Invariant random subgroups of linear groups

Research output: Contribution to journalArticlepeer-review

6 Scopus citations


Let Γ < GLn(F) be a countable non-amenable linear group with a simple, center free Zariski closure. Let Sub(Γ) denote the space of all subgroups of Γ with the compact, metric, Chabauty topology. An invariant random subgroup (IRS) of Γ is a conjugation invariant Borel probability measure on Sub(Γ). An IRS is called non-trivial if it does not have an atom in the trivial group, i.e. if it is non-trivial almost surely. We denote by IRS0(Γ) the collection of all non-trivial IRS on Γ. Theorem 0.1: With the above notation, there exists a free subgroupF < Γ and a non-discrete group topology on Γ such that for everyμ ∈ IRS0(Γ) the following properties hold:μ-almost every subgroup of Γ is openF ·Δ = Γ for μ-almost every Δ ∈ Sub(Γ).F ∩ Δ is infinitely generated, for every open subgroup. In particular, this holds for μ-almost every Δ ∈ Sub(Γ).The map Φ: (Sub(Γ), μ) → (Sub(F),Φ*μ) Δ → Δ ∩ Fis an F-invariant isomorphism of probability spaces. A more technical version of this theorem is valid for general countable linear groups. We say that an action of Γ on a probability space, by measure preserving transformations, is almost surely non-free (ASNF) if almost all point stabilizers are non-trivial. Corollary 0.2: Let Γ be as in the Theorem above. Then the product of finitely many ASNF Γ-spaces, with the diagonal Γ action, is ASNF. Corollary 0.3: Let Γ < GLn(F) be a countable linear group, A Δ Γ the maximal normal amenable subgroup of Γ — its amenable radical. If μ ∈ IRS(Γ) is supported on amenable subgroups of Γ, then in fact it is supported on Sub(A). In particular, if A(Γ) = <e> then Δ = <e>, μ almost surely.

Original languageEnglish
Pages (from-to)215-270
Number of pages56
JournalIsrael Journal of Mathematics
Issue number1
StatePublished - 1 Apr 2017

ASJC Scopus subject areas

  • Mathematics (all)


Dive into the research topics of 'Invariant random subgroups of linear groups'. Together they form a unique fingerprint.

Cite this