Exact Matching Problem Overview

Updated 8 July 2026

Exact Matching Problem is a decision problem in combinatorial optimization that seeks a perfect matching with exactly k red edges.
It leverages algebraic techniques such as Tutte matrices and pfaffians, with randomized algorithms and recent deterministic advances for bipartite graphs.
Recent research highlights parameterized methods and structural decompositions, underscoring its role as a benchmark for testing derandomization hypotheses.

In combinatorial optimization and complexity theory, the Exact Matching Problem usually denotes the following decision problem: given a graph $G=(V,E)$ whose edges are colored red or blue, and an integer $k$ , determine whether $G$ contains a perfect matching $M$ such that exactly $k$ edges of $M$ are red, equivalently $|M \cap R| = k$ where $R\subseteq E$ is the set of red edges (Maalouly} et al., 2022). The problem was introduced in 1982, received a randomized polynomial-time algorithm in 1987, and has long served as a canonical test case for the $\mathbf{RP}=\mathbf{P}$ hypothesis. As of 2026, the bipartite case admits a deterministic polynomial-time algorithm, while the existence of a deterministic polynomial-time algorithm for general graphs remains open (Du, 2 Apr 2026).

1. Formal statement and equivalent formulations

The standard input consists of a graph $G=(V,E)$ , an edge-coloring with two colors, red and blue, and an integer $k$ 0. The objective is to find a perfect matching $k$ 1 satisfying

$k$ 2

This is the exact form of a colored perfect matching problem: the constraint is not “at least” or “at most,” but exactly a prescribed number of red edges (Maalouly} et al., 2022).

A common equivalent formulation replaces colors by $k$ 3 edge weights. In that form, each edge has weight $k$ 4 or $k$ 5, and the task is to decide whether there exists a perfect matching of total weight exactly $k$ 6 (Murakami et al., 2024). This weighted view is especially natural in algebraic treatments based on Tutte matrices and pfaffians, and in reductions to related problems such as linear matroid parity (Sato et al., 6 Aug 2025).

The exactness requirement is the source of the problem’s difficulty. Ordinary perfect matching asks only for existence of a perfect matching, whereas Exact Matching imposes an exact quota on one distinguished edge class. Several later variants weaken that quota in controlled ways. A prominent example is the parity relaxation, which asks only for a perfect matching $k$ 7 such that

$k$ 8

and another is the bounded parity form requiring

$k$ 9

(Maalouly} et al., 2022).

2. Historical development and complexity-theoretic status

Papadimitriou and Yannakakis introduced the problem in 1982 and conjectured it to be $G$ 0-complete. Soon afterward, Mulmuley, Vazirani, and Vazirani gave a randomized polynomial-time algorithm, placing the problem in $G$ 1; later accounts also note a randomized polylogarithmic-time parallel algorithm and membership even in $G$ 2 (Maalouly et al., 2022). Despite decades of work, no deterministic polynomial-time algorithm was known for the general problem, and this made Exact Matching one of the standard natural candidates for testing the $G$ 3 hypothesis (Dürr et al., 2023).

The historical trajectory is unusually sharp: randomized tractability was established early, but deterministic tractability resisted both combinatorial and algebraic attacks. This gap is why the problem occupies a distinctive place in derandomization-oriented complexity theory.

Milestone	Content	Source
1982	Problem introduced by Papadimitriou and Yannakakis	(Maalouly et al., 2022)
1987	Randomized polynomial-time algorithm by Mulmuley, Vazirani, and Vazirani	(Maalouly} et al., 2022)
2026	Deterministic $G$ 4 algorithm for all bipartite graphs	(Du, 2 Apr 2026)

The 2026 bipartite breakthrough changed the status of one major special case. "Bipartite Exact Matching in P" proves the Affine-Slice Nonvanishing Conjecture for all bipartite braces and derives a deterministic $G$ 5 algorithm for Exact Matching on all bipartite graphs via tight-cut decomposition and structural induction on McCuaig’s brace decomposition (Du, 2 Apr 2026). The general-graph problem, however, remains unresolved.

A recurrent misconception is that randomized solvability already makes the problem “essentially settled.” The literature shows the opposite. Exact Matching is easy neither by straightforward derandomization of the 1987 method nor by generic reductions to standard matching algorithms; its status has instead advanced through special graph classes, parameterized algorithms, parity relaxations, and class-specific structural decompositions (Maalouly} et al., 2022).

