Sharp cluster-size decay and logarithmic corrections at phase boundaries in kernel-based spatial random graphs
For supercritical kernel-based spatial random graphs with Pareto marks of tail exponent b = τ − 1, we determine the sharp decay of the rooted finite-cluster tail and the order of the second-largest component whenever long edges strictly dominate surface order. On Poisson and lattice vertex sets we close the gap left by Jorritsma, Komjáthy and Mitsche in the high–high regime: for all b < σ, their upper exponent (σ + 1 − b/α)^(−1) is improved to the sharp exponent ζ_hh = (σ + 1 − b)/(σ + 1 − b/α). Consequently, all three exponent assertions of their conjecture, for the rooted tail, the second-largest component and the lower tail of the giant (the last on Poisson vertex sets), hold whenever the high–high exponent is positive and maximal. We also find the sharp laws at all four maximal long-edge ties above surface order, including a fractional power (log k)^(b/α) at the high–low/high–high tie and a squared logarithm at the triple tie. We construct local seeds at every strictly supercritical weighted-lattice parameter with a positive long-edge exponent, including retention one and dimension one. The main tool is a conditional exterior transfer: once a connected backbone is built, every other vertex receives an independent attachment test whose joint law with the graph is restored exactly, so components avoiding the backbone need no geometric control.
Authors
- Shardul Mishra (ORCID: https://orcid.org/0009-0005-4164-7569)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-15
- DOI
- https://doi.org/10.5281/zenodo.24247017
- Primary Topic
- Stochastic processes and statistical mechanics
- Type
- preprint