A polynomial-time algorithm for recognizing high-bandwidth graphs

An unweighted, undirected graph $G$ on $n$ nodes is said to have \emph{bandwidth} at most $k$ if its nodes can be labelled from $0$ to $n - 1$ such that no two adjacent nodes have labels that differ by more than $k$. It is known that one can decide whether the bandwidth of $G$ is at most $k$ in $O(n^k)$ time and $O(n^k)$ space using dynamic programming techniques. For small $k$ close to $0$, this approach is effectively polynomial, but as $k$ scales with $n$, it becomes superexponential, requiring up to $O(n^{n - 1})$ time (where $n - 1$ is the maximum possible bandwidth). In this paper, we reformulate the problem in terms of bipartite matching for sufficiently large $k \ge \lfloor (n - 1)/2 \rfloor$, allowing us to use Hall’s marriage theorem to develop an algorithm that runs in $O(n^{n - k + 1})$ time and $O(n)$ auxiliary space (beyond storage of the input graph). This yields polynomial complexity for large $k$ close to $n - 1$, demonstrating that the bandwidth recognition problem is solvable in polynomial time whenever either $k$ or $n - k$ remains small.

💡 Research Summary

The paper addresses the long‑standing computational challenge of recognizing whether an undirected, unweighted graph G on n vertices has bandwidth at most k, especially when k is large (close to the maximum possible bandwidth n‑1). Classical dynamic‑programming approaches run in O(n^k) time and space, which is polynomial only for very small k; as k grows with n the runtime becomes super‑exponential, up to O(n^{n‑1}) in the worst case.

The authors observe that when k ≥ ⌊(n‑1)/2⌋ the layout can be split into a left “partial layout” L occupying positions 0,…,n‑k‑2 and a right “partial layout” R occupying positions k+1,…,n‑1. Because the two position intervals are disjoint, any full layout’s feasibility depends solely on which vertices are placed in L and which in R; edges whose endpoints lie entirely within one side automatically satisfy the bandwidth bound, while only edges crossing the two sides can potentially violate it. Lemma 3.1 formalizes this observation, showing that a layout respects the bandwidth bound iff no edge connects a vertex in L(i) with a vertex in R(k+j+1) for any i ≤ j.

Given a fixed left partial layout L, the authors define for each j a candidate set A_j consisting of vertices not used by L that are non‑adjacent to all L(i) for i ≤ j. The existence of a compatible right partial layout R is then equivalent to finding a system of distinct representatives for the family {A_j}. By Hall’s marriage theorem, such a system exists iff |A_j| ≥ n‑k‑j‑1 for every j. This condition can be checked in linear time with respect to n, and if it holds, a feasible R can be constructed greedily in reverse order (Corollary 3.3).

The algorithm therefore enumerates all possible left partial layouts L (there are O(n^{,n‑k‑1}) of them), and for each L performs the O(n) Hall‑condition test. If the test succeeds, the algorithm immediately builds R and returns a full layout, terminating early. The total worst‑case running time is O(n^{,n‑k+1}), and the auxiliary space required is O(n) (to store L, the A_j sets, and temporary data).

The paper also introduces two easily computable lower bounds on bandwidth, α(G) and γ(G), based on k‑hop neighborhoods. Both can be evaluated in O(m n) time and O(n) space and serve as quick pre‑checks to discard instances that are provably out of range.

Experimental results compare the new method against the classic O(n^k) DP algorithm and recent bandwidth‑minimization heuristics that internally invoke recognition subroutines. For large k (e.g., k ≥ 0.8·(n‑1)), the proposed algorithm dramatically outperforms the DP approach, achieving polynomial‑time behavior when n‑k is constant or small. When k is small, the DP method remains preferable, confirming that the two regimes complement each other.

In conclusion, the authors demonstrate that bandwidth recognition is polynomial‑time solvable whenever either k or n‑k is bounded by a constant, filling a gap in the literature that previously focused almost exclusively on low‑bandwidth cases. They suggest future work on hybrid strategies for intermediate k, randomized pruning, and extensions of the Hall‑matching framework to related layout problems such as bandwidth approximation and graph drawing.