3. Algebraic methods and parity-based relaxations

The classical randomized approach is algebraic. In the $G$ 6-weighted formulation, one considers a bivariate Tutte matrix

$G$ 7

where edges in $G$ 8 have weight $G$ 9 and edges in $M$ 0 have weight $M$ 1. The exact matching question reduces to testing whether the coefficient $M$ 2 is nonzero (Sato et al., 6 Aug 2025). This connects Exact Matching to polynomial identity testing and explains its longstanding relationship to randomized algebraic algorithms.

A recent refinement shows that the decision problem for all target values $M$ 3 can be solved with high probability in asymptotically the same time order as matrix multiplication. More precisely, the paper "Exact Matching in Matrix Multiplication Time" gives an $M$ 4-time randomized test for all exact weights simultaneously, and for any fixed feasible $M$ 5 it gives a deterministic $M$ 6-time self-reduction to construct a matching of weight exactly $M$ 7 (Sato et al., 6 Aug 2025). This preserves the randomized character of the fastest known general-graph feasibility tests, but substantially sharpens their complexity.

A separate line of work isolates parity as a tractable surrogate for exact counting. For general graphs, one can compute in polynomial time a perfect matching whose number of red edges has the correct parity,

$M$ 8

even when exact equality is not known to be achievable deterministically in polynomial time (Maalouly} et al., 2022). That paper also shows that an FPT algorithm for Exact Matching parameterized by the independence number reduces to finding, in polynomial time, a perfect matching with at most $M$ 9 red edges and the correct parity.

Parity relaxations are not merely technical conveniences. They identify an invariant that is algorithmically accessible in general graphs and that can be combined with structural restrictions to recover exactness. Related work reformulates the parity-only version through odd alternating cycles: starting from a perfect matching $k$ 0, parity can be flipped exactly when there exists an $k$ 1-alternating cycle of odd weight (Murakami et al., 2024). The same paper also shows that a natural localized variant—odd alternating cycle through a specified matching edge—is NP-hard even in bipartite graphs with exactly one matching edge of weight $k$ 2 (Murakami et al., 2024).

4. Deterministic tractability on special graph classes

Before the full bipartite result, deterministic progress came primarily from dense or highly structured graph classes. A major step was the proof that Exact Matching is solvable in deterministic polynomial time on graphs of bounded independence number and on bipartite graphs of bounded bipartite independence number. These results generalized earlier positive results for complete graphs and complete bipartite graphs, which had been the principal dense-graph cases previously understood (Maalouly et al., 2022).

Subsequent work extended the deterministic frontier well beyond those initial dense classes. "On the Exact Matching Problem in Dense Graphs" proves deterministic polynomial-time solvability for complete $k$ 3-partite graphs, unit interval graphs, bipartite unit interval graphs, graphs of bounded neighborhood diversity, chain graphs, and graphs without a complete bipartite $k$ 4-hole, and gives a deterministic quasi-polynomial-time algorithm for Erdős–Rényi random graphs $k$ 5 (Maalouly et al., 2024). The two main tools are a local search algorithm based on the path-shortening property and a Karzanov-style structural framework based on a chord property (Maalouly et al., 2024).

These deterministic results are structurally heterogeneous. In the path-shortening framework, alternating paths can be locally shortened, yielding locality properties for perfect matchings and enabling bounded-radius local search. In the Karzanov-style framework, certain graph classes satisfy an interpolation principle: between two perfect matchings of the same parity, intermediate red-edge counts of the same parity are also realized (Maalouly et al., 2024). This suggests that exactness becomes tractable when the space of perfect matchings is sufficiently connected under controlled local modifications.

The 2026 bipartite result is qualitatively different from the dense-class results. Rather than relying on a density proxy such as bounded independence number or neighborhood diversity, it gives a deterministic polynomial-time algorithm for all bipartite graphs. The proof passes through tight-cut decomposition into brace blocks, a determinant-based exact- $k$ 6 polynomial $k$ 7, and the Affine-Slice Nonvanishing Conjecture for braces; the entire proof architecture is formally verified in Lean 4 (Du, 2 Apr 2026). This resolves a longstanding open status specifically for bipartite Exact Matching, while leaving the nonbipartite case open.

