Graphons, exchangeability, and relational limit theory, formalized in Lean 4 on top of Mathlib.
Core graphon program complete · zero placeholders · standard axioms only
Research programs
Graphon analysis. Graphons as symmetric measurable functions on a probability space; the cut norm and cut distance; step approximations and Frieze–Kannan regularity; the counting and inverse counting lemmas; compactness of the graphon space. Underneath sits an atomless standard-Borel measure-isomorphism theorem and the overlay theorem, which together replace the coupling step that classical treatments leave implicit.
Diaconis–Janson and Aldous–Hoover. Exchangeable random graphs, graphon mixtures, and the representation theorem identifying the two; the infinite-law correspondence; empirical graphons and their almost-sure convergence; the extremality equivalences characterizing dissociated laws.
Generic relational exchangeability. A multi-sorted relational signature framework: carriers, topology, exchangeable laws, ergodicity, and equality patterns, developed for arbitrary arity rather than graphs alone — the setting of the Aldous–Hoover–Kallenberg theorem.
Digraphons. Directed graph limits: the five-component digraphon with its four reciprocal-edge pair kernels, the directed sampler, and the special families (graphon embeddings, tournaments, asymmetric kernels).
Landmark results
cutDistance_triangle— the cut distance is a pseudometric, via four corrected Rokhlin-style coupling cores.regularity— Frieze–Kannan weak regularity: bounded-complexity step approximation of every graphon.cutDistance_tendsto_iff_homDensity_tendsto— cut-distance convergence is exactly convergence of all homomorphism densities.first_sampling_lemma— a single sample size works for every graphon simultaneously.samplePMF_eq_all_iff_weaklyIsomorphic— the sample laws determine the graphon.graphon_mixture_representation— the Diaconis–Janson theorem: exchangeable graph laws are graphon mixtures, uniquely.tfae_ergodic_extremality— a six-way equivalence: dissociated, restriction independent, tail trivial, ergodic, extreme, Dirac-represented.InfiniteRelExchangeableLaw.condIndep_fixingAlgebra— for every exchangeable relational law, the fixing σ-algebras of two finite vertex sets are conditionally independent given their intersection.ofTournament_sample_isTournament— the tournament digraphon almost surely samples a tournament: no loops, and exactly one direction between any two distinct vertices.
Current frontier
The classical graphon program and the graph-level Diaconis–Janson / Aldous–Hoover theory are complete. The generic functional Aldous–Hoover–Kallenberg theorem for arbitrary relational signatures is in progress: the forward direction and the conditional independence underpinning the converse are proved, while the coherent factor realization and the kernel recursion are still being built. The open issues track this work.
Explore the formalization
- Blueprint (web) — statements, dependencies, and the proof narrative
- Blueprint (pdf)
- API documentation — the complete module and declaration inventory
- Dependency graph
- Repository · Verification status and history
References
- Lovász, L. (2012). Large Networks and Graph Limits. AMS Colloquium Publications, vol. 60.
- Frieze, A. & Kannan, R. (1999). Quick approximation to matrices and applications. Combinatorica, 19(2), 175–220.
- Borgs, C., Chayes, J. T., Lovász, L., Sós, V. T., & Vesztergombi, K. (2008). Convergent sequences of dense graphs I. Advances in Mathematics, 219(6), 1801–1851.
- Diaconis, P. & Janson, S. (2008). Graph limits and exchangeable random graphs. Rendiconti di Matematica, 28, 33–61.
- Kallenberg, O. (2005). Probabilistic Symmetries and Invariance Principles. Springer.
- Austin, T. (2008). On exchangeable random variables and the statistics of large graphs and hypergraphs. Probability Surveys, 5, 80–145.