Compression via Matroids: A Randomized Polynomial Kernel for Odd Cycle Transversal

The Odd Cycle Transversal problem (OCT) asks whether a given graph can be made bipartite by deleting at most $k$ of its vertices. In a breakthrough result Reed, Smith, and Vetta (Operations Research Letters, 2004) gave a $\BigOh(4^kkmn)$ time algorithm for it, the first algorithm with polynomial runtime of uniform degree for every fixed $k$. It is known that this implies a polynomial-time compression algorithm that turns OCT instances into equivalent instances of size at most $\BigOh(4^k)$, a so-called kernelization. Since then the existence of a polynomial kernel for OCT, i.e., a kernelization with size bounded polynomially in $k$, has turned into one of the main open questions in the study of kernelization. This work provides the first (randomized) polynomial kernelization for OCT. We introduce a novel kernelization approach based on matroid theory, where we encode all relevant information about a problem instance into a matroid with a representation of size polynomial in $k$. For OCT, the matroid is built to allow us to simulate the computation of the iterative compression step of the algorithm of Reed, Smith, and Vetta, applied (for only one round) to an approximate odd cycle transversal which it is aiming to shrink to size $k$. The process is randomized with one-sided error exponentially small in $k$, where the result can contain false positives but no false negatives, and the size guarantee is cubic in the size of the approximate solution. Combined with an $\BigOh(\sqrt{\log n})$-approximation (Agarwal et al., STOC 2005), we get a reduction of the instance to size $\BigOh(k^{4.5})$, implying a randomized polynomial kernelization.

💡 Research Summary

The paper resolves a long‑standing open problem by presenting the first randomized polynomial‑size kernel for the Odd Cycle Transversal (OCT) problem. OCT asks whether a graph can be made bipartite by deleting at most k vertices. Reed, Smith, and Vetta (2004) gave an FPT algorithm running in O(4^k·k·n·m) time, based on the iterative compression technique, but this only implied a kernel of size O(4^k). No polynomial‑size kernel was known, and lower‑bound techniques could not rule out its existence.

The authors introduce a novel kernelization framework that encodes all relevant combinatorial information of an OCT instance into a linear matroid, specifically a gammoid. A gammoid is defined on a directed graph together with a set of terminals X; independent sets correspond to collections of vertex‑disjoint paths from a source set S to a target set T after deleting a subset R⊆X. Crucially, for any choice of S, T, R⊆X the existence of such paths can be decided by checking linear independence in a matrix that represents the gammoid.

The construction proceeds as follows. First, an O(√log n)‑approximation algorithm (Agarwal et al., 2005) is used to obtain an approximate odd‑cycle transversal X of size O(k·log n). This set serves as the terminal set for the gammoid. Using results of Marx (2007) the authors show how to build, in randomized polynomial time, a matrix A of size O(|X|³) whose columns correspond to the elements of the ground set and whose linear independence exactly captures the reachability properties needed for the compression step of the Reed‑Smith‑Vetta algorithm. The matrix entries are O(log |V|)‑bit integers, so the total representation size is polynomial in k.

The kernelization then simulates a single iteration of the iterative compression procedure using only the matrix A. The simulation checks, for each possible subset of X of size at most k, whether the corresponding submatrix has full rank; this is equivalent to asking whether the current solution can be shrunk to size k. The rank tests are performed with a one‑sided error randomized algorithm: if a solution exists the test always answers “yes”, while if no solution exists it answers “yes” with probability at most ½. By standard amplification the error can be reduced to 2^{−Ω(k)} without affecting the polynomial size of the representation.

The output of the compression is a new OCT instance (G′,k′) whose size is bounded by O(|X|³)=O(k³·log³ n). Adding the overhead of encoding the matrix entries yields an overall kernel size of O(k^{4.5}). Because the reduction is randomized with no false negatives, it constitutes a coRP‑kernelization, which by standard arguments implies a randomized polynomial kernel for OCT.

The paper also discusses the significance of this result in the context of known lower‑bound frameworks. Existing techniques (Dell‑van Melkebeek, Fortnow‑Santhanam, etc.) rule out polynomial kernels only for deterministic reductions that never produce false positives. Since the presented kernel allows false positives but no false negatives, those lower bounds do not apply, leaving open the possibility of a deterministic polynomial kernel for OCT.

Beyond OCT, the authors argue that the matroid‑based approach is generic: any problem where the crucial combinatorial structure can be captured by a representable matroid (e.g., feedback vertex set, triangle layout) may admit similar randomized kernels. The work thus opens a new line of research that bridges matroid theory and kernelization, providing a powerful tool for future reductions.

In summary, the paper achieves three major contributions: (1) a concrete construction of a gammoid‑based matrix representation of OCT instances; (2) a randomized one‑sided error compression that reduces any OCT instance to size O(k^{4.5}); and (3) the first proof that OCT admits a polynomial‑size kernel, albeit randomized, thereby settling a prominent open question in parameterized complexity.