Preserving Distances in Faulty Colored Graphs

We study color fault-tolerant (CFT) network design problems: Given an $n$-vertex graph $G$ whose edges are arbitrarily colored (with no ``legality'' restrictions), the goal is to find a sparse subgraph $H$ such that, when any color fault causes all edges of some color $c$ to crash, the surviving subgraph $H-c$ remains ``similar'' to the surviving graph $G-c$. The similarity is problem-dependent, usually pertaining to distance preserving. If each color class has size $Δ$ or less, a brute-force approach can disregard the colors and take $H$ to be $Δ$-edge fault-tolerant ($Δ$-EFT), so that $H-F$ is similar to $G-F$ for every set $F$ of $\leq Δ$ edges. We ask if the colors can be utilized to provide a sparser $H$. Our main results concern CFT sourcewise distance preservers, where there is a given set $S \subseteq V$ of $σ$ sources, and all $S \times V$ distances should be exactly equal in $H-c$ and in $G-c$. We give nearly-tight upper and lower bounds of $\tildeΘ (n^{2-1/(Δ+1)} \cdot σ^{1/(Δ+1)})$ on the worst-case size of such preservers. The corresponding $Δ$-EFT problem admits the same lower bound, but the state-of-the-art upper bound for $Δ\geq 3$ is $\tilde{O}(n^{2-1/2^Δ} \cdot σ^{1/2^Δ})$. Our approach also leads to new and arguably simpler constructions that recover these $Δ$-EFT bounds and shed some light on their current gaps. We provide additional results along these lines, showcasing problems where the color structure helps or does not help sparsification. For preserving the distance between a single pair of vertices after a color fault, the brute-force approach via $Δ$-EFT is shown to be suboptimal. In contrast, for preserving reachability from a single source in a directed graph, it is (worst-case) optimal.