5. Parameterized, optimization, and approximation variants

Parameterized complexity has supplied some of the strongest general-graph positive results. Exact Matching is fixed-parameter tractable when parameterized by the independence number $k$ 8. This improves an earlier state of the art in bipartite graphs, where bounded bipartite independence number yielded only an XP algorithm, and it does so by combining new algorithmic techniques with the parity relaxation described above (Maalouly} et al., 2022).

Another parameterized extension treats graphs that are close to bipartite. "An FPT Algorithm for the Exact Matching Problem and NP-hardness of Related Problems" gives a deterministic FPT algorithm for Exact Matching parameterized by the minimum size of an odd cycle transversal together with the independence number, sharpened to the bipartite independence number after removing that transversal (Murakami et al., 2024). The core reduction branches over the odd cycle transversal and reduces the resulting instances to bipartite subproblems.

A substantial reorientation of the subject comes from the optimization problem Top- $k$ 9 Perfect Matching (TkPM), which asks for a perfect matching maximizing the total weight of the $M$ 0 heaviest edges it contains. Work in 2022 showed that TkPM is polynomially equivalent to Exact Matching, first by introducing TkPM and then by giving the missing reduction in the opposite direction (Maalouly, 2022). This equivalence matters because optimization problems often admit approximation and parameterized techniques that are less natural in the original exact-feasibility formulation.

One approximation line studies EM-opt, where the goal is to maximize the number of red edges in a perfect matching subject to not exceeding a budget $M$ 1. For bipartite graphs, there is a deterministic polynomial-time $M$ 2-approximation for EM-opt: the algorithm returns a perfect matching with $M$ 3 red edges satisfying

$M$ 4

whenever $M$ 5 is feasible for Exact Matching, and more generally $M$ 6 for optimum value $M$ 7 (Dürr et al., 2023). This is explicitly a one-sided approximation: it never exceeds the target budget.

More recent work on blow-ups of bounded-bandwidth prototypes studies Exact Matching and TkPM through neighborhood diversity and bandwidth. It gives an FPT algorithm for TkPM parameterized by $M$ 8 and neighborhood diversity, an approximation scheme for the same setting, a subexponential algorithm for TkPM on bounded-bandwidth blow-ups, and an adaptation of the recursive subexponential approach to Exact Matching (Maalouly et al., 14 Oct 2025). This places Exact Matching in a broader algorithmic family where the prototype graph rather than the full blown-up instance governs the complexity.

The problem also has a strong negative polyhedral result. The convex hull of exact matchings in bipartite graphs has exponential extension complexity: there is no sub-exponential sized linear program describing the exact-matching polytope, and in fact the stronger odd-red parity perfect matching polytope already has extension complexity $M$ 9 (Jia et al., 2022). For LP-based approaches, this is a fundamental barrier. It rules out compact extended formulations analogous to those available for ordinary bipartite perfect matching.

That lower bound should not be misread as a barrier to all deterministic algorithms. It excludes compact linear descriptions, not algebraic nonvanishing arguments, structural decompositions, or specialized dynamic programs. The deterministic $|M \cap R| = k$ 0 bipartite algorithm of 2026 is precisely of that non-LP type: it proceeds through tight-cut decomposition, brace structure, and determinant identities rather than through a small polyhedral formulation (Du, 2 Apr 2026). A plausible implication is that Exact Matching resists “easy” convex descriptions even when deterministic decision algorithms do exist.

Within graph algorithms, the phrase Exact Matching conventionally refers to the red/blue perfect matching problem described above. In other literatures, however, the same phrase can denote different tasks. In exact string matching, it means finding all occurrences of a pattern string in a text (Divakaran, 2015). In correlated graph inference, exact matching can mean exact recovery of the hidden vertex permutation between correlated graphs, sometimes with node features (Yang et al., 2024). These are distinct problems with different models and objectives; the classical complexity-theoretic Exact Matching problem is the red/blue perfect matching problem.

The current state of the subject is therefore sharply stratified. Bipartite Exact Matching is in deterministic polynomial time (Du, 2 Apr 2026). Many dense, structured, and parameterized general-graph regimes are also tractable (Maalouly et al., 2022). Fast randomized algebraic algorithms remain available in general graphs, including matrix-multiplication-time feasibility tests (Sato et al., 6 Aug 2025). Yet the existence of a deterministic polynomial-time algorithm for Exact Matching on arbitrary graphs remains open, preserving the problem’s role as a central benchmark for derandomization and exact counting in combinatorial optimization (Maalouly} et al., 2022).