A concrete algebra that cannot fit inside any nuclear norm ultrapower

← All digests

Family 292 · Operator algebras
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-25

1. The question

A complicated object can sometimes be placed inside a simpler universal one. For operator algebras, the Cuntz algebra $\mathcal O_2$ plays a remarkable universal role: every separable unital exact C*-algebra embeds into it.

What if exactness is dropped, but the target is enlarged to an ultrapower? An ultrapower allows an entire sequence of approximate realizations to act as one limiting realization. Kirchberg’s embedding problem asks whether every separable unital C*-algebra embeds into $\mathcal O_2^\omega$.

Family 292 claims an explicit negative answer. Its example is the full C*-algebra of a countable group defined using dyadic rational translations, integer matrices and dilation by two. The obstruction is stronger than failure for $\mathcal O_2$: the same algebra cannot embed unitally into the norm ultrapower of any nonzero unital nuclear C*-algebra.

2. Background and essential definitions

A C*-algebra is a complete normed algebra with an adjoint operation satisfying $\|a^*a\|=\|a\|^2$. An embedding is an injective *-homomorphism. “Unital” additionally requires the identity to map to the identity.

For a discrete group $G$, the full group C-algebra* $C^*(G)$ completes its complex group ring using all unitary representations. This differs from the reduced algebra, which uses only the regular translation representation. The full completion is part of the theorem’s hypothesis.

The Cuntz algebra $\mathcal O_2$ is generated by two isometries $s_1,s_2$ with

$$s_i^*s_i=1,\qquad s_1s_1^*+s_2s_2^*=1.$$

Each isometry copies the entire space into one of two orthogonal ranges. This makes $\mathcal O_2$ very different from a finite matrix algebra.

A free ultrafilter $\omega$ on the positive integers chooses a coherent notion of which infinite sets of indices are large. The norm ultrapower is

$$B^\omega=\ell^\infty(\mathbb N,B)\big/ \{(b_n):\lim_\omega\|b_n\|=0\}.$$

Thus bounded sequences differing by an operator-norm-null sequence represent the same element. The quotient uses operator norm, not a trace-based quadratic norm.

Nuclearity means that finite sets of algebra elements admit arbitrarily accurate factorizations through finite-dimensional algebras by completely positive maps. Exactness means that spatial tensoring preserves short exact sequences. Nuclearity implies exactness, but they are not the same property. A subalgebra of a nuclear algebra can be exact without itself being nuclear.

3. Development history and previous partial results

Kirchberg–Phillips’s embedding theorem established the universality of $\mathcal O_2$ for separable exact algebras. The ultrapower question asks whether sequences in this one nuclear algebra can represent algebras beyond that class. Nonexactness alone does not settle the ultrapower problem: ultrapowers have different permanence behavior from a single copy of the target.

Goldbring and Sinclair studied the problem using model theory. They connected a positive answer with good nuclear witnesses, a local condition asking whether satisfiable finite norm constraints can be realized by tuples admitting good nuclear approximations. They also related the question to existentially closed algebras, in which passing to an extension cannot improve finite existential tests.

More recent approaches explored computability and tensor norms. The source discusses computability consequences obtained by Fox–Goldbring–Hart and a conditional tensor-norm route of Goldbring–Sinclair. M1 instead claims a direct representation-theoretic obstruction from a specified group.

That group configuration has predecessors. De Cornulier recorded the property-(T) affine group and its finite-index dilation maps. Sauers and Eckhardt used related constructions to show nonfiniteness of full group algebras and to construct non-MF groups, meaning groups that cannot be represented faithfully by asymptotic finite matrices in the relevant norm sense.

Those facts alone do not answer the present question. A proper isometry obstructs finiteness, but $\mathcal O_2$ already contains proper isometries. Failure of approximation by matrices also does not automatically imply failure of embedding into a nuclear algebra’s norm ultrapower. M1 adds a finiteness principle for approximate representation types inside arbitrary nuclear targets.

4. What the new manuscripts claim

Let

$$G=\mathbb Z[1/2]^3\rtimes \bigl(\mathrm{SL}_3(\mathbb Z)\times\mathbb Z\bigr),$$

where $\mathbb Z[1/2]$ consists of rational numbers whose denominators are powers of two. The action is

$$(M,k)\cdot v=2^kMv.$$

Here $M$ is an integer three-by-three matrix of determinant one, and $k$ is an integer dilation parameter. The semidirect product combines translations with these linear transformations; explicitly,

$$(v,M,k)(w,N,\ell)=(v+2^kMw,MN,k+\ell).$$

M1 states that the separable unital algebra $A=C^*(G)$ has no unital embedding into $B^\omega$ for any nonzero unital nuclear complex C*-algebra $B$ and any free ultrafilter on $\mathbb N$. No separability assumption is imposed on $B$. The group is fixed independently of both the target and the ultrafilter.

Taking $B=\mathcal O_2$ gives the negative answer to Kirchberg’s embedding problem. The source recalls that the universal question formulated without unit preservation is equivalent to its unital formulation. Its stronger arbitrary-target theorem is explicitly stated for unital embeddings.

