Statistics of bad parts of class groups
Peter Koymans, Yuan Liu
TL;DR
This work develops a statistical genus theory for the $p$-parts of class groups in abelian $p$-extensions. By translating class-group data into generalized Selmer groups with stringent local conditions, the authors prove that the dual Selmer group vanishes for 100% of extensions when ordered by the product of ramified primes, yielding an exact algebraic formula for the $I$-rank of $e\mathrm{Cl}(K)$ in terms of a combinatorial set of special primes and a cohomology-dependent constant. The main technical engine combines local condition analysis, Greenberg–Wiles, and a detailed character-sum and large-sieve framework, together with a thorough combinatorial classification of unlinked index sets. They also provide explicit corollaries for cyclic and elementary abelian $p$-groups, highlighting sharp connections between ramification data and the growth of $p$-parts of class groups. Overall, the results extend genus theory to a broad, probabilistic setting and furnish deterministic leading terms in a statistical context, with potential applications to broader arithmetic-statistics problems.
Abstract
Let $p$ be an odd prime. We give a formula for the bad part of $p$-class groups that is valid for $100\%$ of the abelian $p$-extensions when ordered by product of ramified primes.
