OctoInformBeta

Isomorphism of the free group factors

For a discrete group GG, the group von Neumann algebra L(G)L(G) is the von Neumann algebra on ℓ2(G)\ell^2(G) generated by the left translations λ(g)\lambda(g), g∈Gg\in G, with the trace τ(x)=⟨xδe,δe⟩\tau(x)=\langle x\delta_e,\delta_e\rangle, where δe\delta_e is the basis vector at the identity ee. For 1≤n≤∞1\leq n\leq\infty, 𝔽n\mathbb{F}_n denotes the free group on nn generators, countably infinitely many when n=∞n=\infty. The free group factors are the algebras L(𝔽n)L(\mathbb{F}_n), 2≤n≤∞2\leq n\leq\infty. For a II1\mathrm{II}_1 factor QQ with trace τ\tau and a real t>0t>0, the amplification QtQ^t is the corner p(Q⊗‾B(ℓ2(ℕ)))pp(Q\mathbin{\bar{\otimes}}B(\ell^2(\mathbb{N})))p at a projection pp with (τ⊗Tr)(p)=t(\tau\otimes\mathrm{Tr})(p)=t, where Tr\mathrm{Tr} is the canonical trace on B(ℓ2(ℕ))B(\ell^2(\mathbb{N})). The trace of QtQ^t is t−1(τ⊗Tr)t^{-1}(\tau\otimes\mathrm{Tr}). The fundamental group of QQ is the set of t>0t>0 for which Qt≅QQ^t\cong Q. The free group factor isomorphism problem asks whether L(𝔽m)L(\mathbb{F}_m) and L(𝔽n)L(\mathbb{F}_n) are isomorphic for distinct m,n≥2m,n\geq2.

The free group factors arose in the foundational work of Murray and von Neumann. Their nonhyperfiniteness distinguished them from the hyperfinite factor. As the historical account of Rădulescu recalls, it did not distinguish the factors L(𝔽n)L(\mathbb{F}_n) for different nn. Dykema’s account of the early history credits Kadison with raising the isomorphism problem. Voiculescu developed the framework of reduced free products in which free probability studies these factors through the distributions of their generators. His results on corners of L(𝔽∞)L(\mathbb{F}_\infty) showed that its fundamental group contains every positive rational number. Rădulescu extended this to every positive real number. Dykema and Rădulescu independently constructed the interpolated free group factors, a family of II1\mathrm{II}_1 factors L(𝔽r)L(\mathbb{F}_r) indexed by real r>1r>1. Their results give a dichotomy: the interpolated free group factors are either all isomorphic or pairwise nonisomorphic. The version of the dichotomy that includes L(𝔽∞)L(\mathbb{F}_\infty) is due to Rădulescu. Goldbring and Pi restate it precisely.

The preprint proves that for every integer n≥3n\geq3 there is a unital normal trace-preserving **-isomorphism L(𝔽n)≅L(𝔽n+1)L(\mathbb{F}_n)\cong L(\mathbb{F}_{n+1}). It proves that L(𝔽2)L(\mathbb{F}_2) and L(𝔽3)L(\mathbb{F}_3) are isomorphic as tracial von Neumann algebras, which answers the free group factor isomorphism problem affirmatively. The free entropy dimensions of a finite tuple of self-adjoint elements of a tracial von Neumann algebra are four numerical invariants of the tuple’s joint distribution, defined either through microstates, matricial approximations of that distribution, or without them. From these isomorphisms the preprint also derives, in a result the Lean library does not contain, a negative answer to a question of Voiculescu: none of the four free entropy dimensions of a finite self-adjoint tuple depends only on the tracial von Neumann algebra the tuple generates.

The Lean library InterpolatedFactors proves the consequences for the interpolated free group factors that the preprint draws from the isomorphisms between consecutive free group factors. The isomorphisms L(𝔽n)≅L(𝔽n+1)L(\mathbb{F}_n)\cong L(\mathbb{F}_{n+1}) for n≥2n\geq2 and L(𝔽3)≅L(𝔽∞)L(\mathbb{F}_3)\cong L(\mathbb{F}_\infty) that it uses are proved in Lean, in the same formalization, by the companion Lean library FreeGroupFactor, which these pages do not present result by result. Lean has checked every result of the Lean library, so the Lean library is the reference: each result on these pages is stated as the Lean library states it and linked to the Lean declaration behind it, with the preprint’s statement beside it where the two differ. The preprint is a guide only, which gives the results their order and their names. For a real r>1r>1 that is not an integer, L(𝔽r)L(\mathbb{F}_r) is the amplification L(𝔽2)(r−1)−1/2L(\mathbb{F}_2)^{(r-1)^{-1/2}}. The Lean library proves that L(𝔽2)L(\mathbb{F}_2) and L(𝔽∞)L(\mathbb{F}_\infty) are factors. For every real s>1s>1 and every t>0t>0, the amplification formula L(𝔽s)t≅L(𝔽1+(s−1)t−2)L(\mathbb{F}_s)^t\cong L(\mathbb{F}_{1+(s-1)t^{-2}}) holds. Its main theorem gives, for all 1<r,s≤∞1<r,s\leq\infty, a trace-preserving normal **-isomorphism L(𝔽r)≅L(𝔽s)L(\mathbb{F}_r)\cong L(\mathbb{F}_s). In the Lean library, the fundamental group of every L(𝔽r)L(\mathbb{F}_r), 1<r≤∞1<r\leq\infty, is ℝ>0\mathbb{R}_{>0}. The Lean library also proves Rădulescu’s theorem on the corners of L(𝔽∞)L(\mathbb{F}_\infty) in the form that for every 0<t≤10<t\leq1, L(𝔽∞)L(\mathbb{F}_\infty) is isomorphic to its corner at a projection of trace tt with the normalized trace. For the free group 𝔽I\mathbb{F}_I on any set II, it proves L(𝔽I)L(\mathbb{F}_I) isomorphic to the von Neumann algebra generated by the free semicircular family indexed by II on the full Fock space over ℓ2(I)\ell^2(I), with the trace carried to the vacuum state.