Using established model-theoretic equivalences, M1 derives three further consequences. No exact or nuclear unital C*-algebra is existentially closed among all unital C*-algebra extensions. Embeddability into $\mathcal O_2^\omega$ can fail after taking a crossed product by an action of $\mathbb Z$. And some satisfiable finite norm condition lacks good nuclear witnesses, with a fixed positive approximation-error obstruction for every realizing tuple.

The last two witnesses are existential consequences of the equivalences. M1 does not identify them explicitly through the displayed group construction.

5. Why the advance matters

Importance: 5/5. If correct, this fixes a fundamental limit of nuclear universality. Allowing infinitely many norm-approximate realizations in a nuclear algebra still would not accommodate every separable C*-algebra.

The explicit source algebra makes the claim particularly informative. Its obstruction is encoded by familiar ingredients—integer matrices, dyadic translations and a finite-index self-embedding—rather than only by a nonconstructive existence argument. At the same time, the theorem treats every nuclear target, not just the special geometry of $\mathcal O_2$.

The consequences also show that local approximation and global embeddability interact with algebraic operations. Even a crossed product by the amenable group $\mathbb Z$ can leave the class of embeddable algebras. This is a limitation of that class, not a claim that nuclear algebras themselves fail their defining approximation property.

6. The main idea

A rigid group with a shrinking translation lattice

Inside $G$, consider

$$\Gamma=\mathbb Z^3\rtimes\mathrm{SL}_3(\mathbb Z).$$

It has property (T): sufficiently almost invariant unit vectors in a unitary representation force an actual invariant vector. The map $\theta(v,M)=(2v,M)$ embeds $\Gamma$ as a subgroup of index eight. In the larger group, conjugating by the dilation generator implements $\theta$.

Choose a finite-dimensional irreducible representation $\pi_*$ that factors through reduction modulo two and is nontrivial on translations. Every positive iterate of $\theta$ makes its translation action trivial after that reduction, so $\pi_*$ cannot be equivalent to $\pi_*\circ\theta^j$ for $j\geq1$. This is a built-in nonperiodicity condition.

Nuclearity permits only finitely many approximate matches

Assume an ultrapower embedding exists and choose unitary representatives in $B$ for its group elements. At each coordinate, the representatives satisfy group relations only approximately. Treating them as genuine representations would lose an essential part of the problem.

M1 tests a coordinate tuple against finite-dimensional irreducible representations using a spectral measure of how nearly the corresponding tensor-product operators have invariant vectors. Property (T) supplies a gap separating matches from nonmatches.

The new finiteness lemma says that, inside a nuclear algebra, only finitely many representation classes can pass the chosen test at one coordinate. Its estimate works for arbitrary unitary tuples without requiring exact group relations. It also controls varying matrix dimensions, since the representation class may change with the coordinate.

Finite chains contradict nonperiodicity

Faithfulness ensures that $\pi_*$ appears among the approximate matches on an ultrafilter-large set of coordinates. The dilation relation and finite-index induction force every matched representation to have a predecessor in the same finite set, with predecessor defined through composition with $\theta$.

Iterating predecessors must eventually revisit a representation. M1’s representation-theoretic argument converts that repetition into periodicity of $\pi_*$ under a positive iterate of $\theta$, contradicting its construction.

The logical hinge is simultaneous control at one coordinate. A separate approximation for each representation would not produce a finite closed chain on which the contradiction could operate. The dimension-independent spectral estimate and nuclear finiteness lemma provide that uniformity.

7. Scope, limitations, and open questions

This is a norm-ultrapower embedding question. It is distinct from Connes’ embedding problem for tracial von Neumann algebras and from the tensor-norm formulation often also called Kirchberg’s conjecture. Their names and connections do not make them interchangeable.

The explicit algebra is the full group C*-algebra. The theorem does not automatically transfer to the reduced completion, to nonnuclear targets, or to every type of sequence quotient. The target may be nonseparable, but must be nonzero, unital and nuclear; the ultrapower is indexed by $\mathbb N$.

The example’s nonfiniteness and earlier non-MF constructions are context, not sufficient substitutes for the new obstruction. Conversely, failure to embed here does not prevent representations on Hilbert spaces, which every C*-algebra possesses.

The model-theoretic corollaries establish existence of additional obstructions without explicit witnesses. Producing concrete crossed-product examples and concrete finite norm conditions realizing those consequences would add information beyond the source’s stated construction.

8. Sources and further reading

• Goldbring and Sinclair, On Kirchberg’s Embedding Problem, for the norm-ultrapower formulation and model-theoretic equivalences.
• Eckhardt, Non-MF groups and non-finite full group C*-algebras, for the earlier group configuration and the distinct nonfiniteness obstruction.
• M1 specifies the finite-dimensional representation, the nuclear approximate-matching lemma, and which consequences are explicit versus existential.

Source manuscripts

• M1. OpenAI, An explicit obstruction to nuclear norm-ultrapower embeddings (September 23, 2026).

Sources consulted on 25 October 2026 at repository revision 1581d4c. Download the manuscript citations.