Intensity Dot Product Graphs

Researchers Giulio Valentino Dalla Riva and Matteo Dalla Riva have published a new paper on arXiv introducing Intensity Dot Product Graphs (IDPGs), a significant evolution in statistical network modeling. The work addresses limitations in existing approaches like Random Dot Product Graphs (RDPGs) and graphon models. While RDPGs treat node sets as fixed, IDPGs replace this fixed collection with a Poisson point process on a Euclidean latent space. This fundamental shift yields a model with random node populations while maintaining the geometric interpretability and dot-product affinity structure of RDPGs. The model is parameterized by a population-level 'intensity' that formally links continuous latent structure to the finite graphs we observe in practice.

The technical contributions are substantial. The authors define new continuous analogues of key discrete objects, such as the 'heat map' and 'desire operator' as counterparts to the probability matrix. They prove a spectral consistency result, connecting the singular values of a graph's adjacency matrix to the spectrum of an underlying operator, which is crucial for statistical inference. The framework naturally accommodates temporal extensions, as the evolving intensity can be modeled through partial differential equations (PDEs), opening the door to dynamic network analysis. The paper also rigorously situates IDPGs within the broader landscape, showing how they relate to graphon theory and how classical RDPGs emerge as a special, concentrated limit of the new model.