Abstract:We show a fully dynamic algorithm for maintaining $(1+\epsilon)$-approximate \emph{size} of maximum matching of the graph with $n$ vertices and $m$ edges using $m^{0.5-\Omega_{\epsilon}(1)}$ update time. This is the first polynomial improvement over the long-standing $O(n)$ update time, which can be trivially obtained by periodic recomputation. Thus, we resolve the value version of a major open question of the dynamic graph algorithms literature (see, e.g., [Gupta and Peng FOCS'13], [Bernstein and Stein SODA'16],[Behnezhad and Khanna SODA'22]).
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to maintain an approximation of the maximum matching size in a dynamic graph. Specifically, the goal is to develop an algorithm that can provide a $(1 + \epsilon)$ - approximate maximum matching size with a polynomial - logarithmic time complexity for each edge update. This problem is an important open problem in the field of dynamic graph algorithms, especially for dense graphs, and previous methods have been unable to break through the $O(n)$ update time limit.
### Main contributions of the paper
1. **Proposing a new dynamic algorithm**: The paper proposes a fully dynamic algorithm that can maintain the size of a $(1 + \epsilon)$ - approximate maximum matching in graph $G$, with a worst - case update time of $m^{0.5-\Omega_\epsilon(1)}$. This is the first time that the long - standing $O(n)$ update time limit has been improved at the polynomial level.
2. **Solving an open problem**: The paper affirmatively answers a long - standing open problem, namely whether there exists a dynamic algorithm that can achieve a $(1 + \epsilon)$ - approximate maximum matching with sub - linear time complexity in dense graphs.
3. **Progress in sub - linear time matching algorithms**: The paper also presents the first $(1, \epsilon n)$ - approximate matching algorithm with truly sub - linear time complexity in dense graphs. Specifically, given the adjacency matrix of graph $G$, this algorithm can calculate a $(1, \epsilon n)$ - approximate value of $\mu(G)$ with high probability in $n^{2-\Omega_\epsilon(1)}$ time.
### Technical contributions
- **Design of the sub - routine `LargeMatching`**: The paper introduces a key sub - routine `LargeMatching(S, \delta)`, which can return a matching $M$ such that if $\mu(G[S])\geq\delta n$, then $|M|\geq\Omega(\text{poly}(\delta)n)$. This sub - routine can provide an approximation ratio with an arbitrary difference even in the case of additive approximation.
- **Efficient implementation of matching queries**: The paper implements the function of the `LargeMatching` sub - routine in the sub - linear time model by constructing a sub - routine `LargeMatchingOracle(A, \delta, \epsilon)` for matching queries. This sub - routine can be constructed in $\tilde{O}(\delta(n^{2 - \epsilon}))$ pre - processing time and supports matching queries in $\tilde{O}(\delta(n^{1 + g(\epsilon)}))$ time.
- **Implementation of the dynamic algorithm**: The paper uses the above sub - linear time matching algorithm, combined with known dynamic graph algorithm techniques, to finally implement an algorithm for dynamic $(1 + \epsilon)$ - approximate maximum matching size. Specifically, by calling the sub - linear time algorithm at the beginning of each stage to obtain a $(1, \gamma^2n)$ - approximate estimate, a matching size with a $(1+\Theta(\gamma))$ - approximation is maintained throughout the stage.
### Conclusion
Through innovations in the fields of sub - linear time and dynamic graph algorithms, this paper successfully breaks through the long - standing $O(n)$ update time limit and provides a new solution to the maximum matching problem in dynamic graphs. This result is not only of great theoretical significance, but also provides new ideas for the design of efficient graph algorithms in practical applications.