Ex-post Stability under Two-Sided Matching: Complexity and Characterization
New paper solves a major open problem in algorithmic matching, with implications for AI agents and market design.
A team of computer scientists has solved a fundamental open problem in algorithmic game theory, providing critical insights for designing stable AI-powered matching systems. In their paper 'Ex-post Stability under Two-Sided Matching: Complexity and Characterization,' Haris Aziz, Gergely Csáji, and Péter Biró investigate probabilistic approaches to the classic stable matching problem—a framework used in everything from medical residency placements to dating apps and AI agent coordination.
The researchers' central breakthrough establishes that testing for 'ex-post stability'—a key concept for random matchings—is NP-complete when either side of the matching has ties in their preferences or priorities. This computational hardness holds even in simplified scenarios where both sides have dichotomous preferences (simple 'yes/no' choices). This negative result is balanced by a positive algorithmic contribution: the team developed an integer programming approach that can determine a decomposition with maximum probability of achieving 'weak stability,' providing a practical method for system designers.
Beyond the core complexity result, the paper explores stronger stability concepts. The authors prove that testing for 'robust ex-post stability' and 'ex-post strong stability' can be done in polynomial time, creating a nuanced landscape of what's computationally feasible. This work provides essential guidance for engineers building next-generation matching markets where AI agents must operate reliably under uncertainty, ensuring these systems can be both practically implementable and provably stable.
- Proves testing 'ex-post stability' is NP-complete when preferences have ties, solving a major open problem.
- Provides an integer programming algorithm to find a decomposition with max probability of weak stability.
- Shows stronger stability concepts like 'robust ex-post stability' are testable in polynomial time.
Why It Matters
Provides foundational limits for designing stable, large-scale AI matching systems used in markets, logistics, and multi-agent coordination.