The paper investigates a question about networks (graphs): given a desired set of statistics describing how often certain small patterns appear, can you build an actual finite network that hits those statistics exactly? The patterns in question involve three nodes at a time, such as a triangle, a path connecting three nodes, or a pair of connected nodes with a third isolated node. Researchers have long studied which combinations of these pattern frequencies are theoretically achievable, either approximately (by large networks that get closer and closer to the target) or exactly (by a single concrete network of a specified size). This paper bridges those two perspectives.
The central result is that approximate achievability essentially guarantees exact achievability, at least away from the boundary cases. More precisely, the authors show that if a target combination of pattern frequencies is achievable in the approximate, large-network limit, and if it sits a safe distance away from the edge of the allowed region, then for every sufficiently large network size there exists a concrete network of exactly that size realizing those statistics precisely. The only exceptions are unavoidable technical arithmetic constraints, for instance that certain counts must be whole numbers.
The technical approach works by starting with a theoretical limiting object called a graphon, which encodes the approximate target, sampling a large finite network from it, and then carefully making small local adjustments to the network to correct any discrepancy between the sampled statistics and the exact targets. The authors show these adjustments can always be made without pushing the statistics outside the desired range, as long as you started far enough from the boundary. The upshot is a clean and general correspondence between the continuous world of asymptotic graph theory and the discrete world of finite combinatorics for this family of subgraph statistics.