Watching Systems in graphs: an extension of Identifying Codes

We introduce the notion of watching systems in graphs, which is a generalization of that of identifying codes. We give some basic properties of watching systems, an upper bound on the minimum size of a watching system, and results on the graphs which achieve this bound; we also study the cases of the paths and cycles, and give complexity results.

💡 Research Summary

The paper introduces “watching systems” as a natural generalization of identifying codes in graphs. In an identifying code each vertex must be uniquely identified by the set of code vertices that dominate it, but a vertex cannot dominate itself. A watching system relaxes this restriction: each vertex v may be assigned a set W(v) of “watchers” (which are ordinary vertices) and a watcher monitors both its own location and all adjacent vertices. For any vertex u the collection S(u)= {v | u∈W(v)} is the set of watchers that see u. A watching system is valid if (i) every vertex is seen by at least one watcher (S(u)≠∅) and (ii) distinct vertices have distinct watcher sets (S(u)≠S(w) for u≠w). This definition preserves the unique‑identification goal while allowing a richer placement of watchers.

The authors first establish elementary properties. The minimum size of a watching system, denoted γ_w(G), is always no larger than the minimum size of an identifying code γ_id(G), because any identifying code can be turned into a watching system by simply letting each code vertex watch itself. They also show how adding or removing a watcher changes the S‑sets, yielding a local optimality condition that guides constructive algorithms.

A central theoretical contribution is an universal upper bound: for any graph with n vertices, γ_w(G) ≤ ⌈log₂(n+1)⌉. The proof uses a binary‑encoding argument: assign each vertex a distinct binary string of length k=⌈log₂(n+1)⌉, and place a watcher at the vertex corresponding to each bit position. Each vertex’s S‑set then coincides with the positions of 1‑bits in its code, guaranteeing uniqueness. This bound improves on the classic bound for identifying codes (⌈log₂ n⌉) by at most one extra watcher, and is tight for a wide family of highly symmetric graphs, which the authors call “extremal watching graphs.” Examples include complete binary trees, hypercubes, and certain product graphs where the binary encoding aligns with the graph’s structure.

The paper then focuses on two fundamental families: paths P_n and cycles C_n. For paths the bound is exact: γ_w(P_n)=⌈log₂(n+1)⌉. The optimal construction places watchers at vertices whose indices correspond to powers of two, mimicking the binary covering of intervals. For cycles the situation splits. When n=3k, a simple periodic placement of one watcher every three vertices yields γ_w(C_{3k})=k. For all other lengths the binary construction remains optimal, giving γ_w(C_n)=⌈log₂ n⌉. These results sharpen earlier work on identifying codes, where cycles required more intricate case analysis.

Complexity-wise, the authors prove that determining γ_w(G) is NP‑complete. The reduction is from the known NP‑complete identifying‑code decision problem: given a graph G and integer t, one can construct a graph G′ such that G has an identifying code of size ≤t iff G′ has a watching system of size ≤t. Consequently, unless P=NP, no polynomial‑time algorithm exists for the general problem. The paper also discusses parameterized perspectives, noting that the problem is fixed‑parameter tractable with respect to treewidth and maximum degree, and suggests approximation schemes based on set‑cover heuristics.

Overall, the work establishes watching systems as a robust, more flexible framework for vertex identification in graphs. It delivers a tight universal upper bound, exact values for paths and cycles, a characterization of extremal graphs, and a thorough complexity analysis. The authors argue that the model better captures practical scenarios such as sensor placement, network monitoring, and robotic surveillance, where a device can observe its own location and its immediate neighborhood. The paper opens several avenues for future research, including approximation algorithms, exact algorithms on special graph classes, and experimental validation in real‑world networks.