Von Neumann's Problem for II_1 Factors

Von Neumann's Problem for II_1 Factors

Definitions and mathematical statement

A $\mathrm{II}_1$ factor is a von Neumann algebra with center ${\mathbb{C}}1$, a faithful normal tracial state, and no nonzero minimal projections. Here amenability is the usual injectivity/amenability property of von Neumann algebras. For a group $G$, $L(G)$ is the weak operator closure of the algebra generated by its left translation operators on $\ell^2(G)$.

Let $F_2$ be the free group on two generators. The question is whether every nonamenable $\mathrm{II}_1$ factor $M$ contains a unital von Neumann subalgebra isomorphic to $L(F_2)$, or whether a counterexample exists. Such a subalgebra is itself a factor, so this is also a subfactor-containment question. No finite-index hypothesis is added. The catalogue imposes no separability assumption, and none is silently supplied here.

Formulation and concept references: [S1].

Short English statement

Does every nonamenable finite factor contain the von Neumann algebra of the free group on two generators?

Research attempt

Scope, attribution, and source audit.

This unresolved attempt was prepared by GPT-6 Astra Ultra on 27 September 2026 through primary-source review, explicit group-word constructions, and checks of convergence and trace conventions. The question in Peterson's list [S1] concerns actual unital containment in the given factor, with no separability or index restriction. It is stronger than containment in an ultrapower. The attempt reduces containment to an exact moment problem and constructs finite moment matches inside amenable matrix algebras. That construction is a test of an attempted proof, not a counterexample to the problem, since its ambient examples are amenable.

An exact certificate using two unitaries.

Let $(M,\tau)$ be any finite von Neumann algebra with faithful normal trace. A sufficient and necessary certificate for a trace-preserving copy of $L(F_2)$ is a pair of unitaries $u,v\in M$ such that \begin{equation}\label{wsref-0-61-34-33-39-3a-77-6f-72-64-73} \tau(w(u,v))=0 \quad\text{for every nonidentity reduced word } w\in F_2. \end{equation} The canonical generators of $L(F_2)$ satisfy this because the canonical trace extracts the identity coefficient. Conversely define $\pi(a)=u$, $\pi(b)=v$ on the free group. For group-algebra elements, (a439:words) gives \[ \tau(\pi(x)^*\pi(y))=\langle x,y\rangle_{\ell^2(F_2)}. \] Thus the map sending $\delta_g$ to $\pi(g)$ extends to a unitary from $\ell^2(F_2)$ onto $L^2(N,\tau)$, where $N=W^*(u,v)$. It intertwines left multiplication by every group generator with the corresponding operator on $L^2(N)$. Taking the von Neumann closures therefore identifies $L(F_2)$ normally and trace-preservingly with $N$. Faithfulness of the traces makes this identification injective, and the common identity makes it unital.

In a factor a unital copy automatically has the correct normalized trace, because the trace restricted to a $\mathrm{II}_1$ factor is its unique normalized normal trace. Hence no extra choice of trace weakens the requested containment. Equivalently $u,v$ are free Haar unitaries: the reduced-word condition includes all nonzero powers of each generator and all alternating products of their centered powers.

A complete restricted class: group factors with a free subgroup.

If a group $G$ contains elements $a,b$ generating a subgroup isomorphic to $F_2$, its canonical translation unitaries in $L(G)$ satisfy (a439:words). Every nonidentity reduced word is a nonidentity group element and therefore has trace zero. This proves containment for any such group factor, with no finite-index condition. If $L(G)$ is a $\mathrm{II}_1$ factor, it is a genuine instance of the question.

The contrapositive is unavailable. A nonamenable group without a free subgroup does not automatically yield a counterexample: unitaries generating a copy of $L(F_2)$ inside $L(G)$ need not be individual group elements. They may be operators with large Fourier support. A group-theoretic obstruction to selecting two group generators does not exclude the von Neumann subalgebra certificate above.

Finite moment tests can all hold in finite algebras.

Here is an explicit residual-finiteness construction requiring no asymptotic random-matrix theorem. Fix $N$. Let $B_N$ be the finite set of reduced words of length at most $N$. Left multiplication by $a$ defines a partial bijection of $B_N$ wherever both the input and reduced output remain in $B_N$. Complete it to a permutation $A$ of $B_N$: the unused domain and range have the same cardinality. Do the same for $b$, obtaining $B$.

