Formalizing Randomized Matching Algorithms

Using Je\v{r}'abek ’s framework for probabilistic reasoning, we formalize the correctness of two fundamental RNC^2 algorithms for bipartite perfect matching within the theory VPV for polytime reasoning. The first algorithm is for testing if a bipartite graph has a perfect matching, and is based on the Schwartz-Zippel Lemma for polynomial identity testing applied to the Edmonds polynomial of the graph. The second algorithm, due to Mulmuley, Vazirani and Vazirani, is for finding a perfect matching, where the key ingredient of this algorithm is the Isolating Lemma.

💡 Research Summary

The paper presents a rigorous formalization of two fundamental randomized parallel algorithms for bipartite perfect matching within the bounded‑arithmetic theory VPV, which captures polynomial‑time reasoning. The authors adopt Jan Jeřábek’s framework for probabilistic reasoning, which translates probabilistic events into logical formulas that can be proved inside VPV. This approach bridges the gap between the informal correctness arguments traditionally used for RNC² algorithms and a fully formal proof system.

The first algorithm addresses the decision problem: does a given bipartite graph contain a perfect matching? It relies on the Edmonds polynomial, a determinant‑based polynomial that evaluates to a non‑zero value exactly when a perfect matching exists. By applying a formal version of the Schwartz‑Zippel Lemma, the authors show that evaluating the polynomial on randomly chosen values from a sufficiently large finite field yields a non‑zero result with high probability. Within VPV they define the degree of the polynomial, bound the field size, and prove that the error probability is at most deg / |F|. Consequently, the algorithm runs in RNC² (log² depth parallel circuits) and its correctness is provable in VPV.

The second algorithm is the celebrated Mulmuley‑Vazirani‑Vazirani (MVV) isolating‑weight method for actually constructing a perfect matching. The core of MVV is the Isolating Lemma, which states that assigning random weights to edges from a suitably large range makes the minimum‑weight perfect matching unique with probability at least ½. The paper formalizes the random weight assignment as a probabilistic function w:E→{1,…,N} and encodes the event “there exists an isolated matching” as a VPV‑expressible formula. Using the same determinant trick, they show that if the weighted adjacency matrix has a non‑zero determinant, the unique minimum‑weight matching can be extracted in polynomial time. The authors prove inside VPV that the probability of isolation exceeds ½ when N is larger than the number of edges, and that the subsequent extraction step can be carried out by NC² circuits.

Both algorithms are shown to belong to the class RNC²: they admit parallel implementations with polylogarithmic depth and polynomial size, and their success probabilities are formally bounded. The paper’s contribution is twofold. First, it provides the first complete VPV‑level proofs for these classic randomized matching algorithms, demonstrating that Jeřábek’s probabilistic reasoning framework is powerful enough to capture non‑trivial algebraic arguments such as determinant‑based identity testing and the Isolating Lemma. Second, it establishes a methodology for formalizing other BPP or RNC algorithms, paving the way for automated verification of probabilistic algorithms within bounded‑arithmetic theories. This work thus deepens the connection between complexity theory, algebraic combinatorics, and formal proof systems, and it offers a solid foundation for future research on the mechanized verification of randomized computation.