The paper studies a pursuit game played on a graph (a network of nodes and edges) between a cop and a robber. In the standard version of this "Cops and Robbers" game, the cop tries to catch the robber. In the "damage" variant studied here, the robber instead tries to visit as many distinct nodes as possible before being caught, while the cop tries to minimize that number. The count of nodes the robber visits under perfect play by both sides is called the damage number. The new twist introduced in this paper is the "zombie damage number": the cop must always move directly toward the robber along the shortest available path, like a mindless zombie that cannot be strategic or devious. The question is how much extra damage the robber can cause by exploiting this rigid, predictable pursuit.
The authors prove several concrete results. On trees (graphs with no cycles), the zombie cop does just as well as a fully strategic cop, meaning the rigid pursuit rule costs nothing extra. For paths and cycles the results are clean: on a cycle of five or more nodes, the robber can visit every single node before being caught. For complete multipartite graphs (a broad family of highly connected graphs), the zombie damage number is worked out exactly. The authors also show that for graphs which are sparse in a specific sense, having minimum connectivity at least two and no short cycles, every node ends up being visited, meaning the zombie cop is completely ineffective at limiting damage. This gives a sharp lower bound on zombie damage tied to the length of the shortest cycle in the graph.
A notable aspect of the paper is how the results were found. The authors used a tool they call "Theo-Conjecture," a discovery loop combining automated conjecture software, exact computational checks, a large language model for exploration, and human mathematical judgment to guide the research. This hybrid human-AI process generated many of the conjectures that the humans then proved or refined. The paper is therefore also a case study in how AI-assisted mathematical discovery can work in practice, pointing toward a research methodology where machines help generate ideas and humans provide the rigorous proofs.