Better Diameter Bounds for Efficient Shortcuts and a Structural Criterion for Constructiveness

All parallel algorithms for directed connectivity and shortest paths crucially rely on efficient shortcut constructions that add a linear number of transitive closure edges to a given DAG to reduce its diameter. A long sequence of works has studied both (efficient) shortcut constructions and impossibility results on the best diameter and therefore the best parallelism that can be achieved with this approach. This paper introduces a new conceptual and technical tool, called certified shortcuts, for this line of research in the form of a simple and natural structural criterion that holds for any shortcut constructed by an efficient (combinatorial) algorithm. It allows us to drastically simplify and strengthen existing impossibility results by proving that any near-linear-time shortcut-based algorithm cannot reduce a graph’s diameter below $n^{1/4-o(1)}$. This greatly improves over the $n^{2/9-o(1)}$ lower bound of [HXX25] and seems to be the best bound one can hope for with current techniques. Our structural criterion also precisely captures the constructiveness of all known shortcut constructions: we show that existing constructions satisfy the criterion if and only if they have known efficient algorithms. We believe our new criterion and perspective of looking for certified shortcuts can provide crucial guidance for designing efficient shortcut constructions in the future.

💡 Research Summary

The paper revisits the central role of shortcuts—small sets of transitive‑closure edges added to a directed acyclic graph (DAG) to shrink its diameter—in parallel algorithms for directed connectivity and shortest‑path problems. While prior work has largely focused on the existence of shortcuts and on lower bounds for the achievable diameter, it has not systematically addressed the constructibility of such shortcuts by near‑linear‑time algorithms.

To bridge this gap, the authors introduce the notion of certified shortcuts. A shortcut edge (u,v) is certified if there already exist two edges (u,w) and (w,v) in the original graph or in the set of previously added shortcuts. This exactly captures the iterative “T‑step” procedure used by all known shortcut‑based algorithms, where each iteration shortcuts a length‑2 path. Consequently, any shortcut that can be built by an efficient combinatorial algorithm must be a certified shortcut.

Using this structural restriction, the paper proves two stronger diameter lower bounds. First, for any family of DAGs with n vertices, any certified shortcut of size O(m) (where m is the original number of edges) must have diameter at least Ω(n^{1/4}). This improves the previous Ω(n^{2/9}) bound of HXX25 and implies that any near‑linear‑time shortcut‑based algorithm for directed reachability must have depth at least n^{1/4‑o(1)} (Theorem 1.1). Second, when the shortcut size is limited to O(n), the diameter lower bound rises to Ω(n^{1/3}), strengthening the earlier Ω(n^{1/4}) bound for O(n)‑size shortcuts.

The authors further formalize certification complexity, defined as the size of the smallest certified shortcut that contains a given shortcut set H. They show that all known efficient algorithms (pivot‑based, chain‑cover‑based, flow‑based, etc.) produce shortcuts with certification complexity ~O(m), whereas existing existential constructions (sampling‑based, greedy, non‑constructive chain covers) have certification complexity m^{1+Ω(1)} with high probability. This dichotomy provides the first unconditional evidence that many existential proofs cannot be turned into efficient algorithms without fundamentally new ideas.

The framework is also extended to hopsets, which are weighted analogues of shortcuts. By requiring the weight of a newly added hopset edge to equal the sum of the weights of its two certifying edges, the authors obtain matching lower bounds: any O(m)‑size certified hopset must have diameter Ω(n^{1/3}), and any O(n)‑size certified hopset must have diameter Ω(n^{1/2}). These results close the gap between weighted and unweighted hopset lower bounds that previously existed.

Overall, the paper delivers a unified structural criterion—certified shortcuts—that both captures all known efficient shortcut constructions and yields the strongest known diameter lower bounds for shortcut‑based parallel algorithms. It clarifies the limits of current techniques, introduces certification complexity as a useful measure of constructiveness, and suggests that any future breakthrough in parallel directed connectivity or shortest‑path algorithms will need to either bypass the certified‑shortcut paradigm or discover fundamentally new ways to reduce certification complexity.