Graphons are mathematical objects used to describe the structure of large networks. Think of a graphon as a function that assigns a probability to each pair of potential connections in a network, capturing the overall "shape" of how a large graph is organized. Random graphs generated by graphons are widely used to model real-world networks, and understanding their entropy, a measure of how much randomness or uncertainty is contained in these networks, is an important theoretical question.
The paper focuses on a specific class of graphons where the probability function is "piecewise Holder continuous," meaning it is relatively smooth in pieces but can have sharp jumps between those pieces. The main technical contribution is showing precisely how fast the entropy of a random graph, normalized by the number of nodes, converges to a limiting value as the graph grows larger. Rather than just proving that convergence happens, the authors work out explicit formulas describing how quickly it occurs.
Using this general result, the authors then derive concrete, usable entropy bounds for two popular and well-studied network models: the stochastic block model, where nodes are divided into communities with connection probabilities depending on which communities two nodes belong to, and the random geometric graph, where nodes are placed randomly in space and connect based on their physical proximity. The key contribution here is that these bounds are explicit numerical formulas rather than vague asymptotic statements that only describe behavior in the limit of infinitely large graphs. This makes the results more practically useful for applications in information theory and network science.