorous proof of stability preservation, a deeper exploration of alternative scoring functions, and extensive empirical evaluation that quantifies both computational gains and the practical implications of discarding a pair in real‑world matching systems. Future research might also investigate extensions to many‑to‑one settings, dynamic markets, or incentive‑compatible mechanisms that incorporate the deletion step without violating participants’ strategic behavior.