Node-Disjoint Multipath Spanners and their Relationship with Fault-Tolerant Spanners

Motivated by multipath routing, we introduce a multi-connected variant of spanners. For that purpose we introduce the $p$-multipath cost between two nodes $u$ and $v$ as the minimum weight of a collection of $p$ internally vertex-disjoint paths between $u$ and $v$. Given a weighted graph $G$, a subgraph $H$ is a $p$-multipath $s$-spanner if for all $u,v$, the $p$-multipath cost between $u$ and $v$ in $H$ is at most $s$ times the $p$-multipath cost in $G$. The $s$ factor is called the stretch. Building upon recent results on fault-tolerant spanners, we show how to build $p$-multipath spanners of constant stretch and of $\tO(n^{1+1/k})$ edges, for fixed parameters $p$ and $k$, $n$ being the number of nodes of the graph. Such spanners can be constructed by a distributed algorithm running in $O(k)$ rounds. Additionally, we give an improved construction for the case $p=k=2$. Our spanner $H$ has $O(n^{3/2})$ edges and the $p$-multipath cost in $H$ between any two node is at most twice the corresponding one in $G$ plus $O(W)$, $W$ being the maximum edge weight.

💡 Research Summary

The paper introduces a new class of graph spanners tailored to the needs of multipath routing, called p‑multipath spanners. For a weighted undirected graph G, the authors define the p‑multipath cost δₚᴳ(u,v) as the minimum total weight of a collection of p internally vertex‑disjoint paths connecting u and v. A subgraph H is a p‑multipath s‑spanner if for every pair of vertices the p‑multipath cost in H is at most s times the corresponding cost in G (or more generally α·δₚᴳ + β). This generalizes ordinary spanners, which correspond to the case p = 1.

The main contributions are twofold. First, for any fixed integers p > 1 and k > 1, the authors prove that every n‑vertex weighted graph admits a p‑multipath spanner with stretch
 s = k·p·O(1 + p/k)^{2k‑1}
and size ˜O(p²·n^{1+1/k}). Moreover, such a spanner can be built in the LOCAL distributed model in O(k) synchronous rounds. The construction proceeds by combining three known ingredients:

1. b‑hop spanners – a subgraph in which every original edge can be replaced by a path of at most b edges whose total weight is at most s times the edge weight. The authors show that a (2k‑1)‑hop spanner with O(n^{1+1/k}) edges exists and can be obtained via a variant of Kruskal’s algorithm or via the random‑sampling technique of Thorup–Zwick (as refined by Chechik et al.). This step yields a sparse structure that respects a bounded hop‑count.

2. Fault‑tolerant spanners – using the generic algorithm of Dinitz and Krauthgamer (DK11), which repeatedly runs any s‑spanner algorithm on random vertex‑deletions, one obtains an r‑fault‑tolerant s‑spanner of size O(r³·m(2n/r)·log n) where m(·) is the size bound of the underlying s‑spanner. By plugging the (2k‑1)‑hop spanner algorithm into DK11, the resulting fault‑tolerant spanner is also a (r + 1)‑multipath spanner, because any edge not present in the spanner can be replaced by a set of r + 1 internally disjoint paths (by Menger’s theorem) whose total weight is bounded by the same stretch factor.

3. Parameter mapping – setting r = p − 1 and choosing k appropriately yields the claimed p‑multipath spanner with the stated stretch and size. The crucial observation is that the number of communication rounds does not depend on p or n; it is bounded by the hop‑parameter k of the underlying hop‑spanner.

The second major result focuses on the special case p = k = 2. The generic construction would give a 2‑multipath spanner with stretch 18 and O(n^{3/2} log^{3/2} n) edges. The authors improve this dramatically by designing a dedicated algorithm that achieves stretch 2 plus an additive term O(W), where W is the maximum edge weight, while keeping the edge count at O(n^{3/2}). The key idea is to select a set of “pivot” vertices at random; each vertex connects to its nearest pivot via a shortest‑path tree of depth at most two. For any pair (u,v), the 2‑multipath in the spanner consists of the path from u to its pivot, the (possibly trivial) edge between the two pivots, and the path from the second pivot to v. Because the metric δ₂ does not satisfy the triangle inequality, the authors carefully analyze how the additive term O(W) arises from the possible missing direct edge between the two pivots. Nevertheless, the construction guarantees that for every pair the total cost is at most twice the optimal 2‑multipath cost plus O(W).

The paper also discusses why δₚ for p > 1 lacks the triangle inequality, which makes traditional spanner techniques insufficient. By introducing the notion of b‑hop spanners and leveraging fault‑tolerant constructions, the authors bridge this gap and provide a systematic method to obtain sparse multipath spanners. They further show that any r‑fault‑tolerant s‑spanner that is also b‑hop automatically yields a (r + 1)‑multipath spanner with stretch bounded by a function of b, r, and s.

In summary, the work makes three significant contributions:

• Theoretical existence: It proves that constant‑stretch p‑multipath spanners of near‑optimal size exist for any fixed p and any trade‑off parameter k.
• Distributed construction: It presents a simple O(k)‑round LOCAL algorithm that builds such spanners, improving over previous fault‑tolerant constructions that required Ω(p³ log n) rounds.
• Improved special case: For p = k = 2 it offers a concrete O(n^{3/2})‑edge spanner with stretch 2 + O(W), which is close to optimal for many practical networks.

These results open the door to compact, fault‑resilient multipath routing tables and have potential applications in bandwidth aggregation, load balancing, and rapid recovery from link failures in large‑scale communication networks.