Every nonidentity reduced word $w$ of length at most $N$ acts nontrivially on the empty word under these permutations. Indeed the successive reduced suffixes obtained while evaluating $w(A,B)$ have length at most $N$, so each prescribed partial multiplication agrees with the free group action, including inverse steps. The final image is $w$, not the empty word. Thus in the finite permutation group $Q_N=\langle A,B\rangle$ none of these words is the identity.

Now take the left regular representation of $Q_N$ on $\mathbb C^{|Q_N|}$. A nonidentity element permutes the standard basis without fixed points, so its normalized matrix trace is zero. Consequently the resulting unitaries $U_N,V_N$ satisfy (a439:words) for every word of length at most $N$, exactly, not just approximately.

They cannot satisfy all word conditions simultaneously at this fixed matrix size. Such conditions would make the infinitely many distinct group words orthonormal in the finite-dimensional Hilbert space of matrices. The finite-dimensionality obstruction is immediate even without a polynomial identity for matrices.

Why passing to limits is the actual issue.

The hyperfinite $\mathrm{II}_1$ factor $R$ contains unital matrix algebras of every finite size, so the preceding finite certificates can all be realized in $R$, one pair for each $N$. Yet $R$ cannot contain $L(F_2)$: amenability passes to expected subalgebras, every unital von Neumann subalgebra of a finite tracial algebra has a trace-preserving conditional expectation, and the free group factor is nonamenable. These established operator-algebra facts are recalled in [E1]. They are used only to diagnose this proof strategy.

One can also see where an ultraproduct changes the statement. The sequences $(U_N),(V_N)$ become a pair with all required moments in a tracial matrix ultraproduct, since each fixed word eventually has trace zero. The ultraproduct supplies a compactness device outside any particular original matrix algebra or factor. It has not supplied a pair inside the specified $M$.

Weak compactness inside $M$ does not fix the problem. Weak limits of unitaries may fail to be unitary. For example multiplication by $z^k$ in $L^\infty(S^1)$ converges weak-star to zero, as follows first for trigonometric polynomials and then for $L^1$ tests by approximation. Products are not jointly weakly continuous. Therefore taking separate weakly convergent subsequences and substituting their limits into every word is not a valid moment argument.

A precise sufficient coherence condition.

Suppose one can find pairs $(u_j,v_j)$ of unitaries inside the same $M$ such that \[ \max_{|w|\leq j,\ w\ne e}|\tau(w(u_j,v_j))|\leq2^{-j}, \qquad \sum_j\left(\|u_{j+1}-u_j\|_2+ \|v_{j+1}-v_j\|_2\right)<\infty. \] Then the pairs converge in $L^2$ to unitaries $u,v\in M$. For clarity, a uniformly operator-norm bounded $L^2$-Cauchy sequence has a limit in the same bounded ball of $M$, using weak compactness and uniqueness of the $L^2$ limit. Multiplication is continuous in $L^2$ on bounded sets, so $u^*u=uu^*=1$ and the same holds for $v$.

A telescoping expansion of a word of length $\ell$ bounds its $L^2$ difference by \[ \|w(u_j,v_j)-w(u,v)\|_2 \leq\ell\max(\|u_j-u\|_2,\|v_j-v\|_2), \] with the same estimate for inverse letters because adjoints preserve the $2$-norm. Since $|\tau(x)|\leq\|x\|_2$, all moments pass to zero. The exact certificate proves containment.

This lemma makes the first gap concrete: nonamenability would have to supply an exact free pair, or an approximation scheme with sufficient coherence in the original algebra. Arbitrarily accurate finite moment matching, even exact matching, is inadequate. No implication from nonamenability to the displayed summable perturbation condition is proved here.

Verification and outcome.

The finite-word construction was checked by enumerating short reduced words and their partial permutation paths. The GNS argument keeps the full group law, not merely pairwise orthogonality of $u$ and $v$. All constructions use only countably many operators, but this does not silently assume the ambient factor is separable or prove a reduction from every nonseparable factor to a special separable class. UNRESOLVED. Exact moment and coherent-limit criteria are proved, as is a restricted group-factor case. The universal nonamenable-factor assertion and the existence of a counterexample both remain open in this attempt.

Sources

• S1 Open problems in operator algebras. Accessed 2026-09-01. https://math.vanderbilt.edu/peters10/problems.html. Formulation source.
• E1 Jesse Peterson, Notes on von Neumann algebras, author course notes, especially finite traces, conditional expectations and amenability. https://math.vanderbilt.edu/peters10/teaching/spring2013/vonNeumannAlgebras.pdf. Author reference for the established operator-algebra inputs; consulted 27 September 2026. The finite-word construction and coherent-limit lemma are proved in the attempt.