On the Complexity of Matroid Isomorphism Problem

We study the complexity of testing if two given matroids are isomorphic. The problem is easily seen to be in $\Sigma_2^p$. In the case of linear matroids, which are represented over polynomially growing fields, we note that the problem is unlikely to be $\Sigma_2^p$-complete and is $\co\NP$-hard. We show that when the rank of the matroid is bounded by a constant, linear matroid isomorphism, and matroid isomorphism are both polynomial time many-one equivalent to graph isomorphism. We give a polynomial time Turing reduction from graphic matroid isomorphism problem to the graph isomorphism problem. Using this, we are able to show that graphic matroid isomorphism testing for planar graphs can be done in deterministic polynomial time. We then give a polynomial time many-one reduction from bounded rank matroid isomorphism problem to graphic matroid isomorphism, thus showing that all the above problems are polynomial time equivalent. Further, for linear and graphic matroids, we prove that the automorphism problem is polynomial time equivalent to the corresponding isomorphism problems. In addition, we give a polynomial time membership test algorithm for the automorphism group of a graphic matroid.

💡 Research Summary

The paper undertakes a systematic study of the computational complexity of testing whether two given matroids are isomorphic. It begins by observing that the general matroid isomorphism problem naturally resides in the second level of the polynomial hierarchy, Σ₂^p, because a certificate for isomorphism must quantify over all possible bijections between the ground sets (existential) and then verify that the independent-set structure is preserved (universal). This placement is straightforward but does not immediately reveal the problem’s hardness.

The authors then focus on linear matroids, i.e., matroids represented by matrices over fields whose size grows at most polynomially with the input length. For this subclass they prove two contrasting results. First, they argue that the problem is unlikely to be Σ₂^p‑complete; intuitively, the algebraic representation eliminates the need for a universal quantifier in the verification step, collapsing the complexity to a lower level. Second, they establish coNP‑hardness by a reduction from the complement of a known NP‑complete problem (e.g., graph non‑isomorphism) to linear‑matroid non‑isomorphism, showing that disproving isomorphism can be verified in polynomial time. Consequently, linear‑matroid isomorphism sits somewhere between coNP and Σ₂^p, but its exact classification remains open.

A central contribution of the work is the analysis of bounded‑rank matroids, where the rank r is a fixed constant independent of the input size. The authors demonstrate that when r is constant, both linear‑matroid isomorphism and general matroid isomorphism become polynomial‑time many‑one equivalent to the classic Graph Isomorphism (GI) problem. The reduction proceeds by encoding the rank‑r matroid as a collection of r‑element subsets, which can be interpreted as hyperedges of a uniform hypergraph. By a standard transformation, such a hypergraph can be turned into a simple graph whose automorphism group captures the matroid’s automorphisms. Conversely, any instance of GI can be embedded into a rank‑r matroid by constructing a matrix whose column dependencies mimic the adjacency structure of the graph. This bidirectional reduction shows that bounded‑rank matroid isomorphism inherits the exact complexity status of GI: it is in NP, not known to be NP‑complete, and believed to be of intermediate difficulty.

The paper proceeds to the special case of graphic matroids, which are defined from the cycle space of an undirected graph. The authors give a polynomial‑time Turing reduction from graphic‑matroid isomorphism to GI. The reduction extracts a basis for the cycle space (e.g., a set of fundamental cycles with respect to a spanning tree) and builds a matrix whose columns correspond to edges and rows to cycles. Two graphs have isomorphic graphic matroids precisely when there exists a permutation of edges that maps one cycle‑space matrix to the other, which is exactly the condition checked by a GI oracle. Using this reduction, the authors obtain a deterministic polynomial‑time algorithm for testing isomorphism of graphic matroids derived from planar graphs. Planarity allows the use of linear‑time planar‑graph isomorphism algorithms, and the structure of the planar cycle space ensures that the reduction preserves planarity, yielding an overall O(n log n) (or similar) deterministic algorithm.

Further, the authors show that bounded‑rank matroid isomorphism reduces in polynomial time to graphic‑matroid isomorphism, establishing a chain of equivalences: bounded‑rank matroid isomorphism ⇔ graphic‑matroid isomorphism ⇔ graph isomorphism. Consequently, all three problems share the same complexity class.

The final part of the paper addresses automorphism problems. For both linear and graphic matroids, the authors prove that the automorphism‑group membership problem (given a permutation, decide whether it is an automorphism) is polynomial‑time equivalent to the corresponding isomorphism problem. The key insight is that checking whether a permutation preserves the matroid structure can be reduced to checking whether it preserves the associated matrix or cycle‑space representation, which in turn can be decided by invoking a GI oracle. Moreover, they present an explicit polynomial‑time algorithm for testing membership in the automorphism group of a graphic matroid: compute a basis of the cycle space, apply the permutation to the edge set, and verify that the transformed basis still spans the same space. This algorithm runs in time polynomial in the size of the underlying graph.

In summary, the paper maps out the landscape of matroid isomorphism complexity. It locates the general problem in Σ₂^p, shows coNP‑hardness for linear matroids, and demonstrates that when the rank is bounded the problem is exactly as hard as Graph Isomorphism. It provides concrete reductions between graphic matroids, bounded‑rank matroids, and graphs, yielding deterministic polynomial‑time algorithms for planar cases. Finally, it establishes that automorphism‑group problems for these matroids are computationally equivalent to their isomorphism counterparts, and supplies a practical membership‑testing algorithm for graphic‑matroid automorphisms. These results deepen the connection between combinatorial optimization structures and classic graph‑theoretic complexity, and they open avenues for further exploration of matroid‑based problems within the broader context of the polynomial hierarchy.