The paper studies a question in metric geometry: when can the "distance structure" of certain symmetric grid-like surfaces be faithfully represented in a particularly simple mathematical space? The space in question, called l1-space, is essentially a higher-dimensional version of city-block distance (think of navigating a grid of streets where you can only move horizontally or vertically). A graph or network is called l1-embeddable if you can assign its vertices to points in such a space so that every shortest-path distance in the graph is exactly preserved. This is a meaningful property because l1-embeddable graphs have a clean combinatorial structure and connect to coding theory and optimization.
The surfaces being studied are the torus (the shape of a donut) and the Klein bottle (a surface like a Mobius strip but closed up, which cannot be physically constructed in three dimensions without self-intersection). On each of these surfaces, the paper considers tilings by squares, where every vertex has the same number of edges meeting it. These tilings come in several distinct families depending on how a flat square grid is folded and glued along its edges to form the surface. Each family is indexed by a few integer parameters that control the size and the twist of the gluing. The question is: for which choices of those parameters does the resulting tiled surface have a distance structure that fits into l1-space?
The main result is a complete answer to this question. Out of the five families of such tilings (two on the torus and three on the Klein bottle), the paper precisely identifies which parameter choices lead to l1-embeddable maps and which do not. The embeddable cases are relatively rare and often correspond to small or specially structured examples. For all other parameter choices, the authors prove that no faithful l1-embedding exists. The classification is achieved by a combination of tools: known algebraic relationships between the families, direct construction of embeddings for the embeddable cases, and obstruction arguments (based on structures that are known to be incompatible with l1-embeddability) for the rest.