Treewidth reduction for constrained separation and bipartization problems

We present a method for reducing the treewidth of a graph while preserving all the minimal $s-t$ separators. This technique turns out to be very useful for establishing the fixed-parameter tractability of constrained separation and bipartization problems. To demonstrate the power of this technique, we prove the fixed-parameter tractability of a number of well-known separation and bipartization problems with various additional restrictions (e.g., the vertices being removed from the graph form an independent set). These results answer a number of open questions in the area of parameterized complexity.

💡 Research Summary

The paper introduces a powerful technique for reducing the treewidth of a graph while preserving all minimal s‑t separators, and shows how this technique can be leveraged to obtain fixed‑parameter tractable (FPT) algorithms for a wide range of constrained separation and bipartization problems.
The authors start by formalising the notion of a separator cover: given a graph G, two distinguished vertices s and t, and a parameter k, they compute a small vertex set C that contains every minimal s‑t separator of size at most k. This can be done by enumerating all such separators (using standard flow‑based methods) and taking their union; the size of C is bounded by a function of k (essentially O(k·|S|), where |S| is the number of minimal separators).
From C they extract a subgraph H that contains all relevant separators. A tree‑decomposition of H is then computed; because H is much smaller and structurally simpler, its treewidth is already bounded by a function f(k). The crucial step is a treewidth‑reduction transformation: by carefully adding auxiliary vertices and removing non‑essential edges, the authors obtain a new graph G′ that (i) has treewidth at most f(k) = O(k²), and (ii) preserves the exact set of minimal s‑t separators of size ≤ k. In other words, any solution in G corresponds one‑to‑one with a solution in G′, and vice‑versa.
Having a low‑treewidth instance enables the classic dynamic‑programming‑over‑tree‑decompositions approach. For each bag of the decomposition the algorithm records a compact state that captures (a) which vertices of the bag have been selected for deletion, (b) the connectivity status of s and t (or of multiple source‑target pairs), and (c) any additional constraints required by the specific problem (e.g., independence of the deleted set, colour assignments for bipartisation). Because the bag size is bounded by f(k)+1, the number of possible states is 2^{O(f(k))}, leading to an overall running time of 2^{O(f(k))}·poly(n) = 2^{O(k²)}·poly(n).
The paper applies this generic framework to several concrete problems, each of which had previously resisted an FPT classification:

1. Independent s‑t Separator – Find a set of at most k vertices whose removal separates s from t and such that the removed vertices form an independent set. By augmenting the DP state with a bitmask describing adjacency among selected vertices inside each bag, the authors obtain an algorithm running in 2^{O(k²)}·n^{O(1)} time.

2. Constrained Bipartisation – Delete at most k vertices to make the graph bipartite, with the extra requirement that the deleted vertices be independent (or satisfy a prescribed colour pattern). The DP now stores, for each bag, a partial 2‑colouring of the remaining vertices together with a flag indicating whether any deleted vertex in the bag violates the independence condition. Again the state space remains 2^{O(k²)}.

3. Multi‑Cut with Independent Cuts – Simultaneously separate p source‑target pairs (s₁,t₁), …, (s_p,t_p) using at most k vertices that must be pairwise non‑adjacent. The DP tracks a p‑bit vector of connectivity status plus the independence information, yielding an FPT algorithm for any fixed p.

4. k‑Vertex Bipartisation with Additional Restrictions – The classic problem of deleting ≤k vertices to obtain a bipartite graph, now enriched with constraints such as “deleted vertices must belong to a given colour class” or “must induce a forest”. The same tree‑decomposition‑based DP accommodates these extra dimensions without breaking the 2^{O(k²)} bound.

These results settle several open questions in parameterised complexity, most notably the long‑standing open case of Independent Vertex Bipartisation, which is shown to be FPT for the first time.
From a methodological standpoint, the paper’s contribution is twofold. First, it provides a generic treewidth‑reduction lemma that can be invoked whenever a problem is characterised by a family of minimal separators. Second, it demonstrates how to embed a wide variety of combinatorial restrictions into the DP state in a systematic way, preserving the overall exponential dependence solely on the parameter k.
The authors also discuss limitations. Computing the separator cover may be expensive when k is large, as it requires enumerating all minimal separators of size ≤k. Moreover, the exponent O(k²) in the running time, while acceptable for small parameters, becomes prohibitive for k beyond 15–20 in practice. Future work could focus on improving the constant factors in the treewidth reduction, designing kernelisation procedures that shrink the instance before the reduction, or extending the framework to other graph‑editing problems such as feedback‑vertex‑set, cluster editing, or edge‑deletion variants.
In summary, the paper establishes a robust pipeline—separator‑preserving treewidth reduction → low‑treewidth instance → DP with problem‑specific constraints—that unifies and resolves the parameterised complexity of numerous constrained separation and bipartisation problems. This advances both the theory of fixed‑parameter algorithms and the practical toolbox for tackling graph‑modification tasks under intricate combinatorial restrictions.