On dense totipotent free subgroups in full groups (Q6182525)

The setting considered here is a Borel action \(\beta\) on a standard Borel space \(X\) of a countable discrete group \(\Gamma\) and the associated stabilizer map \(\mathrm{Stab}^{\beta}\) from \(X\) to the space \(\mathrm{Sub}(\Gamma)\) of subgroups of \(\Gamma\) with the compact totally disconnected topology of pointwise convergence, which intertwines the action with the continuous conjugation action. If the \(\Gamma\) action on \(X\) preserves a probability measure \(\mu\) then \(\mathrm{Stab}^{\beta}_{*}\mu\) is a measure on \(\mathrm{Sub}(\Gamma)\) that is invariant under the conjugation action, giving an example of an IRS (invariant random subgroup). The intricate story of how properties of the original action together with an invariant measure relate to properties of the stabilizer map and push-forwards of measures begins with the observation that if \(\beta\) is totally free (\(\mu\) has no atoms and the stabilizer map is essentially injective) then the support of \(\mathrm{Stab}^{\beta}_{*}\mu\) is a perfect set. The main notion studied here comes via the perfect kernel of \(\mathrm{Sub}(\Gamma)\), defined as the largest closed subspace with no isolated points. Then the action \(\beta\) with invariant probability measure \(\mu\) is said to be `totipotent' when the support of the IRS given by \(\mathrm{Stab}^{\beta}_{*}\mu\) is the perfect kernel. The etymology of the word is clear, and it has been borrowed in this context from the study of stem cells in biology. Under the hypothesis of ergodicity this property implies that almost every element of the IRS has dense orbit in the perfect kernel. The action \(\beta\) on \((X,\mu)\) has an associated orbit equivalence relation, full group with induced Polish group structure, and cost (a numerical invariant associated to the orbit equivalence relation). The significance of cost may be seen in the important result of the second author [Invent. Math. 139, No. 1, 41--98 (2000; Zbl 0939.28012)] showing that the cost of \(\beta\) is \(r\) if \(\Gamma\) is the free group on \(r\) generators and \(\beta\) is free and is strictly less than \(r\) if \(\beta\) is non-free. The main result here is a strengthening of work of \textit{F. Le Maître} [Invent. Math. 198, No. 2, 261--268 (2014; Zbl 1338.37006)] on further sharpening of the consequences of the cost being strictly smaller than \(r\): Given an ergodic probability measure-preserving equivalence relation whose cost is strictly smaller than \(r\) for some \(r\geqslant2\), there exists a homomorphism from the free group on \(r\) generators to the full group of the equivalence relation with dense image and totipotent associated action on the free group. The result is sharp in the sense that if the cost was at least \(r\) then it cannot come from a non-free action of the free group on \(r\) generators. Among the consequences is a continuum of pairwise distinct ergodic totipotent IRSs of the free group on \(r\) generators with non-isomorphic associated equivalence relations.